Silence an OpenBSD linker warning

git manual pages render a little bit more sensibly now.

.clang-format

@@ -0,0 +1,32 @@
+# clang-format is fairly limited, and these rules are approximate:
+#  - array initializers can get terribly mangled with clang-format 12.0,
+#  - sometimes it still aligns with space characters,
+#  - struct name NL { NL ... NL } NL name; is unachievable.
+BasedOnStyle: GNU
+ColumnLimit: 80
+IndentWidth: 4
+TabWidth: 4
+UseTab: ForContinuationAndIndentation
+BreakBeforeBraces: Allman
+SpaceAfterCStyleCast: true
+AlignAfterOpenBracket: DontAlign
+AlignOperands: DontAlign
+AlignConsecutiveMacros: Consecutive
+AllowAllArgumentsOnNextLine: false
+AllowAllParametersOfDeclarationOnNextLine: false
+IndentGotoLabels: false
+
+# IncludeCategories has some potential, but it may also break the build.
+# Note that the documentation says the value should be "Never".
+SortIncludes: false
+
+# This is a compromise, it generally works out aesthetically better.
+BinPackArguments: false
+
+# Unfortunately, this can't be told to align to column 40 or so.
+SpacesBeforeTrailingComments: 2
+
+# liberty-specific macro body wrappers.
+MacroBlockBegin: "BLOCK_START"
+MacroBlockEnd: "BLOCK_END"
+ForEachMacros: ["LIST_FOR_EACH"]

.gitignore

@@ -7,3 +7,5 @@
 /liberty.files
 /liberty.creator*
 /liberty.includes
+/liberty.cflags
+/liberty.cxxflags

CMakeLists.txt

@@ -1,12 +1,12 @@
-project (liberty C)
-cmake_minimum_required (VERSION 2.8.5)
+cmake_minimum_required (VERSION 3.0...3.27)
+project (liberty C CXX)
 
 # Moar warnings
 if ("${CMAKE_C_COMPILER_ID}" MATCHES "GNU" OR CMAKE_COMPILER_IS_GNUCC)
 	# -Wunused-function is pretty annoying here, as everything is static
-	set (wdisabled "-Wno-unused-function -Wno-implicit-fallthrough")
+	set (wdisabled "-Wno-unused-function")
 	set (CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -std=c99 -Wall -Wextra ${wdisabled}")
-endif ("${CMAKE_C_COMPILER_ID}" MATCHES "GNU" OR CMAKE_COMPILER_IS_GNUCC)
+endif ()
 
 # Dependencies
 set (CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake)
@@ -16,11 +16,9 @@ find_package (PkgConfig REQUIRED)
 pkg_check_modules (libssl REQUIRED libssl libcrypto)
 
 if ("${CMAKE_SYSTEM_NAME}" MATCHES "BSD")
-	include_directories (/usr/local/include)
-	link_directories (/usr/local/lib)
 	# Our POSIX version macros make these undefined
 	add_definitions (-D__BSD_VISIBLE=1 -D_BSD_SOURCE=1)
-endif ("${CMAKE_SYSTEM_NAME}" MATCHES "BSD")
+endif ()
 
 set (common_libraries ${libssl_LIBRARIES})
 include_directories (${libssl_INCLUDE_DIRS})
@@ -31,16 +29,102 @@ link_directories (${libssl_LIBRARY_DIRS})
 foreach (extra iconv rt)
 	find_library (extra_lib_${extra} ${extra})
 	if (extra_lib_${extra})
-		list (APPEND common_libraries ${extra})
-	endif (extra_lib_${extra})
-endforeach (extra)
+		list (APPEND common_libraries ${extra_lib_${extra}})
+	endif ()
+endforeach ()
 
 # Build some unit tests
 include_directories (${PROJECT_SOURCE_DIR})
 enable_testing ()
-foreach (name liberty proto)
+set (tests liberty proto xdg)
+
+pkg_check_modules (libpulse libpulse)
+if (libpulse_FOUND)
+	list (APPEND tests pulse)
+	list (APPEND common_libraries ${libpulse_LIBRARIES})
+	include_directories (${libpulse_INCLUDE_DIRS})
+	link_directories (${libpulse_LIBRARY_DIRS})
+endif ()
+
+foreach (name ${tests})
 	add_executable (test-${name} tests/${name}.c ${common_sources})
 	add_threads (test-${name})
 	target_link_libraries (test-${name} ${common_libraries})
 	add_test (NAME test-${name} COMMAND test-${name})
-endforeach (name)
+endforeach ()
+
+# --- Tools --------------------------------------------------------------------
+
+# Test the AsciiDoc manual page generator for a successful parse
+set (ASCIIMAN ${PROJECT_SOURCE_DIR}/tools/asciiman.awk)
+add_custom_command (OUTPUT libertyxdr.7
+	COMMAND env LC_ALL=C awk -f ${ASCIIMAN}
+		"${PROJECT_SOURCE_DIR}/libertyxdr.adoc" > libertyxdr.7
+	DEPENDS libertyxdr.adoc ${ASCIIMAN}
+	COMMENT "Generating man page for libertyxdr" VERBATIM)
+add_custom_target (docs ALL DEPENDS libertyxdr.7)
+
+# Test CMake script parsing
+add_test (test-cmake-parser
+	env LC_ALL=C awk -f ${PROJECT_SOURCE_DIR}/tools/cmake-parser.awk
+	-f ${PROJECT_SOURCE_DIR}/tools/cmake-dump.awk ${CMAKE_CURRENT_LIST_FILE})
+
+# Test protocol code generation
+set (lxdrgen_outputs)
+set (lxdrgen_base "${PROJECT_BINARY_DIR}/lxdrgen.lxdr")
+foreach (backend c cpp go mjs swift)
+	list (APPEND lxdrgen_outputs ${lxdrgen_base}.${backend})
+	add_custom_command (OUTPUT ${lxdrgen_base}.${backend}
+		COMMAND env LC_ALL=C awk
+			-f ${PROJECT_SOURCE_DIR}/tools/lxdrgen.awk
+			-f ${PROJECT_SOURCE_DIR}/tools/lxdrgen-${backend}.awk
+			-v PrefixCamel=ProtoGen
+			${PROJECT_SOURCE_DIR}/tests/lxdrgen.lxdr
+			> ${lxdrgen_base}.${backend}
+		DEPENDS
+			${PROJECT_SOURCE_DIR}/tools/lxdrgen.awk
+			${PROJECT_SOURCE_DIR}/tools/lxdrgen-${backend}.awk
+			${PROJECT_SOURCE_DIR}/tests/lxdrgen.lxdr
+		COMMENT "Generating test protocol code (${backend})" VERBATIM)
+endforeach ()
+add_custom_target (test-lxdrgen-outputs ALL DEPENDS ${lxdrgen_outputs})
+
+set_source_files_properties (${lxdrgen_base}.c
+	PROPERTIES HEADER_FILE_ONLY TRUE)
+add_executable (test-lxdrgen-c tests/lxdrgen.c ${lxdrgen_base}.c)
+target_include_directories (test-lxdrgen-c PUBLIC ${PROJECT_BINARY_DIR})
+add_test (NAME test-lxdrgen-c COMMAND test-lxdrgen-c)
+
+set_source_files_properties (${lxdrgen_base}.cpp
+	PROPERTIES HEADER_FILE_ONLY TRUE)
+if (WIN32)
+	add_executable (test-lxdrgen-cpp tests/lxdrgen.cpp
+		${lxdrgen_base}.cpp tools/lxdrgen-cpp-win32.cpp)
+else ()
+	add_executable (test-lxdrgen-cpp tests/lxdrgen.cpp
+		${lxdrgen_base}.cpp tools/lxdrgen-cpp-posix.cpp)
+endif ()
+target_include_directories (test-lxdrgen-cpp PUBLIC ${PROJECT_BINARY_DIR})
+add_test (NAME test-lxdrgen-cpp COMMAND test-lxdrgen-cpp)
+
+find_program (GO_EXECUTABLE go)
+if (GO_EXECUTABLE)
+	add_test (test-lxdrgen-go ${GO_EXECUTABLE} vet ${lxdrgen_base}.go)
+else ()
+	message (WARNING "Cannot test generated protocol code for Go")
+endif ()
+
+find_program (NODE_EXECUTABLE node)
+if (NODE_EXECUTABLE)
+	add_test (test-lxdrgen-mjs ${NODE_EXECUTABLE} -c ${lxdrgen_base}.mjs)
+else ()
+	message (WARNING "Cannot test generated protocol code for Javascript")
+endif ()
+
+find_program (SWIFTC_EXECUTABLE swiftc)
+if (SWIFTC_EXECUTABLE)
+	add_test (test-lxdrgen-swift
+		${SWIFTC_EXECUTABLE} -typecheck ${lxdrgen_base}.swift)
+else ()
+	message (WARNING "Cannot test generated protocol code for Swift")
+endif ()

LICENSE

@@ -1,4 +1,4 @@
-Copyright (c) 2014 - 2020, Přemysl Eric Janouch <p@janouch.name>
+Copyright (c) 2014 - 2024, Přemysl Eric Janouch <p@janouch.name>
 
 Permission to use, copy, modify, and/or distribute this software for any
 purpose with or without fee is hereby granted.

README.adoc

@@ -1,21 +1,67 @@
 liberty
 =======
 
-'liberty' is a pseudolibrary of all the common C code I have written for various
-projects.  I used to copy-paste large swaths of code with minimal changes to it
+'liberty' is a pseudolibrary largely consisting of reusable C code for my
+various projects.  I used to copy-paste large swaths of it with minimal changes,
 and it slowly became awfully painful to synchronize.  The project can be thought
 of as a successor to my other C library, libxtnd.
 
 You are supposed to import it as a git submodule and include the main source
-file directly everywhere you need it.  Everything is declared "static".  I have
-come to the conclusion that this style of C programming suits me the best, as it
-allows me to nearly forget about the mess that are header files.
+file directly everywhere you need it, setting feature flags as appropriate.
+Everything is declared "static".  I have come to the conclusion that this style
+of C programming suits me the best, as it allows me to nearly forget about the
+mess that are header files.
 
 The API is intentionally unstable, which allows for easy refactoring.
 
 All development is done on Linux, but other POSIX-compatible operating systems
-should be supported as well.  They have an extremely low priority, however, and
-I'm not testing them at all, with the exception of OpenBSD.
+should be generally supported as well.  They have an extremely low priority,
+however, and I'm not testing them at all, perhaps with the exception of macOS.
+
+Tools
+-----
+This project also hosts a number of supporting scripts written in portable AWK:
+
+asciiman.awk::
+	A fallback manual page generator for AsciiDoc documents,
+	motivated by the hugeness of AsciiDoc's and Asciidoctor's dependency trees.
+	Just like them, it uses the _man_ macro package.
+
+cmake-parser.awk::
+	Parses the CMake language to the extent that is necessary to reliably
+	extract project versions.  Its greatest limitation is its inability
+	to expand variables, which would require a full interpreter.
+
+cmake-dump.awk::
+	This can be used in conjunction with the previous script to dump CMake
+	scripts in a normalized format for further processing.
+
+lxdrgen.awk::
+	Protocol code generator for a variant of XDR,
+	which is link:libertyxdr.adoc[documented separately].
+	Successfully employed in https://git.janouch.name/p/xK[xK].
+
+lxdrgen-c.awk::
+	LibertyXDR backend that builds on top of the C pseudolibrary.
+
+lxdrgen-cpp.awk::
+lxdrgen-cpp-win32.cpp::
+lxdrgen-cpp-posix.cpp::
+	LibertyXDR backend for C++, primarily targeting Win32 and its wide strings.
+	Link the result together with one of the accompanied source files.
+
+lxdrgen-go.awk::
+	LibertyXDR backend for Go, supporting _encoding/json_ interfaces.  It also
+	produces optimized JSON marshallers (however, note that the _json.Marshaler_
+	interface is bound to be underperforming, due to the amount of otherwise
+	avoidable memory allocations it necessitates).
+
+lxdrgen-mjs.awk::
+	LibertyXDR backend for Javascript, currently for decoding only.
+	It cuts a corner by not using BigInts, on par with `JSON.parse()`.
+
+lxdrgen-swift.awk::
+	LibertyXDR backend for the Swift programming language.
 
 Contributing and Support
 ------------------------

cmake/AddThreads.cmake

@@ -9,15 +9,15 @@ find_package (Threads)
 function (add_threads target)
 	if (NOT Threads_FOUND OR NOT CMAKE_USE_PTHREADS_INIT)
 		message (FATAL_ERROR "pthreads not found")
-	endif (NOT Threads_FOUND OR NOT CMAKE_USE_PTHREADS_INIT)
+	endif ()
 
 	if (THREADS_HAVE_PTHREAD_ARG)
 		set_property (TARGET ${target} PROPERTY
 			COMPILE_OPTIONS "-pthread")
 		set_property (TARGET ${target} PROPERTY
 			INTERFACE_COMPILE_OPTIONS "-pthread")
-	endif (THREADS_HAVE_PTHREAD_ARG)
+	endif ()
 	if (CMAKE_THREAD_LIBS_INIT)
 		target_link_libraries (${target} "${CMAKE_THREAD_LIBS_INIT}")
-	endif (CMAKE_THREAD_LIBS_INIT)
-endfunction (add_threads)
+	endif ()
+endfunction ()

cmake/FindNcursesw.cmake

@@ -7,7 +7,7 @@ pkg_check_modules (Ncursesw QUIET ncursesw)
 set (required_vars Ncursesw_LIBRARIES)
 if (NOT Ncursesw_FOUND)
 	find_library (Ncursesw_LIBRARIES NAMES ncursesw)
-	find_path (Ncursesw_INCLUDE_DIRS ncurses.h)
+	find_path (Ncursesw_INCLUDE_DIRS ncurses.h PATH_SUFFIXES ncurses)
 	list (APPEND required_vars Ncursesw_INCLUDE_DIRS)
 endif (NOT Ncursesw_FOUND)
 

cmake/IconUtils.cmake

@@ -0,0 +1,40 @@
+# Public Domain
+
+function (icon_to_png name svg size output_dir output)
+	set (_dimensions "${size}x${size}")
+	set (_png_path "${output_dir}/hicolor/${_dimensions}/apps")
+	set (_png "${_png_path}/${name}.png")
+	set (${output} "${_png}" PARENT_SCOPE)
+
+	set (_find_program_REQUIRE)
+	if (NOT ${CMAKE_VERSION} VERSION_LESS 3.18.0)
+		set (_find_program_REQUIRE REQUIRED)
+	endif ()
+
+	find_program (rsvg_convert_EXECUTABLE rsvg-convert ${_find_program_REQUIRE})
+	add_custom_command (OUTPUT "${_png}"
+		COMMAND ${CMAKE_COMMAND} -E make_directory "${_png_path}"
+		COMMAND ${rsvg_convert_EXECUTABLE} "--output=${_png}"
+			"--width=${size}" "--height=${size}" -- "${svg}"
+		DEPENDS "${svg}"
+		COMMENT "Generating ${name} ${_dimensions} application icon" VERBATIM)
+endfunction ()
+
+# You should include a 256x256 icon--which takes less space as raw PNG.
+function (icon_for_win32 ico pngs pngs_raw)
+	set (_raws)
+	foreach (png ${pngs_raw})
+		list (APPEND _raws "--raw=${png}")
+	endforeach ()
+
+	set (_find_program_REQUIRE)
+	if (NOT ${CMAKE_VERSION} VERSION_LESS 3.18.0)
+		set (_find_program_REQUIRE REQUIRED)
+	endif ()
+
+	find_program (icotool_EXECUTABLE icotool ${_find_program_REQUIRE})
+	add_custom_command (OUTPUT "${ico}"
+		COMMAND ${icotool_EXECUTABLE} -c -o "${ico}" ${_raws} -- ${pngs}
+		DEPENDS ${pngs} ${pngs_raw}
+		COMMENT "Generating Windows program icon" VERBATIM)
+endfunction ()

cmake/toolchains/MinGW-w64-x64.cmake

@@ -0,0 +1,15 @@
+set (CMAKE_SYSTEM_NAME "Windows")
+set (CMAKE_SYSTEM_PROCESSOR "x86_64")
+
+set (CMAKE_C_COMPILER "x86_64-w64-mingw32-gcc")
+set (CMAKE_CXX_COMPILER "x86_64-w64-mingw32-g++")
+set (CMAKE_RC_COMPILER "x86_64-w64-mingw32-windres")
+
+# Remember to set WINEPATH for library dependencies
+set (CMAKE_CROSSCOMPILING_EMULATOR "wine64")
+
+set (CMAKE_FIND_ROOT_PATH "/usr/x86_64-w64-mingw32")
+
+set (CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
+set (CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
+set (CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)

cmake/toolchains/MinGW-w64-x86.cmake

@@ -0,0 +1,15 @@
+set (CMAKE_SYSTEM_NAME "Windows")
+set (CMAKE_SYSTEM_PROCESSOR "x86")
+
+set (CMAKE_C_COMPILER "i686-w64-mingw32-gcc")
+set (CMAKE_CXX_COMPILER "i686-w64-mingw32-g++")
+set (CMAKE_RC_COMPILER "i686-w64-mingw32-windres")
+
+# Remember to set WINEPATH for library dependencies
+set (CMAKE_CROSSCOMPILING_EMULATOR "wine")
+
+set (CMAKE_FIND_ROOT_PATH "/usr/i686-w64-mingw32")
+
+set (CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
+set (CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
+set (CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)

liberty-proto.c

@@ -148,7 +148,7 @@ irc_free_message (struct irc_message *msg)
 
 static void
 irc_process_buffer (struct str *buf,
-	void (*callback)(const struct irc_message *, const char *, void *),
+	void (*callback) (const struct irc_message *, const char *, void *),
 	void *user_data)
 {
 	char *start = buf->str, *end = start + buf->len;
@@ -1321,7 +1321,6 @@ enum mpd_subsystem
 #define XX(a, b, c) MPD_SUBSYSTEM_ ## a = (1 << b),
 	MPD_SUBSYSTEM_TABLE (XX)
 #undef XX
-	MPD_SUBSYSTEM_MAX
 };
 
 static const char *mpd_subsystem_names[] =
@@ -1912,14 +1911,14 @@ mpd_client_connect_unix (struct mpd_client *self, const char *address,
 	// Expand tilde if needed
 	char *expanded = resolve_filename (address, xstrdup);
 
-	struct sockaddr_un sun;
-	sun.sun_family = AF_UNIX;
-	strncpy (sun.sun_path, expanded, sizeof sun.sun_path);
-	sun.sun_path[sizeof sun.sun_path - 1] = 0;
+	struct sockaddr_un sau;
+	sau.sun_family = AF_UNIX;
+	strncpy (sau.sun_path, expanded, sizeof sau.sun_path);
+	sau.sun_path[sizeof sau.sun_path - 1] = 0;
 
 	free (expanded);
 
-	if (connect (fd, (struct sockaddr *) &sun, sizeof sun))
+	if (connect (fd, (struct sockaddr *) &sau, sizeof sau))
 	{
 		error_set (e, "%s: %s", "connect", strerror (errno));
 		xclose (fd);

liberty-pulse.c

@@ -0,0 +1,348 @@
+/*
+ * liberty-pulse.c: PulseAudio mainloop abstraction
+ *
+ * Copyright (c) 2016 - 2021, Přemysl Eric Janouch <p@janouch.name>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
+ * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
+ * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
+ * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ */
+
+#include <pulse/mainloop.h>
+
+// --- PulseAudio mainloop abstraction -----------------------------------------
+
+struct pa_io_event
+{
+	LIST_HEADER (pa_io_event)
+
+	pa_mainloop_api *api;               ///< Parent structure
+	struct poller_fd fd;                ///< Underlying FD event
+
+	pa_io_event_cb_t dispatch;          ///< Dispatcher
+	pa_io_event_destroy_cb_t free;      ///< Destroyer
+	void *user_data;                    ///< User data
+};
+
+struct pa_time_event
+{
+	LIST_HEADER (pa_time_event)
+
+	pa_mainloop_api *api;               ///< Parent structure
+	struct poller_timer timer;          ///< Underlying timer event
+
+	pa_time_event_cb_t dispatch;        ///< Dispatcher
+	pa_time_event_destroy_cb_t free;    ///< Destroyer
+	void *user_data;                    ///< User data
+};
+
+struct pa_defer_event
+{
+	LIST_HEADER (pa_defer_event)
+
+	pa_mainloop_api *api;               ///< Parent structure
+	struct poller_idle idle;            ///< Underlying idle event
+
+	pa_defer_event_cb_t dispatch;       ///< Dispatcher
+	pa_defer_event_destroy_cb_t free;   ///< Destroyer
+	void *user_data;                    ///< User data
+};
+
+struct poller_pa
+{
+	struct poller *poller;              ///< The underlying event loop
+	pa_io_event *io_list;               ///< I/O events
+	pa_time_event *time_list;           ///< Timer events
+	pa_defer_event *defer_list;         ///< Deferred events
+};
+
+// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+
+static short
+poller_pa_flags_to_events (pa_io_event_flags_t flags)
+{
+	short result = 0;
+	if (flags & PA_IO_EVENT_ERROR)   result |= POLLERR;
+	if (flags & PA_IO_EVENT_HANGUP)  result |= POLLHUP;
+	if (flags & PA_IO_EVENT_INPUT)   result |= POLLIN;
+	if (flags & PA_IO_EVENT_OUTPUT)  result |= POLLOUT;
+	return result;
+}
+
+static pa_io_event_flags_t
+poller_pa_events_to_flags (short events)
+{
+	pa_io_event_flags_t result = 0;
+	if (events & POLLERR)  result |= PA_IO_EVENT_ERROR;
+	if (events & POLLHUP)  result |= PA_IO_EVENT_HANGUP;
+	if (events & POLLIN)   result |= PA_IO_EVENT_INPUT;
+	if (events & POLLOUT)  result |= PA_IO_EVENT_OUTPUT;
+	return result;
+}
+
+static struct timeval
+poller_pa_get_current_time (void)
+{
+	struct timeval tv;
+#ifdef _POSIX_TIMERS
+	struct timespec tp;
+	hard_assert (clock_gettime (CLOCK_REALTIME, &tp) != -1);
+	tv.tv_sec = tp.tv_sec;
+	tv.tv_usec = tp.tv_nsec / 1000;
+#else
+	gettimeofday (&tv, NULL);
+#endif
+	return tv;
+}
+
+// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+
+static void
+poller_pa_io_dispatcher (const struct pollfd *pfd, void *user_data)
+{
+	pa_io_event *self = user_data;
+	self->dispatch (self->api, self,
+		pfd->fd, poller_pa_events_to_flags (pfd->revents), self->user_data);
+}
+
+static void
+poller_pa_io_enable (pa_io_event *self, pa_io_event_flags_t events)
+{
+	struct poller_fd *fd = &self->fd;
+	if (events)
+		poller_fd_set (fd, poller_pa_flags_to_events (events));
+	else
+		poller_fd_reset (fd);
+}
+
+static pa_io_event *
+poller_pa_io_new (pa_mainloop_api *api, int fd_, pa_io_event_flags_t events,
+	pa_io_event_cb_t cb, void *userdata)
+{
+	pa_io_event *self = xcalloc (1, sizeof *self);
+	self->api = api;
+	self->dispatch = cb;
+	self->user_data = userdata;
+
+	struct poller_pa *data = api->userdata;
+	self->fd = poller_fd_make (data->poller, fd_);
+	self->fd.user_data = self;
+	self->fd.dispatcher = poller_pa_io_dispatcher;
+
+	// FIXME: under x2go PA tries to register twice for the same FD,
+	//   which fails with our curent poller implementation;
+	//   we could maintain a list of { poller_fd, listeners } structures;
+	//   or maybe we're doing something wrong, which is yet to be determined
+	poller_pa_io_enable (self, events);
+	LIST_PREPEND (data->io_list, self);
+	return self;
+}
+
+static void
+poller_pa_io_free (pa_io_event *self)
+{
+	if (self->free)
+		self->free (self->api, self, self->user_data);
+
+	struct poller_pa *data = self->api->userdata;
+	poller_fd_reset (&self->fd);
+	LIST_UNLINK (data->io_list, self);
+	free (self);
+}
+
+static void
+poller_pa_io_set_destroy (pa_io_event *self, pa_io_event_destroy_cb_t cb)
+{
+	self->free = cb;
+}
+
+// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+
+static void
+poller_pa_time_dispatcher (void *user_data)
+{
+	pa_time_event *self = user_data;
+	// XXX: the meaning of the time argument is undocumented,
+	//   so let's just put current Unix time in there
+	struct timeval now = poller_pa_get_current_time ();
+	self->dispatch (self->api, self, &now, self->user_data);
+}
+
+static void
+poller_pa_time_restart (pa_time_event *self, const struct timeval *tv)
+{
+	struct poller_timer *timer = &self->timer;
+	if (tv)
+	{
+		struct timeval now = poller_pa_get_current_time ();
+		poller_timer_set (timer,
+			(tv->tv_sec  - now.tv_sec)  * 1000 +
+			(tv->tv_usec - now.tv_usec) / 1000);
+	}
+	else
+		poller_timer_reset (timer);
+}
+
+static pa_time_event *
+poller_pa_time_new (pa_mainloop_api *api, const struct timeval *tv,
+	pa_time_event_cb_t cb, void *userdata)
+{
+	pa_time_event *self = xcalloc (1, sizeof *self);
+	self->api = api;
+	self->dispatch = cb;
+	self->user_data = userdata;
+
+	struct poller_pa *data = api->userdata;
+	self->timer = poller_timer_make (data->poller);
+	self->timer.user_data = self;
+	self->timer.dispatcher = poller_pa_time_dispatcher;
+
+	poller_pa_time_restart (self, tv);
+	LIST_PREPEND (data->time_list, self);
+	return self;
+}
+
+static void
+poller_pa_time_free (pa_time_event *self)
+{
+	if (self->free)
+		self->free (self->api, self, self->user_data);
+
+	struct poller_pa *data = self->api->userdata;
+	poller_timer_reset (&self->timer);
+	LIST_UNLINK (data->time_list, self);
+	free (self);
+}
+
+static void
+poller_pa_time_set_destroy (pa_time_event *self, pa_time_event_destroy_cb_t cb)
+{
+	self->free = cb;
+}
+
+// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+
+static void
+poller_pa_defer_dispatcher (void *user_data)
+{
+	pa_defer_event *self = user_data;
+	self->dispatch (self->api, self, self->user_data);
+}
+
+static pa_defer_event *
+poller_pa_defer_new (pa_mainloop_api *api,
+	pa_defer_event_cb_t cb, void *userdata)
+{
+	pa_defer_event *self = xcalloc (1, sizeof *self);
+	self->api = api;
+	self->dispatch = cb;
+	self->user_data = userdata;
+
+	struct poller_pa *data = api->userdata;
+	self->idle = poller_idle_make (data->poller);
+	self->idle.user_data = self;
+	self->idle.dispatcher = poller_pa_defer_dispatcher;
+
+	poller_idle_set (&self->idle);
+	LIST_PREPEND (data->defer_list, self);
+	return self;
+}
+
+static void
+poller_pa_defer_enable (pa_defer_event *self, int enable)
+{
+	struct poller_idle *idle = &self->idle;
+	if (enable)
+		poller_idle_set (idle);
+	else
+		poller_idle_reset (idle);
+}
+
+static void
+poller_pa_defer_free (pa_defer_event *self)
+{
+	if (self->free)
+		self->free (self->api, self, self->user_data);
+
+	struct poller_pa *data = self->api->userdata;
+	poller_idle_reset (&self->idle);
+	LIST_UNLINK (data->defer_list, self);
+	free (self);
+}
+
+static void
+poller_pa_defer_set_destroy (pa_defer_event *self,
+	pa_defer_event_destroy_cb_t cb)
+{
+	self->free = cb;
+}
+
+// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+
+static void
+poller_pa_quit (pa_mainloop_api *api, int retval)
+{
+	(void) api;
+	(void) retval;
+
+	// This is not called from within libpulse
+	hard_assert (!"quitting the libpulse event loop is unimplemented");
+}
+
+// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+
+static struct pa_mainloop_api g_poller_pa_template =
+{
+	.io_new            = poller_pa_io_new,
+	.io_enable         = poller_pa_io_enable,
+	.io_free           = poller_pa_io_free,
+	.io_set_destroy    = poller_pa_io_set_destroy,
+
+	.time_new          = poller_pa_time_new,
+	.time_restart      = poller_pa_time_restart,
+	.time_free         = poller_pa_time_free,
+	.time_set_destroy  = poller_pa_time_set_destroy,
+
+	.defer_new         = poller_pa_defer_new,
+	.defer_enable      = poller_pa_defer_enable,
+	.defer_free        = poller_pa_defer_free,
+	.defer_set_destroy = poller_pa_defer_set_destroy,
+
+	.quit              = poller_pa_quit,
+};
+
+static struct pa_mainloop_api *
+poller_pa_new (struct poller *self)
+{
+	struct poller_pa *data = xcalloc (1, sizeof *data);
+	data->poller = self;
+
+	struct pa_mainloop_api *api = xmalloc (sizeof *api);
+	*api = g_poller_pa_template;
+	api->userdata = data;
+	return api;
+}
+
+static void
+poller_pa_destroy (struct pa_mainloop_api *api)
+{
+	struct poller_pa *data = api->userdata;
+
+	LIST_FOR_EACH (pa_io_event, iter, data->io_list)
+		poller_pa_io_free (iter);
+	LIST_FOR_EACH (pa_time_event, iter, data->time_list)
+		poller_pa_time_free (iter);
+	LIST_FOR_EACH (pa_defer_event, iter, data->defer_list)
+		poller_pa_defer_free (iter);
+
+	free (data);
+	free (api);
+}

liberty-tui.c

@@ -1,270 +0,0 @@
-/*
- * liberty-tui.c: the ultimate C unlibrary: TUI
- *
- * Copyright (c) 2016 - 2017, Přemysl Eric Janouch <p@janouch.name>
- *
- * Permission to use, copy, modify, and/or distribute this software for any
- * purpose with or without fee is hereby granted.
- *
- * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
- * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
- * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
- * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
- * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
- * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
- * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
- *
- */
-
-// This file includes some common stuff to build TUI applications with
-
-#include <ncurses.h>
-
-// It is surprisingly hard to find a good library to handle Unicode shenanigans,
-// and there's enough of those for it to be impractical to reimplement them.
-//
-//                         GLib          ICU     libunistring    utf8proc
-// Decently sized            .            .            x            x
-// Grapheme breaks           .            x            .            x
-// Character width           x            .            x            x
-// Locale handling           .            .            x            .
-// Liberal license           .            x            .            x
-//
-// Also note that the ICU API is icky and uses UTF-16 for its primary encoding.
-//
-// Currently we're chugging along with libunistring but utf8proc seems viable.
-// Non-Unicode locales can mostly be handled with simple iconv like in sdtui.
-// Similarly grapheme breaks can be guessed at using character width (a basic
-// test here is Zalgo text).
-//
-// None of this is ever going to work too reliably anyway because terminals
-// and Unicode don't go awfully well together.  In particular, character cell
-// devices have some problems with double-wide characters.
-
-#include <unistr.h>
-#include <uniwidth.h>
-#include <uniconv.h>
-#include <unicase.h>
-
-// --- Configurable display attributes -----------------------------------------
-
-struct attrs
-{
-	short fg;                           ///< Foreground colour index
-	short bg;                           ///< Background colour index
-	chtype attrs;                       ///< Other attributes
-};
-
-/// Decode attributes in the value using a subset of the git config format,
-/// ignoring all errors since it doesn't affect functionality
-static struct attrs
-attrs_decode (const char *value)
-{
-	struct strv v = strv_make ();
-	cstr_split (value, " ", true, &v);
-
-	int colors = 0;
-	struct attrs attrs = { -1, -1, 0 };
-	for (char **it = v.vector; *it; it++)
-	{
-		char *end = NULL;
-		long n = strtol (*it, &end, 10);
-		if (*it != end && !*end && n >= SHRT_MIN && n <= SHRT_MAX)
-		{
-			if (colors == 0) attrs.fg = n;
-			if (colors == 1) attrs.bg = n;
-			colors++;
-		}
-		else if (!strcmp (*it, "bold"))    attrs.attrs |= A_BOLD;
-		else if (!strcmp (*it, "dim"))     attrs.attrs |= A_DIM;
-		else if (!strcmp (*it, "ul"))      attrs.attrs |= A_UNDERLINE;
-		else if (!strcmp (*it, "blink"))   attrs.attrs |= A_BLINK;
-		else if (!strcmp (*it, "reverse")) attrs.attrs |= A_REVERSE;
-#ifdef A_ITALIC
-		else if (!strcmp (*it, "italic"))  attrs.attrs |= A_ITALIC;
-#endif  // A_ITALIC
-	}
-	strv_free (&v);
-	return attrs;
-}
-
-// --- Terminal output ---------------------------------------------------------
-
-// Necessary abstraction to simplify aligned, formatted character output
-
-// This callback you need to implement in the application
-static bool app_is_character_in_locale (ucs4_t ch);
-
-struct row_char
-{
-	ucs4_t c;                           ///< Unicode codepoint
-	chtype attrs;                       ///< Special attributes
-	int width;                          ///< How many cells this takes
-};
-
-struct row_buffer
-{
-	ARRAY (struct row_char, chars)      ///< Characters
-	int total_width;                    ///< Total width of all characters
-};
-
-static struct row_buffer
-row_buffer_make (void)
-{
-	struct row_buffer self = {};
-	ARRAY_INIT_SIZED (self.chars, 256);
-	return self;
-}
-
-static void
-row_buffer_free (struct row_buffer *self)
-{
-	free (self->chars);
-}
-
-/// Replace invalid chars and push all codepoints to the array w/ attributes.
-static void
-row_buffer_append (struct row_buffer *self, const char *str, chtype attrs)
-{
-	// The encoding is only really used internally for some corner cases
-	const char *encoding = locale_charset ();
-
-	// Note that this function is a hotspot, try to keep it decently fast
-	struct row_char current = { .attrs = attrs };
-	struct row_char invalid = { .attrs = attrs, .c = '?', .width = 1 };
-	const uint8_t *next = (const uint8_t *) str;
-	while ((next = u8_next (&current.c, next)))
-	{
-		current.width = uc_width (current.c, encoding);
-		if (current.width < 0 || !app_is_character_in_locale (current.c))
-			current = invalid;
-
-		ARRAY_RESERVE (self->chars, 1);
-		self->chars[self->chars_len++] = current;
-		self->total_width += current.width;
-	}
-}
-
-static void
-row_buffer_append_args (struct row_buffer *self, const char *s, ...)
-	ATTRIBUTE_SENTINEL;
-
-static void
-row_buffer_append_args (struct row_buffer *self, const char *s, ...)
-{
-	va_list ap;
-	va_start (ap, s);
-
-	while (s)
-	{
-		row_buffer_append (self, s, va_arg (ap, chtype));
-		s = va_arg (ap, const char *);
-	}
-	va_end (ap);
-}
-
-static void
-row_buffer_append_buffer (struct row_buffer *self, const struct row_buffer *rb)
-{
-	ARRAY_RESERVE (self->chars, rb->chars_len);
-	memcpy (self->chars + self->chars_len, rb->chars,
-		rb->chars_len * sizeof *rb->chars);
-
-	self->chars_len   += rb->chars_len;
-	self->total_width += rb->total_width;
-}
-
-/// Pop as many codepoints as needed to free up "space" character cells.
-/// Given the suffix nature of combining marks, this should work pretty fine.
-static int
-row_buffer_pop_cells (struct row_buffer *self, int space)
-{
-	int made = 0;
-	while (self->chars_len && made < space)
-		made += self->chars[--self->chars_len].width;
-	self->total_width -= made;
-	return made;
-}
-
-static void
-row_buffer_space (struct row_buffer *self, int width, chtype attrs)
-{
-	if (width < 0)
-		return;
-
-	ARRAY_RESERVE (self->chars, (size_t) width);
-
-	struct row_char space = { .attrs = attrs, .c = ' ', .width = 1 };
-	self->total_width += width;
-	while (width-- > 0)
-		self->chars[self->chars_len++] = space;
-}
-
-static void
-row_buffer_ellipsis (struct row_buffer *self, int target)
-{
-	if (self->total_width <= target
-	 || !row_buffer_pop_cells (self, self->total_width - target))
-		return;
-
-	// We use attributes from the last character we've removed,
-	// assuming that we don't shrink the array (and there's no real need)
-	ucs4_t ellipsis = 0x2026; // …
-	if (app_is_character_in_locale (ellipsis))
-	{
-		if (self->total_width >= target)
-			row_buffer_pop_cells (self, 1);
-		if (self->total_width + 1 <= target)
-			row_buffer_append (self, "…",   self->chars[self->chars_len].attrs);
-	}
-	else if (target >= 3)
-	{
-		if (self->total_width >= target)
-			row_buffer_pop_cells (self, 3);
-		if (self->total_width + 3 <= target)
-			row_buffer_append (self, "...", self->chars[self->chars_len].attrs);
-	}
-}
-
-static void
-row_buffer_align (struct row_buffer *self, int target, chtype attrs)
-{
-	row_buffer_ellipsis (self, target);
-	row_buffer_space (self, target - self->total_width, attrs);
-}
-
-static void
-row_buffer_print (uint32_t *ucs4, chtype attrs)
-{
-	// This assumes that we can reset the attribute set without consequences
-	char *str = u32_strconv_to_locale (ucs4);
-	if (str)
-	{
-		attrset (attrs);
-		addstr (str);
-		attrset (0);
-		free (str);
-	}
-}
-
-static void
-row_buffer_flush (struct row_buffer *self)
-{
-	if (!self->chars_len)
-		return;
-
-	// We only NUL-terminate the chunks because of the libunistring API
-	uint32_t chunk[self->chars_len + 1], *insertion_point = chunk;
-	for (size_t i = 0; i < self->chars_len; i++)
-	{
-		struct row_char *iter = self->chars + i;
-		if (i && iter[0].attrs != iter[-1].attrs)
-		{
-			row_buffer_print (chunk, iter[-1].attrs);
-			insertion_point = chunk;
-		}
-		*insertion_point++ = iter->c;
-		*insertion_point = 0;
-	}
-	row_buffer_print (chunk, self->chars[self->chars_len - 1].attrs);
-}

liberty-xdg.c

@@ -0,0 +1,772 @@
+/*
+ * liberty-xdg.c: the ultimate C unlibrary: freedesktop.org specifications
+ *
+ * Copyright (c) 2023 - 2024, Přemysl Eric Janouch <p@janouch.name>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
+ * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
+ * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
+ * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ */
+
+// This files assumes you've already included liberty.c.
+
+#ifdef LIBERTY_XDG_WANT_X11
+#include <X11/Xatom.h>
+#include <X11/Xlib.h>
+#endif
+
+#ifdef LIBERTY_XDG_WANT_ICONS
+#include <png.h>
+#endif
+
+// --- XSettings ---------------------------------------------------------------
+
+#ifdef LIBERTY_XDG_WANT_X11
+
+struct xdg_xsettings_setting
+{
+	enum xdg_xsettings_type
+	{
+		XDG_XSETTINGS_INTEGER,
+		XDG_XSETTINGS_STRING,
+		XDG_XSETTINGS_COLOR,
+	}
+	type;                               ///< What's stored in the union
+	uint32_t serial;                    ///< Serial of the last change
+	union
+	{
+		int32_t integer;
+		struct str string;
+		struct { uint16_t red, green, blue, alpha; } color;
+	};
+};
+
+static void
+xdg_xsettings_setting_destroy (struct xdg_xsettings_setting *self)
+{
+	if (self->type == XDG_XSETTINGS_STRING)
+		str_free (&self->string);
+	free (self);
+}
+
+// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+
+struct xdg_xsettings
+{
+	struct str_map settings;            ///< Name -> xdg_xsettings_setting
+};
+
+static void
+xdg_xsettings_free (struct xdg_xsettings *self)
+{
+	str_map_free (&self->settings);
+}
+
+static struct xdg_xsettings
+xdg_xsettings_make (void)
+{
+	return (struct xdg_xsettings)
+	{
+		.settings =
+			str_map_make ((str_map_free_fn) xdg_xsettings_setting_destroy),
+	};
+}
+
+static void
+xdg_xsettings_update (struct xdg_xsettings *self, Display *dpy)
+{
+	// TODO: We're supposed to lock the server.
+	// TODO: We're supposed to trap X errors.
+	char *selection = xstrdup_printf ("_XSETTINGS_S%d", DefaultScreen (dpy));
+	Window owner
+		= XGetSelectionOwner (dpy, XInternAtom (dpy, selection, True));
+	free (selection);
+	if (!owner)
+		return;
+
+	Atom actual_type = None;
+	int actual_format = 0;
+	unsigned long nitems = 0, bytes_after = 0;
+	unsigned char *buffer = NULL;
+	Atom xsettings = XInternAtom (dpy, "_XSETTINGS_SETTINGS", True);
+	int status = XGetWindowProperty (dpy,
+		owner,
+		xsettings,
+		0L,
+		LONG_MAX,
+		False,
+		xsettings,
+		&actual_type,
+		&actual_format,
+		&nitems,
+		&bytes_after,
+		&buffer);
+	if (status != Success || !buffer)
+		return;
+
+	if (actual_type != xsettings
+	 || actual_format != 8
+	 || nitems < 12)
+		goto fail;
+
+	const struct peeker *peeker = NULL;
+	if (buffer[0] == LSBFirst)
+		peeker = &peeker_le;
+	else if (buffer[0] == MSBFirst)
+		peeker = &peeker_be;
+	else
+		goto fail;
+
+	// We're ignoring the serial for now.
+	uint32_t n_settings = peeker->u32 (buffer + 8);
+	size_t offset = 12;
+	struct str name = str_make ();
+	struct xdg_xsettings_setting *setting = xcalloc (1, sizeof *setting);
+	while (n_settings--)
+	{
+		if (nitems < offset + 4)
+			goto fail_item;
+
+		setting->type = buffer[offset];
+		uint16_t name_len = peeker->u16 (buffer + offset + 2);
+		offset += 4;
+		if (nitems < offset + name_len)
+			goto fail_item;
+
+		str_append_data (&name, buffer + offset, name_len);
+		offset += ((name_len + 3) & ~3);
+		if (nitems < offset + 4)
+			goto fail_item;
+
+		setting->serial = peeker->u32 (buffer + offset);
+		offset += 4;
+		switch (setting->type)
+		{
+		case XDG_XSETTINGS_INTEGER:
+			if (nitems < offset + 4)
+				goto fail_item;
+
+			setting->integer = (int32_t) peeker->u32 (buffer + offset);
+			offset += 4;
+			break;
+		case XDG_XSETTINGS_STRING:
+		{
+			if (nitems < offset + 4)
+				goto fail_item;
+
+			uint32_t value_len = peeker->u32 (buffer + offset);
+			offset += 4;
+			if (nitems < offset + value_len)
+				goto fail_item;
+
+			setting->string = str_make ();
+			str_append_data (&setting->string, buffer + offset, value_len);
+			offset += ((value_len + 3) & ~3);
+			break;
+		}
+		case XDG_XSETTINGS_COLOR:
+			if (nitems < offset + 8)
+				goto fail_item;
+
+			setting->color.red   = peeker->u16 (buffer + offset);
+			setting->color.green = peeker->u16 (buffer + offset + 2);
+			setting->color.blue  = peeker->u16 (buffer + offset + 4);
+			setting->color.alpha = peeker->u16 (buffer + offset + 6);
+			offset += 8;
+			break;
+		default:
+			goto fail_item;
+		}
+
+		// TODO(p): Change detection, by comparing existence and serials.
+		str_map_set (&self->settings, name.str, setting);
+		setting = xcalloc (1, sizeof *setting);
+		str_reset (&name);
+	}
+fail_item:
+	xdg_xsettings_setting_destroy (setting);
+	str_free (&name);
+fail:
+	XFree (buffer);
+}
+
+#endif // LIBERTY_XDG_WANT_X11
+
+// --- Desktop file parser -----------------------------------------------------
+
+// Useful for parsing desktop-entry-spec, icon-theme-spec, trash-spec,
+// mime-apps-spec.  This code is not designed for making changes to the files.
+
+struct desktop_file
+{
+	struct str_map groups;              ///< Group name → Key → Value
+};
+
+static void
+desktop_file_free_group (void *value)
+{
+	str_map_free (value);
+	free (value);
+}
+
+static void
+desktop_file_free (struct desktop_file *self)
+{
+	str_map_free (&self->groups);
+}
+
+// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+
+static void
+desktop_file_parse_line (struct desktop_file *self,
+	char **group_name, const char *line, const char *end)
+{
+	struct str_map *group = NULL;
+	if (*group_name)
+		group = str_map_find (&self->groups, *group_name);
+
+	if (*line == '[')
+	{
+		bool ok = *--end == ']';
+		for (const char *p = ++line; ok && p != end; p++)
+			ok = (unsigned char) *p >= 32 && (unsigned char) *p <= 127
+				&& *p != '[' && *p != ']';
+		if (!ok)
+		{
+			cstr_set (group_name, NULL);
+			print_debug ("invalid desktop file group header");
+			return;
+		}
+
+		cstr_set (group_name, xstrndup (line, end - line));
+		if (str_map_find (&self->groups, *group_name))
+		{
+			print_debug ("duplicate desktop file group: %s", *group_name);
+			return;
+		}
+
+		group = xcalloc (1, sizeof *group);
+		*group = str_map_make (free);
+		str_map_set (&self->groups, *group_name, group);
+		return;
+	}
+	if (!group)
+	{
+		print_debug ("unexpected desktop file entry outside of a group");
+		return;
+	}
+
+	const char *key_end = line;
+	while (key_end != end && (isalnum_ascii (*key_end) || *key_end == '-'))
+		key_end++;
+
+	// We could validate these further, but we just search in them anyway.
+	if (key_end != end && *key_end == '[')
+	{
+		while (++key_end != end && *key_end != ']')
+			;
+		if (key_end != end && *key_end == ']')
+			key_end++;
+	}
+
+	const char *value = key_end;
+	while (value != end && *value == ' ')
+		value++;
+	if (value == end || *value++ != '=')
+	{
+		print_debug ("invalid desktop file entry");
+		return;
+	}
+	while (value != end && *value == ' ')
+		value++;
+
+	char *key = xstrndup (line, key_end - line);
+	if (str_map_find (group, key))
+		print_debug ("duplicate desktop file entry for: %s", key);
+	else
+		str_map_set (group, key, xstrndup (value, end - value));
+	free (key);
+}
+
+static struct desktop_file
+desktop_file_make (const char *data, size_t len)
+{
+	struct desktop_file self = (struct desktop_file)
+		{ .groups = str_map_make (desktop_file_free_group) };
+
+	char *group_name = NULL;
+	const char *p = data, *data_end = p + len;
+	while (p != data_end)
+	{
+		const char *line = p, *line_end = line;
+		while (line_end != data_end && *line_end != '\n')
+			line_end++;
+		if ((p = line_end) != data_end && *p == '\n')
+			p++;
+
+		if (line != line_end && *line != '#')
+			desktop_file_parse_line (&self, &group_name, line, line_end);
+	}
+	free (group_name);
+	return self;
+}
+
+static const char *
+desktop_file_get (struct desktop_file *self, const char *group, const char *key)
+{
+	// TODO(p): Ideally, also implement localised keys.
+	struct str_map *group_map = str_map_find (&self->groups, group);
+	if (!group_map)
+		return NULL;
+
+	return str_map_find (group_map, key);
+}
+
+static struct strv
+desktop_file_unescape (const char *value, bool is_list)
+{
+	struct strv result = strv_make ();
+	struct str s = str_make ();
+
+	// XXX: The unescaping behaviour is underspecified.
+	//   It might make sense to warn about unrecognised escape sequences.
+	bool escape = false;
+	for (const char *p = value; *p; p++)
+	{
+		if (escape)
+		{
+			switch (*p)
+			{
+			break; case 's': str_append_c (&s, ' ');
+			break; case 'n': str_append_c (&s, '\n');
+			break; case 't': str_append_c (&s, '\t');
+			break; case 'r': str_append_c (&s, '\r');
+			break; default:  str_append_c (&s, *p);
+			}
+			escape = false;
+		}
+		else if (*p == '\\' && p[1])
+			escape = true;
+		else if (*p == ';' && is_list)
+		{
+			strv_append_owned (&result, str_steal (&s));
+			s = str_make ();
+		}
+		else
+			str_append_c (&s, *p);
+	}
+
+	if (!is_list || s.len != 0)
+		strv_append_owned (&result, str_steal (&s));
+	else
+		str_free (&s);
+	return result;
+}
+
+static char *
+desktop_file_get_string (struct desktop_file *self,
+	const char *group, const char *key)
+{
+	const char *value = desktop_file_get (self, group, key);
+	if (!value)
+		return NULL;
+
+	struct strv values = desktop_file_unescape (value, false /* is_list */);
+	char *unescaped = strv_steal (&values, 0);
+	strv_free (&values);
+	return unescaped;
+}
+
+static struct strv
+desktop_file_get_stringv (struct desktop_file *self,
+	const char *group, const char *key)
+{
+	const char *value = desktop_file_get (self, group, key);
+	if (!value)
+		return strv_make ();
+
+	return desktop_file_unescape (value, true /* is_list */);
+}
+
+static bool
+desktop_file_get_bool (struct desktop_file *self,
+	const char *group, const char *key)
+{
+	const char *value = desktop_file_get (self, group, key);
+	if (!value)
+		return false;
+
+	// Let's be compatible with pre-1.0 files when it costs us so little.
+	if (!strcmp (value, "true")
+	 || !strcmp (value, "1"))
+		return true;
+	if (!strcmp (value, "false")
+	 || !strcmp (value, "0"))
+		return false;
+
+	print_debug ("invalid desktop file boolean for '%s': %s", key, value);
+	return false;
+}
+
+// Nothing uses the "numeric" type.
+// "icon-theme-spec" uses "integer" and doesn't say what it is.
+static long
+desktop_file_get_integer (struct desktop_file *self,
+	const char *group, const char *key)
+{
+	const char *value = desktop_file_get (self, group, key);
+	if (!value)
+		return 0;
+
+	char *end = NULL;
+	long parsed = (errno = 0, strtol (value, &end, 10));
+	if (errno != 0 || *end)
+		print_debug ("invalid desktop file integer for '%s': %s", key, value);
+	return parsed;
+}
+
+// --- Icon themes -------------------------------------------------------------
+
+// This implements part of the Icon Theme Specification.
+
+#ifdef LIBERTY_XDG_WANT_ICONS
+
+struct icon_theme_icon
+{
+	uint32_t width;                     ///< Width of argb in pixels
+	uint32_t height;                    ///< Height of argb in pixels
+	uint32_t argb[];                    ///< ARGB32 data, unassociated alpha
+};
+
+static void
+icon_theme_open_on_error (png_structp pngp, const char *error)
+{
+	print_debug ("%s: %s", (const char *) png_get_error_ptr (pngp), error);
+	png_longjmp (pngp, 1);
+}
+
+static void
+icon_theme_open_on_warning (png_structp pngp, const char *warning)
+{
+	(void) pngp;
+	(void) warning;
+	// Fuck your "gamma value does not match libpng estimate".
+}
+
+// For simplicity, only support PNG icons, using the most popular library.
+static struct icon_theme_icon *
+icon_theme_open (const char *path)
+{
+	volatile png_bytep buffer = NULL;
+	volatile png_bytepp row_pointers = NULL;
+	struct icon_theme_icon *volatile result = NULL;
+	FILE *fp = fopen (path, "rb");
+	if (!fp)
+	{
+		if (errno != ENOENT)
+			print_debug ("%s: %s", path, strerror (errno));
+		return NULL;
+	}
+
+	// The simplified and high-level APIs aren't powerful enough.
+	png_structp pngp = png_create_read_struct (PNG_LIBPNG_VER_STRING,
+		(png_voidp) path, icon_theme_open_on_error, icon_theme_open_on_warning);
+	png_infop infop = png_create_info_struct (pngp);
+	if (!infop)
+	{
+		print_debug ("%s: %s", path, strerror (errno));
+		goto fail;
+	}
+	if (setjmp (png_jmpbuf (pngp)))
+		goto fail;
+
+	png_init_io (pngp, fp);
+	png_read_info (pngp, infop);
+
+	// Asking for at least 8-bit channels.  This call is a superset of:
+	//  - png_set_palette_to_rgb(),
+	//  - png_set_tRNS_to_alpha(),
+	//  - png_set_expand_gray_1_2_4_to_8().
+	png_set_expand (pngp);
+
+	// Reduce the possibilities further to RGB or RGBA...
+	png_set_gray_to_rgb (pngp);
+
+	// ...and /exactly/ 8-bit channels.
+	// Alternatively, use png_set_expand_16() above to obtain 16-bit channels.
+	png_set_scale_16 (pngp);
+
+	// PNG uses RGBA order, let's change that to ARGB (both in memory order).
+	// This doesn't change a row's `color_type` in png_do_read_filler(),
+	// and the following transformation thus ignores it.
+	png_set_add_alpha (pngp, 0xFFFF, PNG_FILLER_BEFORE);
+	png_set_swap_alpha (pngp);
+
+	(void) png_set_interlace_handling (pngp);
+
+	png_read_update_info (pngp, infop);
+	if (png_get_bit_depth (pngp, infop) != 8
+	 || png_get_channels (pngp, infop) != 4
+	 || png_get_color_type (pngp, infop) != PNG_COLOR_TYPE_RGB_ALPHA)
+		png_error (pngp, "result not A8R8G8B8");
+
+	size_t row_bytes = png_get_rowbytes (pngp, infop);
+	size_t height = png_get_image_height (pngp, infop);
+	buffer = xcalloc (row_bytes, height);
+	row_pointers = xcalloc (height, sizeof buffer);
+	for (size_t y = 0; y < height; y++)
+		row_pointers[y] = buffer + y * row_bytes;
+
+	png_read_image (pngp, row_pointers);
+
+	result = xcalloc (1, sizeof *result + row_bytes * height);
+	result->width = png_get_image_width (pngp, infop);
+	result->height = height;
+
+	uint32_t *dst = (uint32_t *) result->argb, *src = (uint32_t *) buffer;
+	for (size_t pixels = result->width * result->height; pixels--; )
+		*dst++ = ntohl (*src++);
+
+fail:
+	free (buffer);
+	free (row_pointers);
+	png_destroy_read_struct (&pngp, &infop, NULL);
+	fclose (fp);
+	return result;
+}
+
+// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+
+struct icon_theme_find_context
+{
+	struct strv base;                   ///< Base directories
+	struct str_map visited;             ///< Cycle prevention
+
+	ARRAY (struct icon_theme_icon *, icons)
+};
+
+static void
+icon_theme_find__fallback (struct icon_theme_find_context *ctx,
+	const char *name)
+{
+	for (size_t i = 0; i < ctx->base.len; i++)
+	{
+		char *path = xstrdup_printf ("%s/%s.png", ctx->base.vector[i], name);
+		struct icon_theme_icon *icon = icon_theme_open (path);
+		free (path);
+		if (icon)
+		{
+			ARRAY_RESERVE (ctx->icons, 1);
+			ctx->icons[ctx->icons_len++] = icon;
+			return;
+		}
+	}
+}
+
+static struct desktop_file
+icon_theme_find__index (struct icon_theme_find_context *ctx, const char *theme)
+{
+	struct str data = str_make ();
+	for (size_t i = 0; i < ctx->base.len; i++)
+	{
+		struct error *e = NULL;
+		char *path = xstrdup_printf ("%s/%s/index.theme",
+			ctx->base.vector[i], theme);
+		read_file (path, &data, &e);
+		free (path);
+		if (!e)
+			break;
+
+		if (errno != ENOENT)
+			print_debug ("%s", e->message);
+		error_free (e);
+	}
+
+	struct desktop_file index = desktop_file_make (data.str, data.len);
+	str_free (&data);
+	return index;
+}
+
+static void
+icon_theme_find__named (struct icon_theme_find_context *ctx,
+	const char *theme, const char *name)
+{
+	// Either a cycle, or a common ancestor of inherited themes, which is valid.
+	if (str_map_find (&ctx->visited, theme))
+		return;
+
+	str_map_set (&ctx->visited, theme, (void *) (intptr_t) 1);
+	struct desktop_file index = icon_theme_find__index (ctx, theme);
+
+	char *directories =
+		desktop_file_get_string (&index, "Icon Theme", "Directories");
+	if (!directories)
+		goto out;
+
+	// NOTE: The sizes are not deduplicated, and priorities are uncertain.
+	struct strv dirs = strv_make ();
+	cstr_split (directories, ",", true, &dirs);
+	free (directories);
+	for (size_t d = 0; d < dirs.len; d++)
+	{
+		// The hicolor icon theme stuffs everything in Directories.
+		if (desktop_file_get (&index, dirs.vector[d], "Scale")
+		 && desktop_file_get_integer (&index, dirs.vector[d], "Scale") != 1)
+			continue;
+
+		for (size_t i = 0; i < ctx->base.len; i++)
+		{
+			char *path = xstrdup_printf ("%s/%s/%s/%s.png",
+				ctx->base.vector[i], theme, dirs.vector[d], name);
+			struct icon_theme_icon *icon = icon_theme_open (path);
+			free (path);
+			if (icon)
+			{
+				ARRAY_RESERVE (ctx->icons, 1);
+				ctx->icons[ctx->icons_len++] = icon;
+				break;
+			}
+		}
+	}
+	strv_free (&dirs);
+	if (ctx->icons_len)
+		goto out;
+
+	char *inherits =
+		desktop_file_get_string (&index, "Icon Theme", "Inherits");
+	if (inherits)
+	{
+		struct strv parents = strv_make ();
+		cstr_split (inherits, ",", true, &parents);
+		free (inherits);
+		for (size_t i = 0; i < parents.len; i++)
+		{
+			icon_theme_find__named (ctx, parents.vector[i], name);
+			if (ctx->icons_len)
+				break;
+		}
+		strv_free (&parents);
+	}
+
+out:
+	desktop_file_free (&index);
+}
+
+/// Return all base directories appropriate for icon search.
+static struct strv
+icon_theme_get_base_directories (void)
+{
+	struct strv dirs = strv_make ();
+	struct str icons = str_make ();
+	(void) str_append_env_path (&icons, "HOME", false);
+	str_append (&icons, "/.icons");
+	strv_append_owned (&dirs, str_steal (&icons));
+
+	// Note that we use XDG_CONFIG_HOME as well, which might be intended.
+	struct strv xdg = strv_make ();
+	get_xdg_data_dirs (&xdg);
+	for (size_t i = 0; i < xdg.len; i++)
+		strv_append_owned (&dirs, xstrdup_printf ("%s/icons", xdg.vector[i]));
+	strv_free (&xdg);
+
+	strv_append (&dirs, "/usr/share/pixmaps");
+	return dirs;
+}
+
+static int
+icon_theme_find__compare (const void *a, const void *b)
+{
+	const struct icon_theme_icon **ia = (const struct icon_theme_icon **) a;
+	const struct icon_theme_icon **ib = (const struct icon_theme_icon **) b;
+	double pa = (double) (*ia)->width * (*ia)->height;
+	double pb = (double) (*ib)->width * (*ib)->height;
+	return (pa > pb) - (pa < pb);
+}
+
+/// Return all sizes of the named icon.  When the theme name is not NULL,
+/// use it as the preferred theme.  Always consult fallbacks locations.
+/// Ignore icon scales other than 1.
+static struct icon_theme_icon **
+icon_theme_find (const char *theme, const char *name, size_t *len)
+{
+	struct icon_theme_find_context ctx = {};
+	ctx.base = icon_theme_get_base_directories ();
+	ctx.visited = str_map_make (NULL);
+	ARRAY_INIT (ctx.icons);
+
+	if (theme)
+		icon_theme_find__named (&ctx, theme, name);
+	if (!ctx.icons_len)
+		icon_theme_find__named (&ctx, "hicolor", name);
+	if (!ctx.icons_len)
+		icon_theme_find__fallback (&ctx, name);
+
+	strv_free (&ctx.base);
+	str_map_free (&ctx.visited);
+
+	ARRAY_RESERVE (ctx.icons, 1);
+	ctx.icons[ctx.icons_len] = NULL;
+	if (!ctx.icons_len)
+	{
+		free (ctx.icons);
+		return NULL;
+	}
+
+	qsort (ctx.icons,
+		ctx.icons_len, sizeof *ctx.icons, icon_theme_find__compare);
+	*len = ctx.icons_len;
+	return ctx.icons;
+}
+
+static void
+icon_theme_free (struct icon_theme_icon **icons)
+{
+	for (struct icon_theme_icon **p = icons; *p; p++)
+		free (*p);
+	free (icons);
+}
+
+// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+
+#ifdef LIBERTY_XDG_WANT_X11
+
+static void
+icon_theme_set_window_icon (Display *dpy,
+	Window window, const char *theme, const char *name)
+{
+	size_t icons_len = 0;
+	struct icon_theme_icon **icons = icon_theme_find (theme, name, &icons_len);
+	if (!icons)
+		return;
+
+	size_t n = 0;
+	for (size_t i = 0; i < icons_len; i++)
+		n += 2 + icons[i]->width * icons[i]->height;
+
+	unsigned long *data = xcalloc (n, sizeof *data), *p = data;
+	for (size_t i = 0; i < icons_len; i++)
+	{
+		*p++ = icons[i]->width;
+		*p++ = icons[i]->height;
+
+		uint32_t *q = icons[i]->argb;
+		for (size_t k = icons[i]->width * icons[i]->height; k--; )
+			*p++ = *q++;
+	}
+
+	XChangeProperty (dpy, window, XInternAtom (dpy, "_NET_WM_ICON", False),
+		XA_CARDINAL, 32, PropModeReplace, (const unsigned char *) data, n);
+	free (data);
+	icon_theme_free (icons);
+}
+
+#endif // LIBERTY_XDG_WANT_X11
+#endif // LIBERTY_XDG_WANT_ICONS

liberty.c

@@ -1,7 +1,7 @@
 /*
  * liberty.c: the ultimate C unlibrary
  *
- * Copyright (c) 2014 - 2020, Přemysl Eric Janouch <p@janouch.name>
+ * Copyright (c) 2014 - 2024, Přemysl Eric Janouch <p@janouch.name>
  *
  * Permission to use, copy, modify, and/or distribute this software for any
  * purpose with or without fee is hereby granted.
@@ -16,8 +16,13 @@
  *
  */
 
-#define _POSIX_C_SOURCE 199309L
+#ifndef _POSIX_C_SOURCE
+#define _POSIX_C_SOURCE 200112L
+#endif
+
+#ifndef _XOPEN_SOURCE
 #define _XOPEN_SOURCE 600
+#endif
 
 #include <stdio.h>
 #include <stddef.h>
@@ -287,7 +292,8 @@ xreallocarray (void *o, size_t n, size_t m)
 static char *
 xstrdup (const char *s)
 {
-	return strcpy (xmalloc (strlen (s) + 1), s);
+	size_t len = strlen (s) + 1;
+	return memcpy (xmalloc (len), s, len);
 }
 
 static char *
@@ -647,6 +653,60 @@ str_pack_u64 (struct str *self, uint64_t x)
 #define str_pack_i32(self, x)  str_pack_u32 ((self), (uint32_t) (x))
 #define str_pack_i64(self, x)  str_pack_u64 ((self), (uint64_t) (x))
 
+// --- Reading binary numbers --------------------------------------------------
+
+// Doing this byte by byte prevents unaligned memory access issues.
+
+static uint64_t
+peek_u64be (const uint8_t *p)
+{
+	return (uint64_t) p[0] << 56 | (uint64_t) p[1] << 48
+		| (uint64_t) p[2] << 40 | (uint64_t) p[3] << 32
+		| (uint64_t) p[4] << 24 | (uint64_t) p[5] << 16 | p[6] << 8 | p[7];
+}
+
+static uint32_t
+peek_u32be (const uint8_t *p)
+{
+	return (uint32_t) p[0] << 24 | (uint32_t) p[1] << 16 | p[2] << 8 | p[3];
+}
+
+static uint16_t
+peek_u16be (const uint8_t *p)
+{
+	return (uint16_t) p[0] << 8 | p[1];
+}
+
+static uint64_t
+peek_u64le (const uint8_t *p)
+{
+	return (uint64_t) p[7] << 56 | (uint64_t) p[6] << 48
+		| (uint64_t) p[5] << 40 | (uint64_t) p[4] << 32
+		| (uint64_t) p[3] << 24 | (uint64_t) p[2] << 16 | p[1] << 8 | p[0];
+}
+
+static uint32_t
+peek_u32le (const uint8_t *p)
+{
+	return (uint32_t) p[3] << 24 | (uint32_t) p[2] << 16 | p[1] << 8 | p[0];
+}
+
+static uint16_t
+peek_u16le (const uint8_t *p)
+{
+	return (uint16_t) p[1] << 8 | p[0];
+}
+
+struct peeker
+{
+	uint64_t (*u64) (const uint8_t *);
+	uint32_t (*u32) (const uint8_t *);
+	uint16_t (*u16) (const uint8_t *);
+};
+
+static const struct peeker peeker_be = {peek_u64be, peek_u32be, peek_u16be};
+static const struct peeker peeker_le = {peek_u64le, peek_u32le, peek_u16le};
+
 // --- Errors ------------------------------------------------------------------
 
 // Error reporting utilities.  Inspired by GError, only much simpler.
@@ -1749,10 +1809,9 @@ poller_run (struct poller *self)
 	self->revents_len = 0;
 }
 
-#elif defined (BSD)
-
-// Mac OS X's kqueue is fatally broken, or so I've been told; leaving it out.
-// Otherwise this is sort of similar to the epoll version.
+// Sort of similar to the epoll version.  Let's hope Darwin isn't broken,
+// that'd mean reimplementing this in terms of select() just because of Crapple.
+#elif defined (BSD) || defined (__APPLE__)
 
 #include <sys/types.h>
 #include <sys/event.h>
@@ -2451,7 +2510,7 @@ write_queue_processed (struct write_queue *self, size_t len)
 }
 
 static bool
-write_queue_is_empty (struct write_queue *self)
+write_queue_is_empty (const struct write_queue *self)
 {
 	return self->head == NULL;
 }
@@ -2499,12 +2558,7 @@ msg_reader_get (struct msg_reader *self, size_t *len)
 		return NULL;
 
 	uint8_t *x = (uint8_t *) self->buf.str + self->offset;
-	uint64_t msg_len
-		= (uint64_t) x[0] << 56 | (uint64_t) x[1] << 48
-		| (uint64_t) x[2] << 40 | (uint64_t) x[3] << 32
-		| (uint64_t) x[4] << 24 | (uint64_t) x[5] << 16
-		| (uint64_t) x[6] << 8  | (uint64_t) x[7];
-
+	uint64_t msg_len = peek_u64be (x);
 	if (msg_len < sizeof msg_len)
 	{
 		// The message is shorter than its header
@@ -2546,7 +2600,7 @@ msg_unpacker_make (const void *data, size_t len)
 }
 
 static size_t
-msg_unpacker_get_available (struct msg_unpacker *self)
+msg_unpacker_get_available (const struct msg_unpacker *self)
 {
 	return self->len - self->offset;
 }
@@ -2569,8 +2623,7 @@ static bool
 msg_unpacker_u16 (struct msg_unpacker *self, uint16_t *value)
 {
 	UNPACKER_INT_BEGIN
-	*value
-		= (uint16_t) x[0] << 8  | (uint16_t) x[1];
+	*value = peek_u16be (x);
 	return true;
 }
 
@@ -2578,9 +2631,7 @@ static bool
 msg_unpacker_u32 (struct msg_unpacker *self, uint32_t *value)
 {
 	UNPACKER_INT_BEGIN
-	*value
-		= (uint32_t) x[0] << 24 | (uint32_t) x[1] << 16
-		| (uint32_t) x[2] << 8  | (uint32_t) x[3];
+	*value = peek_u32be (x);
 	return true;
 }
 
@@ -2588,11 +2639,7 @@ static bool
 msg_unpacker_u64 (struct msg_unpacker *self, uint64_t *value)
 {
 	UNPACKER_INT_BEGIN
-	*value
-		= (uint64_t) x[0] << 56 | (uint64_t) x[1] << 48
-		| (uint64_t) x[2] << 40 | (uint64_t) x[3] << 32
-		| (uint64_t) x[4] << 24 | (uint64_t) x[5] << 16
-		| (uint64_t) x[6] << 8  | (uint64_t) x[7];
+	*value = peek_u64be (x);
 	return true;
 }
 
@@ -2748,16 +2795,12 @@ utf8_decode (const char **s, size_t len)
 	}
 
 	// In the middle of a character
-	if (sequence_len == 1)
+	// or an overlong sequence (subset, possibly MUTF-8, not supported)
+	if (sequence_len == 1 || *p == 0xC0 || *p == 0xC1)
 		return -1;
 
 	// Check the rest of the sequence
 	uint32_t cp = *p++ & ~mask;
-
-	// Overlong sequence (possibly MUTF-8, not supported)
-	if (!cp && sequence_len)
-		return -1;
-
 	while (sequence_len && --sequence_len)
 	{
 		if (p == end)
@@ -3251,7 +3294,7 @@ get_xdg_config_dirs (struct strv *out)
 	str_free (&config_home);
 
 	const char *xdg_config_dirs;
-	if (!(xdg_config_dirs = getenv ("XDG_CONFIG_DIRS")))
+	if (!(xdg_config_dirs = getenv ("XDG_CONFIG_DIRS")) || !*xdg_config_dirs)
 		xdg_config_dirs = "/etc/xdg";
 	cstr_split (xdg_config_dirs, ":", true, out);
 }
@@ -3276,7 +3319,7 @@ get_xdg_data_dirs (struct strv *out)
 	str_free (&data_home);
 
 	const char *xdg_data_dirs;
-	if (!(xdg_data_dirs = getenv ("XDG_DATA_DIRS")))
+	if (!(xdg_data_dirs = getenv ("XDG_DATA_DIRS")) || !*xdg_data_dirs)
 		xdg_data_dirs = "/usr/local/share/:/usr/share/";
 	cstr_split (xdg_data_dirs, ":", true, out);
 }
@@ -3552,6 +3595,8 @@ write_file_safe (const char *filename, const void *data, size_t data_len,
 {
 	// XXX: ideally we would also open the directory, use *at() versions
 	//   of functions and call fsync() on the directory as appropriate
+	// FIXME: this should behave similarly to mkstemp(), just with 0666;
+	//   as it is, this function is not particularly safe
 	char *temp = xstrdup_printf ("%s.new", filename);
 	bool success = write_file (temp, data, data_len, e);
 	if (success && !(success = !rename (temp, filename)))
@@ -4417,11 +4462,13 @@ socket_io_try_write (int socket_fd, struct str *wb)
 //   object  = lws '{' entries endobj
 //   endobj  = lws '}'
 //
-//   string  = lws '"' ('\\' escape / ![\\"] char)* '"'
+//   quoted  = lws '"' (!["\\] char / '\\' escape)* '"'
+//           / lws '`' (![`] char)* '`'
+//   string  = (quoted)+
 //   char    = [\0-\177]  # or any Unicode codepoint in the UTF-8 encoding
 //   escape  = [\\"abfnrtv] / [xX][0-9A-Fa-f][0-9A-Fa-f]? / [0-7][0-7]?[0-7]?
 //
-//   integer = lws '-'? [0-9]+  # whatever strtoll() accepts on your system
+//   integer = lws [-+]? [0-9]+  # whatever strtoll() accepts on your system
 //   null    = lws 'null'
 //   boolean = lws 'yes'  / lws 'YES'  / lws 'no'    / lws 'NO'
 //           / lws 'on'   / lws 'ON'   / lws 'off'   / lws 'OFF'
@@ -4712,13 +4759,15 @@ config_item_write_string (struct str *output, const struct str *s)
 	for (size_t i = 0; i < s->len; i++)
 	{
 		unsigned char c = s->str[i];
-		if      (c == '\n')  str_append        (output, "\\n");
-		else if (c == '\r')  str_append        (output, "\\r");
-		else if (c == '\t')  str_append        (output, "\\t");
-		else if (c == '\\')  str_append        (output, "\\\\");
-		else if (c == '"')   str_append        (output, "\\\"");
-		else if (c < 32)     str_append_printf (output, "\\x%02x", c);
-		else                 str_append_c      (output, c);
+		if      (c == '\n')  str_append (output, "\\n");
+		else if (c == '\r')  str_append (output, "\\r");
+		else if (c == '\t')  str_append (output, "\\t");
+		else if (c == '\\')  str_append (output, "\\\\");
+		else if (c == '"')   str_append (output, "\\\"");
+		else if (iscntrl_ascii (c))
+			str_append_printf (output, "\\x%02x", c);
+		else
+			str_append_c (output, c);
 	}
 	str_append_c (output, '"');
 }
@@ -5042,10 +5091,10 @@ config_tokenizer_escape_sequence
 }
 
 static bool
-config_tokenizer_string
-	(struct config_tokenizer *self, struct str *output, struct error **e)
+config_tokenizer_dq_string (struct config_tokenizer *self, struct str *output,
+	struct error **e)
 {
-	unsigned char c;
+	unsigned char c = config_tokenizer_advance (self);
 	while (self->len)
 	{
 		if ((c = config_tokenizer_advance (self)) == '"')
@@ -5059,6 +5108,44 @@ config_tokenizer_string
 	return false;
 }
 
+static bool
+config_tokenizer_bt_string (struct config_tokenizer *self, struct str *output,
+	struct error **e)
+{
+	unsigned char c = config_tokenizer_advance (self);
+	while (self->len)
+	{
+		if ((c = config_tokenizer_advance (self)) == '`')
+			return true;
+		str_append_c (output, c);
+	}
+	config_tokenizer_error (self, e, "premature end of string");
+	return false;
+}
+
+static bool
+config_tokenizer_string (struct config_tokenizer *self, struct str *output,
+	struct error **e)
+{
+	// Go-like strings, with C/AWK-like automatic concatenation
+	while (self->len)
+	{
+		bool ok = true;
+		if (isspace_ascii (*self->p) && *self->p != '\n')
+			config_tokenizer_advance (self);
+		else if (*self->p == '"')
+			ok = config_tokenizer_dq_string (self, output, e);
+		else if (*self->p == '`')
+			ok = config_tokenizer_bt_string (self, output, e);
+		else
+			break;
+
+		if (!ok)
+			return false;
+	}
+	return true;
+}
+
 static enum config_token
 config_tokenizer_next (struct config_tokenizer *self, struct error **e)
 {
@@ -5083,7 +5170,7 @@ config_tokenizer_next (struct config_tokenizer *self, struct error **e)
 		return CONFIG_T_ABORT;
 
 	case '"':
-		config_tokenizer_advance (self);
+	case '`':
 		str_reset (&self->string);
 		if (!config_tokenizer_string (self, &self->string, e))
 			return CONFIG_T_ABORT;

libertyxdr.adoc

@@ -0,0 +1,108 @@
+libertyxdr(7)
+=============
+:doctype: manpage
+
+Name
+----
+LibertyXDR - an XDR-derived IDL and data serialization format
+
+Description
+-----------
+*LibertyXDR* is an interface description language, as well as a data
+serialization format.  It is largely derived from XDR, though notably
+simplified.
+
+Conventions
+~~~~~~~~~~~
+User-defined types should be named in *CamelCase*, field names in *snake_case*,
+and constants in *SCREAMING_SNAKE_CASE*.  Code generators will convert these to
+whatever is appropriate in their target language.
+
+Primitive data types
+~~~~~~~~~~~~~~~~~~~~
+Like in XDR, all data is serialized in the network byte order, i.e., big-endian.
+
+ * *void*: 0 bytes
++
+This is a dummy type that cannot be assigned a field name.
+
+ * *bool*: 1 byte 
++
+This is a boolean value: 0 means _false_, any other value means _true_.
+
+ * *u8*, *u16*, *u32*, *u64*: 1, 2, 4, and 8 bytes respectively
++
+These are unsigned integers.
+
+ * *i8*, *i16*, *i32*, *i64*: 1, 2, 4, and 8 bytes respectively
++
+These are signed integers in two's complement.
+
+ * *string*: implicitly prefixed by its length as a *u32*,
+   then immediately followed by its contents, with no trailing NUL byte
++
+This is a valid UTF-8 string without a byte order mark.  Note that strings are
+always unbounded, unlike in XDR.
+
+Constants
+~~~~~~~~~
+At the top level of a document, outside other definitions, you can define
+typeless integer constants:
+
+ const VERSION = 1;
+
+The value can be either a name of another previously defined constant,
+or an immediate decimal value, which may not contain leading zeros.
+
+Enumerations
+~~~~~~~~~~~~
+An *enum* is an *i8* with uniquely named values, in their own namespace.
+
+Values can be either specified explicitly, in the same way as with a constant,
+or they can be left implicit, in which case names assume a value that is one
+larger than their predecessor.  Zero is reserved for internal use, thus
+enumerations implicitly begin with a value of one.  For example, these form
+a sequence from one to three:
+
+ enum Vehicle { CAR, LORRY = 2, PLANE, };
+
+Structures
+~~~~~~~~~~
+A *struct* is a sequence of fields, specified by their type, and their chosen
+name.  You can add a *<>* suffix to change a field to an array, in which case
+it is implicitly preceded by a *u32* specifying its length in terms of its
+elements.
+
+Unlike in XDR, there is no padding between subsequent fields, and type
+definitions can be arbitrarily syntactically nested, as in C.
+
+ struct StockReport {
+   u8 version;       // Version of this report.
+   struct Item {
+     Vehicle kind;   // The vehicle in question.
+     i32 count;      // How many vehicle of that kind there are.
+   } items<>;        // Reported items.
+ };
+
+Unions
+~~~~~~
+A *union* is a kind of structure whose fields depend on the value of its first
+and always-present field, which must be a tag *enum*:
+
+ union VehicleDetails switch (Vehicle kind) {
+ case CAR:   void;
+ case LORRY: i8 axles;
+ case PLANE: i8 engines;
+ };
+
+All possible enumeration values must be named, and there is no *case*
+fall-through.
+
+Framing
+-------
+Unless this role is already filled by, e.g., WebSocket, _LibertyXDR_ structures
+should be prefixed by their byte length in the *u32* format, once serialized.
+
+See also
+--------
+_XDR: External Data Representation Standard_, RFC 4506

libertyxdr.vim

@@ -0,0 +1,21 @@
+" filetype.vim: au! BufNewFile,BufRead *.lxdr setf libertyxdr
+if exists("b:current_syntax")
+	finish
+endif
+
+syn match libertyxdrError "[^[:space:]:;,(){}<>=]\+"
+syn region libertyxdrBlockComment start=+/[*]+ end=+[*]/+
+syn match libertyxdrComment "//.*"
+syn match libertyxdrIdentifier "\<[[:alpha:]][[:alnum:]_]*\>"
+syn match libertyxdrNumber "\<0\>\|\(-\|\<\)[1-9][[:digit:]]*\>"
+syn keyword libertyxdrKeyword const enum struct union switch case
+syn keyword libertyxdrType bool u8 u16 u32 u64 i8 i16 i32 i64 string void
+
+let b:current_syntax = "libertyxdr"
+hi def link libertyxdrError Error
+hi def link libertyxdrBlockComment Comment
+hi def link libertyxdrComment Comment
+hi def link libertyxdrIdentifier Identifier
+hi def link libertyxdrNumber Number
+hi def link libertyxdrKeyword Statement
+hi def link libertyxdrType Type

siphash.c

@@ -82,4 +82,3 @@ siphash (const unsigned char key[16], const unsigned char *m, size_t len)
 
 	return v0 ^ v1 ^ v2 ^ v3;
 }
-

tests/fuzz.c

@@ -32,13 +32,6 @@
 #define LIBERTY_WANT_PROTO_MPD
 
 #include "../liberty.c"
-#include "../liberty-tui.c"
-
-static bool
-app_is_character_in_locale (ucs4_t ch)
-{
-	return ch < 128;
-}
 
 // --- UTF-8 -------------------------------------------------------------------
 
@@ -211,19 +204,6 @@ test_config_item_parse (const uint8_t *data, size_t size)
 		config_item_destroy (item);
 }
 
-// --- TUI ---------------------------------------------------------------------
-
-static void
-test_attrs_decode (const uint8_t *data, size_t size)
-{
-	struct str wrap = str_make ();
-	str_append_data (&wrap, data, size);
-
-	attrs_decode (wrap.str);
-
-	str_free (&wrap);
-}
-
 // --- MPD ---------------------------------------------------------------------
 
 static void
@@ -266,7 +246,6 @@ LLVMFuzzerInitialize (int *argcp, char ***argvp)
 	REGISTER (fcgi_parser_push)
 	REGISTER (fcgi_nv_parser_push)
 	REGISTER (config_item_parse)
-	REGISTER (attrs_decode)
 	REGISTER (mpd_client_process_input)
 
 	char **argv = *argvp, *option = "-test=", *name = NULL;

tests/liberty.c

@@ -1,7 +1,7 @@
 /*
  * tests/liberty.c
  *
- * Copyright (c) 2015 - 2016, Přemysl Eric Janouch <p@janouch.name>
+ * Copyright (c) 2015 - 2022, Přemysl Eric Janouch <p@janouch.name>
  *
  * Permission to use, copy, modify, and/or distribute this software for any
  * purpose with or without fee is hereby granted.
@@ -331,10 +331,12 @@ test_utf8 (void)
 	soft_assert (utf8_decode (&partial, 1) == -2);
 	soft_assert (utf8_decode (&empty,   0) == -1);
 
-	const char valid[] = "2H₂ + O₂ ⇌ 2H₂O, R = 4.7 kΩ, ⌀ 200 mm";
+	const char valid_1[] = "2H₂ + O₂ ⇌ 2H₂O, R = 4.7 kΩ, ⌀ 200 mm";
+	const char valid_2[] = "\xf0\x93\x82\xb9";
 	const char invalid_1[] = "\xf0\x90\x28\xbc";
 	const char invalid_2[] = "\xc0\x80";
-	soft_assert ( utf8_validate (valid,     sizeof valid));
+	soft_assert ( utf8_validate (valid_1,   sizeof valid_1));
+	soft_assert ( utf8_validate (valid_2,   sizeof valid_2));
 	soft_assert (!utf8_validate (invalid_1, sizeof invalid_1));
 	soft_assert (!utf8_validate (invalid_2, sizeof invalid_2));
 
@@ -647,7 +649,7 @@ static struct config_schema g_config_test[] =
 	  .default_  = "1" },
 	{ .name      = "foobar",
 	  .type      = CONFIG_ITEM_STRING,
-	  .default_  = "\"qux\\x01\"" },
+	  .default_  = "\"qux\\x01`\" \"\"`a`" },
 	{}
 };
 
@@ -673,6 +675,9 @@ test_config (void)
 		"top.bar", NULL), invalid, NULL));
 	config_item_destroy (invalid);
 
+	hard_assert (!strcmp ("qux\001`a",
+		config_item_get (config.root, "top.foobar", NULL)->value.string.str));
+
 	struct str s = str_make ();
 	config_item_write (config.root, true, &s);
 	struct config_item *parsed = config_item_parse (s.str, s.len, false, NULL);

tests/lxdrgen.c

@@ -0,0 +1,134 @@
+/*
+ * tests/lxdrgen.c
+ *
+ * Copyright (c) 2022, Přemysl Eric Janouch <p@janouch.name>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
+ * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
+ * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
+ * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ */
+
+#define PROGRAM_NAME "test"
+#define PROGRAM_VERSION "0"
+
+#include "../liberty.c"
+#include "lxdrgen.lxdr.c"
+
+static void
+test_ser_deser_free (void)
+{
+	hard_assert (PROTO_GEN_VERSION == 1);
+
+	enum { CASES = 3 };
+
+	struct proto_gen_struct a = {}, b = {};
+	a.u = xcalloc ((a.u_len = CASES + rand () % 100), sizeof *a.u);
+	for (size_t i = 0; i < a.u_len; i++)
+	{
+		union proto_gen_union *u = a.u + i;
+		switch (i % CASES)
+		{
+		case 0:
+			u->tag = PROTO_GEN_ENUM_NUMBERS;
+			u->numbers.a = rand () % UINT8_MAX;
+			u->numbers.b = rand () % UINT16_MAX;
+			u->numbers.c = rand () % UINT32_MAX;
+			u->numbers.d = rand () % UINT64_MAX;
+			u->numbers.e = rand () % UINT8_MAX;
+			u->numbers.f = rand () % UINT16_MAX;
+			u->numbers.g = rand () % UINT32_MAX;
+			u->numbers.h = rand () % UINT64_MAX;
+			break;
+		case 1:
+			u->tag = PROTO_GEN_ENUM_OTHERS;
+			u->others.foo = rand () % 2;
+			u->others.bar = str_make ();
+			for (int i = rand () % 0x30; i > 0; i--)
+				str_append_c (&u->others.bar, 0x30 + i);
+			u->others.baz_len = rand () % 0x30;
+			u->others.baz = xcalloc (1, u->others.baz_len);
+			for (uint32_t i = 0; i < u->others.baz_len; i++)
+				u->others.baz[i] = 0x30 + i;
+			break;
+		case 2:
+			u->tag = PROTO_GEN_ENUM_NOTHING;
+			break;
+		default:
+			hard_assert (!"unhandled case");
+		}
+	}
+
+	a.o.tag = PROTO_GEN_ENUM_NOTHING;
+
+	struct str buf = str_make ();
+	hard_assert (proto_gen_struct_serialize (&a, &buf));
+	struct msg_unpacker r = msg_unpacker_make (buf.str, buf.len);
+	hard_assert (proto_gen_struct_deserialize (&b, &r));
+	hard_assert (!msg_unpacker_get_available (&r));
+	str_free (&buf);
+
+	hard_assert (a.u_len == b.u_len);
+	for (size_t i = 0; i < a.u_len; i++)
+	{
+		union proto_gen_union *ua = a.u + i;
+		union proto_gen_union *ub = b.u + i;
+		hard_assert (ua->tag == ub->tag);
+		switch (ua->tag)
+		{
+		case PROTO_GEN_ENUM_NUMBERS:
+			hard_assert (ua->numbers.a == ub->numbers.a);
+			hard_assert (ua->numbers.b == ub->numbers.b);
+			hard_assert (ua->numbers.c == ub->numbers.c);
+			hard_assert (ua->numbers.d == ub->numbers.d);
+			hard_assert (ua->numbers.e == ub->numbers.e);
+			hard_assert (ua->numbers.f == ub->numbers.f);
+			hard_assert (ua->numbers.g == ub->numbers.g);
+			hard_assert (ua->numbers.h == ub->numbers.h);
+			break;
+		case PROTO_GEN_ENUM_OTHERS:
+			hard_assert (ua->others.foo == ub->others.foo);
+			hard_assert (ua->others.bar.len == ub->others.bar.len);
+			hard_assert (!memcmp (ua->others.bar.str, ub->others.bar.str,
+				ua->others.bar.len));
+			hard_assert (ua->others.baz_len == ub->others.baz_len);
+			hard_assert (!memcmp (ua->others.baz, ub->others.baz,
+				ua->others.baz_len));
+			break;
+		case PROTO_GEN_ENUM_NOTHING:
+			break;
+		default:
+			hard_assert (!"unexpected case");
+		}
+	}
+
+	hard_assert (a.o.tag == b.o.tag);
+
+	// Emulate partially deserialized data to test disposal of that.
+	for (size_t i = b.u_len - CASES; i < b.u_len; i++)
+	{
+		proto_gen_union_free (&b.u[i]);
+		memset (&b.u[i], 0, sizeof b.u[i]);
+	}
+
+	proto_gen_struct_free (&a);
+	proto_gen_struct_free (&b);
+}
+
+int
+main (int argc, char *argv[])
+{
+	struct test test;
+	test_init (&test, argc, argv);
+
+	test_add_simple (&test, "/ser-deser-free", NULL, test_ser_deser_free);
+
+	return test_run (&test);
+}

tests/lxdrgen.cpp

@@ -0,0 +1,132 @@
+/*
+ * tests/lxdrgen.cpp
+ *
+ * Copyright (c) 2023, Přemysl Eric Janouch <p@janouch.name>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
+ * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
+ * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
+ * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ */
+
+#include "lxdrgen.lxdr.cpp"
+
+#include <cstdlib>
+
+static void
+hard_assert (bool condition, const char *description)
+{
+	if (!condition)
+	{
+		fprintf (stderr, "assertion failed: %s\n", description);
+		abort ();
+	}
+}
+
+#define hard_assert(condition) hard_assert (condition, #condition)
+
+int
+main (int argc, char *argv[])
+{
+	hard_assert (ProtoGen::VERSION == 1);
+
+	enum { CASES = 3 };
+
+	ProtoGen::Struct a = {}, b = {};
+	a.u.resize (CASES + rand () % 100);
+	for (size_t i = 0; i < a.u.size (); i++)
+	{
+		std::unique_ptr<ProtoGen::Union> &u = a.u[i];
+		switch (i % CASES)
+		{
+		case 0:
+		{
+			auto numbers = new ProtoGen::Union_Numbers ();
+			numbers->a = rand () % UINT8_MAX;
+			numbers->b = rand () % UINT16_MAX;
+			numbers->c = rand () % UINT32_MAX;
+			numbers->d = rand () % UINT64_MAX;
+			numbers->e = rand () % UINT8_MAX;
+			numbers->f = rand () % UINT16_MAX;
+			numbers->g = rand () % UINT32_MAX;
+			numbers->h = rand () % UINT64_MAX;
+			u.reset (numbers);
+			break;
+		}
+		case 1:
+		{
+			auto others = new ProtoGen::Union_Others ();
+			others->foo = rand () % 2;
+			for (int i = rand () % 0x30; i > 0; i--)
+				others->bar += 0x30 + i;
+			for (int i = rand () % 0x30; i > 0; i--)
+				others->baz.push_back (0x30 + i);
+			u.reset (others);
+			break;
+		}
+		case 2:
+			u.reset (new ProtoGen::Union_Nothing ());
+			break;
+		default:
+			hard_assert (!"unhandled case");
+		}
+	}
+
+	a.o.reset (new ProtoGen::Onion_Nothing ());
+
+	LibertyXDR::Writer buf;
+	hard_assert (a.serialize (buf));
+	LibertyXDR::Reader r;
+	r.data = buf.data.data ();
+	r.length = buf.data.size ();
+	hard_assert (b.deserialize (r));
+	hard_assert (!r.length);
+
+	hard_assert (a.u.size () == b.u.size ());
+	for (size_t i = 0; i < a.u.size (); i++)
+	{
+		ProtoGen::Union *ua = a.u[i].get ();
+		ProtoGen::Union *ub = b.u[i].get ();
+		hard_assert (ua->tag == ub->tag);
+		switch (ua->tag)
+		{
+		case ProtoGen::Enum::NUMBERS:
+		{
+			auto a = dynamic_cast<ProtoGen::Union_Numbers *> (ua);
+			auto b = dynamic_cast<ProtoGen::Union_Numbers *> (ub);
+			hard_assert (a->a == b->a);
+			hard_assert (a->b == b->b);
+			hard_assert (a->c == b->c);
+			hard_assert (a->d == b->d);
+			hard_assert (a->e == b->e);
+			hard_assert (a->f == b->f);
+			hard_assert (a->g == b->g);
+			hard_assert (a->h == b->h);
+			break;
+		}
+		case ProtoGen::Enum::OTHERS:
+		{
+			auto a = dynamic_cast<ProtoGen::Union_Others *> (ua);
+			auto b = dynamic_cast<ProtoGen::Union_Others *> (ub);
+			hard_assert (a->foo == b->foo);
+			hard_assert (a->bar == b->bar);
+			hard_assert (a->baz == b->baz);
+			break;
+		}
+		case ProtoGen::Enum::NOTHING:
+			break;
+		default:
+			hard_assert (!"unexpected case");
+		}
+	}
+
+	hard_assert (a.o->tag == b.o->tag);
+	return 0;
+}

tests/lxdrgen.lxdr

@@ -0,0 +1,29 @@
+/*
+ * tests/lxdrgen.lxdr: a test protocol for the generator
+ */
+const VERSION = 1;
+const NOISREV = -1;
+
+// TODO: Test failure paths, and in general go for full coverage.
+struct Struct {
+	union Union switch (enum Enum {
+		NUMBERS = VERSION,
+		OTHERS = 2,
+		NOTHING,
+	} tag) {
+	case NUMBERS:
+		i8 a; i16 b; i32 c; i64 d;
+		u8 e; u16 f; u32 g; u64 h;
+	case OTHERS:
+		bool foo;
+		string bar;
+		u8 baz<>;
+	case NOTHING:
+		void;
+	} u<>;
+
+	union Onion switch (Enum tag) {
+	case NOTHING:
+		void;
+	} o;
+};

tests/pulse.c

@@ -0,0 +1,127 @@
+/*
+ * tests/pulse.c
+ *
+ * Copyright (c) 2021, Přemysl Eric Janouch <p@janouch.name>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
+ * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
+ * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
+ * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ */
+
+#define PROGRAM_NAME "test"
+#define PROGRAM_VERSION "0"
+
+#define LIBERTY_WANT_POLLER
+
+#include "../liberty.c"
+#include "../liberty-pulse.c"
+
+// --- Tests -------------------------------------------------------------------
+
+enum
+{
+	EVENT_IO    =  1 << 0,
+	EVENT_TIME  =  1 << 1,
+	EVENT_DEFER =  1 << 2,
+	EVENT_ALL   = (1 << 3) - 1
+};
+
+static intptr_t g_events = 0;
+static intptr_t g_destroys = 0;
+
+static void
+io_event_cb (pa_mainloop_api *a,
+	pa_io_event *e, int fd, pa_io_event_flags_t events, void *userdata)
+{
+	(void) a; (void) e; (void) fd; (void) events;
+	g_events |= (intptr_t) userdata;
+}
+
+static void
+io_event_destroy_cb (pa_mainloop_api *a, pa_io_event *e, void *userdata)
+{
+	(void) a; (void) e;
+	g_destroys += (intptr_t) userdata;
+}
+
+static void
+time_event_cb (pa_mainloop_api *a,
+	pa_time_event *e, const struct timeval *tv, void *userdata)
+{
+	(void) a; (void) e; (void) tv;
+	g_events |= (intptr_t) userdata;
+}
+
+static void
+time_event_destroy_cb (pa_mainloop_api *a, pa_time_event *e, void *userdata)
+{
+	(void) a; (void) e;
+	g_destroys += (intptr_t) userdata;
+}
+
+static void
+defer_event_cb (pa_mainloop_api *a, pa_defer_event *e, void *userdata)
+{
+	(void) a; (void) e;
+	g_events |= (intptr_t) userdata;
+}
+
+static void
+defer_event_destroy_cb (pa_mainloop_api *a, pa_defer_event *e, void *userdata)
+{
+	(void) a; (void) e;
+	g_destroys += (intptr_t) userdata;
+}
+
+static void
+test_pulse (void)
+{
+	struct poller poller;
+	poller_init (&poller);
+
+	// Let's just get this over with, not aiming for high test coverage here
+	pa_mainloop_api *api = poller_pa_new (&poller);
+
+	pa_io_event *ie = api->io_new (api, STDOUT_FILENO, PA_IO_EVENT_OUTPUT,
+		io_event_cb, (void *) EVENT_IO);
+	api->io_set_destroy (ie, io_event_destroy_cb);
+
+	const struct timeval tv = poller_pa_get_current_time ();
+	pa_time_event *te = api->time_new (api, &tv,
+		time_event_cb, (void *) EVENT_TIME);
+	api->time_set_destroy (te, time_event_destroy_cb);
+	api->time_restart (te, &tv);
+
+	pa_defer_event *de = api->defer_new (api,
+		defer_event_cb, (void *) EVENT_DEFER);
+	api->defer_set_destroy (de, defer_event_destroy_cb);
+	api->defer_enable (api->defer_new (api,
+		defer_event_cb, (void *) EVENT_DEFER), false);
+
+	alarm (1);
+	while (g_events != EVENT_ALL)
+		poller_run (&poller);
+
+	poller_pa_destroy (api);
+	soft_assert (g_destroys == EVENT_ALL);
+	poller_free (&poller);
+}
+
+// --- Main --------------------------------------------------------------------
+
+int
+main (int argc, char *argv[])
+{
+	struct test test;
+	test_init (&test, argc, argv);
+	test_add_simple (&test, "/pulse", NULL, test_pulse);
+	return test_run (&test);
+}

tests/xdg.c

@@ -0,0 +1,67 @@
+/*
+ * tests/xdg.c
+ *
+ * Copyright (c) 2024, Přemysl Eric Janouch <p@janouch.name>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
+ * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
+ * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
+ * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ */
+
+#define PROGRAM_NAME "test"
+#define PROGRAM_VERSION "0"
+
+#include "../liberty.c"
+#include "../liberty-xdg.c"
+
+static const char file[] =
+	"# This only tests the happy paths\n"
+	"[Desktop Entry]\n"
+	"Version = 1.0\n"
+	"Name=\\s\\n\\t\\r\\\\\n"
+	"Name[fr]=Nom\n"
+	"Hidden=true\n"
+	"Categories=Utility;TextEditor;\n"
+	"Number=42";
+
+static void
+test_desktop_file (void)
+{
+	struct desktop_file entry = desktop_file_make (file, sizeof file - 1);
+	const char *group = "Desktop Entry";
+
+	char *value = desktop_file_get_string (&entry, group, "Version");
+	hard_assert (!strcmp (value, "1.0"));
+	cstr_set (&value, desktop_file_get_string (&entry, group, "Name"));
+	hard_assert (!strcmp (value, " \n\t\r\\"));
+	free (value);
+
+	hard_assert (desktop_file_get_bool (&entry, group, "Hidden"));
+	struct strv values = desktop_file_get_stringv (&entry, group, "Categories");
+	hard_assert (values.len == 2);
+	hard_assert (!strcmp (values.vector[0], "Utility"));
+	hard_assert (!strcmp (values.vector[1], "TextEditor"));
+	strv_free (&values);
+	hard_assert (desktop_file_get_integer (&entry, group, "Number") == 42);
+
+	desktop_file_free (&entry);
+}
+
+int
+main (int argc, char *argv[])
+{
+	struct test test;
+	test_init (&test, argc, argv);
+
+	test_add_simple (&test, "/desktop-file", NULL, test_desktop_file);
+
+	return test_run (&test);
+}

tools/asciiman.awk

@@ -0,0 +1,336 @@
+# asciiman.awk: simplified AsciiDoc to manual page converter
+#
+# Copyright (c) 2022 - 2024, Přemysl Eric Janouch <p@janouch.name>
+# SPDX-License-Identifier: 0BSD
+#
+# This is not intended to produce great output, merely useful output.
+# As such, input documents should restrict themselves as follows:
+#
+#  - In-line formatting sequences must not overlap,
+#    cannot be escaped, and cannot span lines.
+#  - Heading underlines must match in byte length exactly.
+#  - Only a small subset of syntax is supported overall.
+#
+# Also beware that the output has only been tested with GNU troff and mandoc.
+# Attributes can be passed via environment variables starting with "asciidoc-".
+
+function fatal(message) {
+	print ".\\\" " FILENAME ":" FNR ": fatal error: " message
+	print FILENAME ":" FNR ": fatal error: " message > "/dev/stderr"
+	exit 1
+}
+
+BEGIN {
+	for (name in ENVIRON)
+		if (match(name, /^asciidoc-/))
+			Attrs[substr(name, RSTART + RLENGTH)] = ENVIRON[name]
+}
+
+function expand(s,   attrname, v) {
+	while (match(s, /[{][^{}]+[}]/)) {
+		attrname = substr(s, RSTART + 1, RLENGTH - 2)
+		if (attrname in Attrs)
+			v = v substr(s, 1, RSTART - 1) Attrs[attrname]
+		else
+			v = v substr(s, 1, RSTART + RLENGTH - 1)
+		s = substr(s, RSTART + RLENGTH)
+	}
+	return v s
+}
+
+function escape(s) {
+	gsub(/\\/, "\\\\", s)
+	gsub(/-/, "\\-", s)
+	sub(/^[.']/, "\\\\\\&&", s)
+	return s
+}
+
+function readattribute(line,    attrname) {
+	if (match(line, /^:[^:]+:$/)) {
+		Attrs[substr(line, RSTART + 1, RLENGTH - 2)] = ""
+	} else if (match(line, /^:[^:]+!:$/)) {
+		delete Attrs[substr(line, RSTART + 1, RLENGTH - 3)]
+	} else if (match(line, /^:![^:]+:$/)) {
+		delete Attrs[substr(line, RSTART + 2, RLENGTH - 3)]
+	} else if (match(line, /^:[^:]+: /)) {
+		attrname = substr(line, RSTART + 1, RLENGTH - 3)
+		Attrs[attrname] = expand(substr(line, RSTART + RLENGTH))
+	} else {
+		return 0
+	}
+	return 1
+}
+
+NR == 1 {
+	nameline = $0
+	if (match(nameline, /[(][[:digit:]][)]$/)) {
+		name = substr(nameline, 1, RSTART - 1)
+		section = substr(nameline, RSTART + 1, RLENGTH - 2)
+	} else {
+		fatal("invalid header line")
+	}
+
+	getline
+	if (length(nameline) != length($0) || /[^=]/)
+		fatal("invalid header underline")
+
+	getline
+	while (readattribute($0))
+		getline
+	if ($0)
+		fatal("expected an empty line after the header")
+
+	# Requesting tbl(1), even though we currently do not support tables.
+	print "'\\\" t"
+	printf ".TH \"%s\" \"%s\" \"\" \"%s\"",
+		toupper(name), section, Attrs["mansource"]
+	if ("manmanual" in Attrs)
+		printf " \"%s\"", Attrs["manmanual"]
+	print ""
+
+	# Hyphenation is indeed rather annoying, in particular with long links.
+	print ".nh"
+}
+
+function readattrlist(line, posattrs, namedattrs,    name, value, n) {
+	if (!match(line, /^\[.*\]$/))
+		return 0
+
+	line = expand(substr(line, RSTART + 1, RLENGTH - 2))
+	while (line) {
+		name = ""
+		if (match(line, /^[[:alnum:]][[:alnum:]-]*/)) {
+			value = substr(line, RSTART, RLENGTH)
+			if (match(substr(line, RSTART + RLENGTH),
+					/^[[:space:]]*=[[:space:]]*/)) {
+				name = value
+				line = substr(line, 1 + length(name) + RLENGTH)
+			}
+		}
+
+		# The quoting syntax actually is awful like this.
+		if (match(line, /^"(\\.|[^"\\])*"/)) {
+			value = substr(line, RSTART + 1, RLENGTH - 2)
+			gsub(/\\"/, "\"", value)
+		} else if (match(line, /^'(\\.|[^'\\])*'/)) {
+			value = substr(line, RSTART + 1, RLENGTH - 2)
+			gsub(/\\'/, "'", value)
+		} else {
+			match(line, /^[^,]*/)
+			value = substr(line, RSTART, RLENGTH)
+			sub(/[[:space:]]*$/, "", value)
+		}
+
+		line = substr(line, RSTART + RLENGTH)
+		sub(/^[[:space:]]*,[[:space:]]*/, "", line)
+		if (!name)
+			posattrs[++n] = value
+		else if (value == "None")
+			delete namedattrs[name]
+		else
+			namedattrs[name] = value
+	}
+	return 1
+}
+
+function format(line,    v) {
+	# Pass-through, otherwise useful for hacks, is a bit of a lie here,
+	# and formatting doesn't fully respect word boundaries.
+	while (line) {
+		if (match(line, /^[+][+][+][^+]+[+][+][+]/)) {
+			v = v substr(line, RSTART + 3, RLENGTH - 6)
+		} else if (match(line, /^__[^_]+__/)) {
+			v = v "\\fI" substr(line, RSTART + 2, RLENGTH - 4) "\\fP"
+		} else if (match(line, /^[*][*][^*]+[*][*]/)) {
+			v = v "\\fB" substr(line, RSTART + 2, RLENGTH - 4) "\\fP"
+		} else if (match(line, /^_[^_]+_/) &&
+			substr(line, RSTART + RLENGTH) !~ /^[[:alnum:]]/) {
+			v = v "\\fI" substr(line, RSTART + 1, RLENGTH - 2) "\\fP"
+		} else if (match(line, /^[*][^*]+[*]/) &&
+			substr(line, RSTART + RLENGTH) !~ /^[[:alnum:]]/) {
+			v = v "\\fB" substr(line, RSTART + 1, RLENGTH - 2) "\\fP"
+		} else if (match(line, /^`[^`]+`/) &&
+			substr(line, RSTART + RLENGTH) !~ /^[[:alnum:]]/) {
+			# Manual pages are usually already rendered in monospace;
+			# follow others, and render this in boldface.
+			v = v "\\fB" substr(line, RSTART + 1, RLENGTH - 2) "\\fP"
+		} else {
+			v = v substr(line, 1, 1)
+			line = substr(line, 2)
+			continue
+		}
+		line = substr(line, RSTART + RLENGTH)
+	}
+	return v
+}
+
+function flushspace() {
+	if (NeedSpace) {
+		print ".sp"
+		NeedSpace = 0
+	}
+}
+
+function inline(line) {
+	if (!line) {
+		NeedSpace = 1
+		return
+	}
+
+	flushspace()
+	line = format(escape(expand(line)))
+
+	# Strip empty URL descriptions, otherwise useful for demarking the end.
+	while (match(line, /[^[:space:]]+\[\]/)) {
+		line = substr(line, 1, RSTART + RLENGTH - 3) \
+			 substr(line, RSTART + RLENGTH)
+	}
+
+	# Enable double-spacing after the end of a sentence.
+	gsub(/[.][[:space:]]+/, ".\n", line)
+	gsub(/[!][[:space:]]+/, "!\n", line)
+	gsub(/[?][[:space:]]+/, "?\n", line)
+
+	# Quote commands resulting from that, as well as from expand().
+	gsub(/\n[.]/, "\n\\\\\\&.", line)
+	gsub(/\n[']/, "\n\\\\\\&'", line)
+
+	sub(/[[:space:]]+[+]$/, "\n.br", line)
+	print line
+}
+
+# Returns 1 iff the left-over $0 should be processed further.
+function process(firstline,     posattrs, namedattrs) {
+	if (readattribute(firstline))
+		return 0
+	if (getline <= 0) {
+		inline(firstline)
+		return 0
+	}
+
+	# Block attribute list lines.
+	delete posattrs[0]
+	delete namedattrs[0]
+	while (readattrlist(firstline, posattrs, namedattrs)) {
+		firstline = $0
+		if (getline <= 0) {
+			inline(firstline)
+			return 0
+		}
+	}
+
+	# mandoc(1) automatically precedes section headers with blank lines.
+	if (length(firstline) == length($0) && /^-+$/) {
+		print ".SH \"" escape(toupper(expand(firstline))) "\""
+		NeedSpace = 0
+		return 0
+	}
+	if (length(firstline) == length($0) && /^~+$/) {
+		print ".SS \"" escape(expand(firstline)) "\""
+		NeedSpace = 0
+		return 0
+	}
+
+	if (firstline ~ /^--$/) {
+		flushspace()
+
+		# For now, recognize, but do not process open block delimiters.
+		InOpenBlock = !InOpenBlock
+		return 1
+	}
+	if (firstline ~ /^(-{4,}|[.]{4,})$/) {
+		flushspace()
+
+		print ".if n .RS 4"
+		print ".nf"
+		print ".fam C"
+		do {
+			print escape($0)
+		} while (getline > 0 && $0 != firstline)
+		print ".fam"
+		print ".fi"
+		print ".if n .RE"
+		return 0
+	}
+	if (firstline ~ /^\/{4,}$/) {
+		do {
+			print ".\\\" " $0
+		} while (getline > 0 && $0 != firstline)
+		return 0
+	}
+	if (match(firstline, /^\/\//)) {
+		print ".\\\"" substr(firstline, RSTART + RLENGTH)
+		return 1
+	}
+
+	# We generally assume these blocks end with a blank line.
+	if (match(firstline, /^[[:space:]]*[*][[:space:]]+/)) {
+		flushspace()
+
+		# Bullet magic copied over from AsciiDoc/Asciidoctor generators.
+		print ".RS 4"
+		print ".ie n \\{\\"
+		print "\\h'-04'\\(bu\\h'+03'\\c"
+		print ".\\}"
+		print ".el \\{\\"
+		print ".sp -1"
+		print ".IP \\(bu 2.3"
+		print ".\\}"
+
+		inline(substr(firstline, RSTART + RLENGTH))
+		while ($0) {
+			sub(/^[[:space:]]+/, "")
+			sub(/^[+]$/, "")
+			if (!process($0) && getline <= 0)
+				fatal("unexpected EOF")
+			if (match($0, /^[[:space:]]*[*][[:space:]]+/))
+				break
+		}
+		print ".RE"
+		NeedSpace = 1
+		return !!$0
+	}
+	if (match(firstline, /^[[:space:]]+/)) {
+		flushspace()
+
+		print ".if n .RS 4"
+		print ".nf"
+		print ".fam C"
+		do {
+			print escape(substr(firstline, RLENGTH + 1))
+			firstline = $0
+		} while ($0 && getline > 0)
+		print ".fam"
+		print ".fi"
+		print ".if n .RE"
+		return 1
+	}
+	if (match(firstline, /::$/)) {
+		inline(substr(firstline, 1, RSTART - 1))
+		while (match($0, /::$/)) {
+			print ".br"
+			inline(substr($0, 1, RSTART - 1))
+			if (getline <= 0)
+				fatal("unexpected EOF")
+		}
+
+		print ".RS 4"
+		while ($0) {
+			sub(/^[[:space:]]+/, "")
+			sub(/^[+]$/, "")
+			if (!process($0) && getline <= 0)
+				fatal("unexpected EOF")
+			if (match($0, /::$/))
+				break
+		}
+		print ".RE"
+		NeedSpace = 1
+		return !!$0
+	}
+	inline(firstline)
+	return 1
+}
+
+{
+	while (process($0)) {}
+}

tools/cmake-dump.awk

@@ -0,0 +1,24 @@
+# cmake-dump.awk: dump parsed CMake scripts as tables
+#
+# Copyright (c) 2022, Přemysl Eric Janouch <p@janouch.name>
+# SPDX-License-Identifier: 0BSD
+#
+# Parsed scripts are output in a table, with commands separated using ASCII
+# Record Separators, and arguments using Unit Separators.
+#
+# Example usage: awk -f cmake-parser.awk -f cmake-dump.awk CMakeLists.txt \
+#  | sed 'y/\x1F\x1E\t\n/\t\n  /' \
+#  | sed -n '/^project\t\([^\t]*\).*\tVERSION\t\([^\t]*\).*/{s//\1 \2/p;q;}'
+
+function sanitize(s) {
+	if (s ~ /[\x1E\x1F]/)
+		fatal("conflicting ASCII control characters found in source")
+	return s
+}
+
+Command {
+	out = sanitize(Command)
+	for (i in Args)
+		out = out "\x1F" sanitize(Args[i])
+	printf "%s\x1E", out
+}

tools/cmake-parser.awk

@@ -0,0 +1,252 @@
+# cmake-parser.awk: rudimentary CMake script parser
+#
+# Copyright (c) 2022, Přemysl Eric Janouch <p@janouch.name>
+# SPDX-License-Identifier: 0BSD
+#
+# Implemented roughly according to the grammar described in cmake-language(7),
+# which is self-conflicting, and not an accurate description.
+#
+# The result of parsing is stored in the case-normalized Command variable,
+# and the Args array.  These can be used by subsequent scripts.
+
+function warning(message) {
+	print FILENAME ":" FNR ": warning: " message > "/dev/stderr"
+}
+
+function fatal(message) {
+	print FILENAME ":" FNR ": fatal error: " message > "/dev/stderr"
+	exit 1
+}
+
+function expect(v) {
+	if (!v && v == 0)
+		fatal("broken expectations at `" $0 "'")
+	return v
+}
+
+function literal(v) {
+	if (substr($0, 1, length(v)) != v)
+		return 0
+	$0 = substr($0, length(v) + 1)
+	return 1
+}
+
+function regexp(re) {
+	if (!match($0, "^" re))
+		return 0
+	$0 = substr($0, RLENGTH + 1)
+	return 1
+}
+
+function space() {
+	return regexp("[ \t]+")
+}
+
+function unbracket(len,    v) {
+	do {
+		if (match($0, "]={" len "}]")) {
+			v = v substr($0, 1, RSTART - 1)
+			$0 = substr($0, RSTART + RLENGTH)
+			return v
+		}
+		v = v $0 RS
+	} while (getline > 0)
+	fatal("unterminated bracket")
+}
+
+function bracket_comment() {
+	if (!match($0, /^#\[=*\[/))
+		return 0
+	$0 = substr($0, RSTART + RLENGTH)
+	unbracket(RLENGTH - 3)
+	return 1
+}
+
+function line_ending() {
+	while (space() || bracket_comment()) {}
+	if (/^#/)
+		$0 = ""
+	return !$0
+}
+
+# ------------------------------------------------------------------------------
+
+# While elementary expansion of previously set variables is implementable,
+# it doesn't seem to be worth the effort.
+function expand(s,    v) {
+	v = s
+	while (match(v, /\\*[$](|ENV|CACHE)[{]/)) {
+		if (index(substr(v, RSTART), "$") % 2 != 0) {
+			warning("variable expansion is not supported: " s)
+			return s
+		}
+		v = substr(v, RSTART + RLENGTH)
+	}
+	return s
+}
+
+function escape_sequence(    v) {
+	if (!literal("\\"))
+		return 0
+
+	if (literal("t")) return "\t"
+	if (literal("r")) return "\r"
+	if (literal("n")) return "\n"
+
+	# escape_semicolon isn't treated any specially here.
+	if (regexp("[A-Za-z0-9]"))
+		fatal("unsupported escape sequence")
+
+	if ($0) {
+		v = substr($0, 1, 1)
+		$0 = substr($0, 2)
+		return v
+	}
+	if (getline > 0)
+		return ""
+	fatal("premature end of file")
+}
+
+function quoted_argument(    v, unescaped) {
+	if (!literal("\""))
+		return 0
+
+	v = ""
+	while (!literal("\"")) {
+		if (!$0) {
+			if (getline <= 0)
+				fatal("premature end of file")
+			v = v RS
+		} else if ((unescaped = escape_sequence())) {
+			if (unescaped == "\\" || unescaped == "$")
+				v = v "\\"
+			else if (unescaped == ";")
+				v = v "\\\\"
+			v = v unescaped
+		} else if (unescaped == "") {
+			# quoted_continuation
+		} else {
+			v = v substr($0, 1, 1)
+			$0 = substr($0, 2)
+		}
+	}
+	return v
+}
+
+function finalize_quoted(expanded,    v) {
+	while (match(expanded, /\\./)) {
+		v = v substr(expanded, 1, RSTART - 1) \
+			substr(expanded, RSTART + 1, 1)
+		expanded = substr(expanded, RSTART + RLENGTH)
+	}
+	Args[++N] = v expanded
+}
+
+function unquoted_argument(    v, unescaped) {
+	while (1) {
+		if (match($0, /^[^[:space:]()#"\\]+/)) {
+			v = v substr($0, RSTART, RLENGTH)
+			$0 = substr($0, RSTART + RLENGTH)
+		} else if ((unescaped = escape_sequence())) {
+			if (unescaped == "\\" || unescaped == "$" || unescaped == ";")
+				v = v "\\"
+			v = v unescaped
+		} else if (unescaped == "") {
+			fatal("unexpected backslash in an unquoted argument")
+		} else {
+			# unquoted_legacy is not supported.
+			return v
+		}
+	}
+}
+
+function finalize_unquoted(expanded,    v) {
+	while (expanded) {
+		if (expanded ~ /^;/) {
+			if (v)
+				Args[++N] = v
+			v = ""
+			expanded = substr(expanded, 2)
+		} else if (expanded ~ /^\\./) {
+			v = v substr(expanded, 2, 1)
+			expanded = substr(expanded, 3)
+		} else {
+			v = v substr(expanded, 1, 1)
+			expanded = substr(expanded, 2)
+		}
+	}
+	if (v)
+		Args[++N] = v
+}
+
+# We keep and reprocess some escape sequences in here.
+function argument(    arg, expanded, v) {
+	if (regexp("\\[=*\\["))
+		Args[++N] = unbracket(RLENGTH - 2)
+	else if ((arg = quoted_argument()) || arg == "")
+		finalize_quoted(expand(arg))
+	else if ((arg = unquoted_argument()))
+		finalize_unquoted(expand(arg))
+	else
+		return 0
+	return 1
+}
+
+# ------------------------------------------------------------------------------
+
+function identifier(    v) {
+	if (!match($0, /^[A-Za-z_][A-Za-z0-9_]*/))
+		return 0
+	v = substr($0, 1, RLENGTH)
+	$0 = substr($0, RLENGTH + 1)
+	return v
+}
+
+function separation() {
+	if (space() || bracket_comment())
+		return 1
+
+	if (!line_ending())
+		return 0
+	if (getline > 0)
+		return 1
+	fatal("premature end of file")
+}
+
+function command_invocation(    level) {
+	while (space()) {}
+	Command = identifier()
+	if (!Command)
+		return 0
+	while (space()) {}
+
+	Command = tolower(Command)
+	for (N in Args)
+		delete Args[N]
+
+	N = 0
+	expect(literal("("))
+	while (1) {
+		while (separation()) {}
+		if (literal(")")) {
+			if (!level--)
+				break
+			Args[++N] = ")"
+			continue
+		}
+		if (literal("(")) {
+			level++
+			Args[++N] = "("
+			continue
+		}
+		expect(argument())
+		if (!/^[()]/)
+			expect(separation())
+	}
+	return 1
+}
+
+{
+	command_invocation()
+	expect(line_ending())
+}

tools/lxdrgen-c.awk

@@ -0,0 +1,324 @@
+# lxdrgen-c.awk: C backend for lxdrgen.awk.
+#
+# Copyright (c) 2022, Přemysl Eric Janouch <p@janouch.name>
+# SPDX-License-Identifier: 0BSD
+#
+# Neither *_new() nor *_destroy() functions are provided, because they'd only
+# be useful for top-levels, and are merely extra malloc()/free() calls.
+# Users are expected to reuse buffers.
+#
+# Similarly, no constructors are produced--those are easy to write manually.
+#
+# All arrays are deserialized zero-terminated, so u8<> and i8<> can be directly
+# used as C strings.
+#
+# All types must be able to dispose partially zero values going from the back,
+# i.e., in the reverse order of deserialization.
+
+function define_internal(name, ctype) {
+	Types[name] = "internal"
+	CodegenCType[name] = ctype
+}
+
+function define_int(shortname, ctype) {
+	define_internal(shortname, ctype)
+	CodegenSerialize[shortname] = \
+		"\tstr_pack_" shortname "(w, %s);\n"
+	CodegenDeserialize[shortname] = \
+		"\tif (!msg_unpacker_" shortname "(r, &%s))\n" \
+		"\t\treturn false;\n"
+}
+
+function define_sint(size) { define_int("i" size, "int" size "_t") }
+function define_uint(size) { define_int("u" size, "uint" size "_t") }
+
+function codegen_begin() {
+	define_sint("8")
+	define_sint("16")
+	define_sint("32")
+	define_sint("64")
+	define_uint("8")
+	define_uint("16")
+	define_uint("32")
+	define_uint("64")
+
+	define_internal("string", "struct str")
+	CodegenDispose["string"] = "\tstr_free(&%s);\n"
+	CodegenSerialize["string"] = \
+		"\tif (!proto_string_serialize(&%s, w))\n" \
+		"\t\treturn false;\n"
+	CodegenDeserialize["string"] = \
+		"\tif (!proto_string_deserialize(&%s, r))\n" \
+		"\t\treturn false;\n"
+
+	define_internal("bool", "bool")
+	CodegenSerialize["bool"] = \
+		"\tstr_pack_u8(w, !!%s);\n"
+	CodegenDeserialize["bool"] = \
+		"\t{\n" \
+		"\t\tuint8_t v = 0;\n" \
+		"\t\tif (!msg_unpacker_u8(r, &v))\n" \
+		"\t\t\treturn false;\n" \
+		"\t\t%s = !!v;\n" \
+		"\t}\n"
+
+	print "// Code generated from " FILENAME ". DO NOT EDIT."
+	print "// This file directly depends on liberty.c, but doesn't include it."
+	print ""
+	print "static bool"
+	print "proto_string_serialize(const struct str *s, struct str *w) {"
+	print "\tif (s->len > UINT32_MAX)"
+	print "\t\treturn false;"
+	print "\tstr_pack_u32(w, s->len);"
+	print "\tstr_append_str(w, s);"
+	print "\treturn true;"
+	print "}"
+	print ""
+	print "static bool"
+	print "proto_string_deserialize(struct str *s, struct msg_unpacker *r) {"
+	print "\tuint32_t len = 0;"
+	print "\tif (!msg_unpacker_u32(r, &len))"
+	print "\t\treturn false;"
+	print "\tif (msg_unpacker_get_available(r) < len)"
+	print "\t\treturn false;"
+	print "\t*s = str_make();"
+	print "\tstr_append_data(s, r->data + r->offset, len);"
+	print "\tr->offset += len;"
+	print "\tif (!utf8_validate (s->str, s->len))"
+	print "\t\treturn false;"
+	print "\treturn true;"
+	print "}"
+}
+
+function codegen_constant(name, value) {
+	print ""
+	print "enum { " PrefixUpper name " = " value " };"
+}
+
+function codegen_enum_value(name, subname, value, cg) {
+	append(cg, "fields",
+		"\t" PrefixUpper toupper(cameltosnake(name)) "_" subname \
+		" = " value ",\n")
+}
+
+function codegen_enum(name, cg,    ctype) {
+	ctype = "enum " PrefixLower cameltosnake(name)
+	print ""
+	print ctype " {"
+	print cg["fields"] "};"
+
+	# XXX: This should also check if it isn't out-of-range for any reason,
+	# but our usage of sprintf() stands in the way a bit.
+	CodegenSerialize[name] = "\tstr_pack_i8(w, %s);\n"
+	CodegenDeserialize[name] = \
+		"\t{\n" \
+		"\t\tint8_t v = 0;\n" \
+		"\t\tif (!msg_unpacker_i8(r, &v) || !v)\n" \
+		"\t\t\treturn false;\n" \
+		"\t\t%s = v;\n" \
+		"\t}\n"
+
+	CodegenCType[name] = ctype
+	for (i in cg)
+		delete cg[i]
+}
+
+function codegen_struct_tag(d, cg,    f) {
+	f = "self->" d["name"]
+	append(cg, "fields", "\t" CodegenCType[d["type"]] " " d["name"] ";\n")
+	append(cg, "dispose", sprintf(CodegenDispose[d["type"]], f))
+	append(cg, "serialize", sprintf(CodegenSerialize[d["type"]], f))
+	# Do not deserialize here, that would be out of order.
+}
+
+function codegen_struct_field(d, cg,    f, dispose, serialize, deserialize) {
+	f = "self->" d["name"]
+	dispose = CodegenDispose[d["type"]]
+	serialize = CodegenSerialize[d["type"]]
+	deserialize = CodegenDeserialize[d["type"]]
+	if (!d["isarray"]) {
+		append(cg, "fields", "\t" CodegenCType[d["type"]] " " d["name"] ";\n")
+		append(cg, "dispose", sprintf(dispose, f))
+		append(cg, "serialize", sprintf(serialize, f))
+		append(cg, "deserialize", sprintf(deserialize, f))
+		return
+	}
+
+	append(cg, "fields",
+		"\t" CodegenCType["u32"] " " d["name"] "_len;\n" \
+		"\t" CodegenCType[d["type"]] " *" d["name"] ";\n")
+
+	if (dispose)
+		append(cg, "dispose", "\tif (" f ")\n" \
+			"\t\tfor (size_t i = 0; i < " f "_len; i++)\n" \
+			indent(indent(sprintf(dispose, f "[i]"))))
+	append(cg, "dispose", "\tfree(" f ");\n")
+
+	append(cg, "serialize", sprintf(CodegenSerialize["u32"], f "_len"))
+	if (d["type"] == "u8" || d["type"] == "i8") {
+		append(cg, "serialize",
+			"\tstr_append_data(w, " f ", " f "_len);\n")
+	} else if (serialize) {
+		append(cg, "serialize",
+			"\tfor (size_t i = 0; i < " f "_len; i++)\n" \
+			indent(sprintf(serialize, f "[i]")))
+	}
+
+	append(cg, "deserialize", sprintf(CodegenDeserialize["u32"], f "_len") \
+		"\tif (!(" f " = calloc(" f "_len + 1, sizeof *" f ")))\n" \
+		"\t\treturn false;\n")
+	if (d["type"] == "u8" || d["type"] == "i8") {
+		append(cg, "deserialize",
+			"\tif (msg_unpacker_get_available(r) < " f "_len)\n" \
+			"\t\treturn false;\n" \
+			"\tmemcpy(" f ", r->data + r->offset, " f "_len);\n" \
+			"\tr->offset += " f "_len;\n")
+	} else if (deserialize) {
+		append(cg, "deserialize",
+			"\tfor (size_t i = 0; i < " f "_len; i++)\n" \
+			indent(sprintf(deserialize, f "[i]")))
+	}
+}
+
+function codegen_struct(name, cg,    ctype, funcname) {
+	ctype = "struct " PrefixLower cameltosnake(name)
+	print ""
+	print ctype " {"
+	print cg["fields"] "};"
+
+	if (cg["dispose"]) {
+		funcname = PrefixLower cameltosnake(name) "_free"
+		print ""
+		print "static void\n" funcname "(" ctype " *self) {"
+		print cg["dispose"] "}"
+
+		CodegenDispose[name] = "\t" funcname "(&%s);\n"
+	}
+	if (cg["serialize"]) {
+		funcname = PrefixLower cameltosnake(name) "_serialize"
+		print ""
+		print "static bool\n" \
+			  funcname "(\n\t\tconst " ctype " *self, struct str *w) {"
+		print cg["serialize"] "\treturn true;"
+		print "}"
+
+		CodegenSerialize[name] = "\tif (!" funcname "(&%s, w))\n" \
+			"\t\treturn false;\n"
+	}
+	if (cg["deserialize"]) {
+		funcname = PrefixLower cameltosnake(name) "_deserialize"
+		print ""
+		print "static bool\n" \
+			  funcname "(\n\t\t" ctype " *self, struct msg_unpacker *r) {"
+		print cg["deserialize"] "\treturn true;"
+		print "}"
+
+		CodegenDeserialize[name] = "\tif (!" funcname "(&%s, r))\n" \
+			"\t\treturn false;\n"
+	}
+
+	CodegenCType[name] = ctype
+	for (i in cg)
+		delete cg[i]
+}
+
+function codegen_union_tag(name, d, cg) {
+	cg["tagtype"] = d["type"]
+	cg["tagname"] = d["name"]
+	append(cg, "fields", "\t" CodegenCType[d["type"]] " " d["name"] ";\n")
+}
+
+function codegen_union_struct( \
+		name, casename, cg, scg,     structname, fieldname, fullcasename) {
+	# Don't generate obviously useless structs.
+	fullcasename = toupper(cameltosnake(cg["tagtype"])) "_" casename
+	if (!scg["dispose"] && !scg["deserialize"]) {
+		append(cg, "structless", "\tcase " PrefixUpper fullcasename ":\n")
+		for (i in scg)
+			delete scg[i]
+		return
+	}
+
+	# And thus not all generated structs are present in Types.
+	structname = name "_" casename
+	fieldname = tolower(casename)
+	codegen_struct(structname, scg)
+
+	append(cg, "fields", "\t" CodegenCType[structname] " " fieldname ";\n")
+	if (CodegenDispose[structname])
+		append(cg, "dispose", "\tcase " PrefixUpper fullcasename ":\n" \
+			indent(sprintf(CodegenDispose[structname], "self->" fieldname)) \
+			"\t\tbreak;\n")
+
+	# With no de/serialization code, this will simply recognize the tag.
+	append(cg, "serialize", "\tcase " PrefixUpper fullcasename ":\n" \
+		indent(sprintf(CodegenSerialize[structname], "self->" fieldname)) \
+		"\t\tbreak;\n")
+	append(cg, "deserialize", "\tcase " PrefixUpper fullcasename ":\n" \
+		indent(sprintf(CodegenDeserialize[structname], "self->" fieldname)) \
+		"\t\tbreak;\n")
+}
+
+function codegen_union(name, cg, exhaustive,    f, ctype, funcname) {
+	ctype = "union " PrefixLower cameltosnake(name)
+	print ""
+	print ctype " {"
+	print cg["fields"] "};"
+
+	f = "self->" cg["tagname"]
+	if (cg["dispose"]) {
+		funcname = PrefixLower cameltosnake(name) "_free"
+		print ""
+		print "static void\n" funcname "(" ctype " *self) {"
+		print "\tswitch (" f ") {"
+		if (cg["structless"])
+			print cg["structless"] \
+				indent(sprintf(CodegenDispose[cg["tagtype"]], f)) "\t\tbreak;"
+		print cg["dispose"] "\tdefault:"
+		print "\t\tbreak;"
+		print "\t}"
+		print "}"
+
+		CodegenDispose[name] = "\t" funcname "(&%s);\n"
+	}
+	{
+		funcname = PrefixLower cameltosnake(name) "_serialize"
+		print ""
+		print "static bool\n" \
+			  funcname "(\n\t\tconst " ctype " *self, struct str *w) {"
+		print "\tswitch (" f ") {"
+		if (cg["structless"])
+			print cg["structless"] \
+				indent(sprintf(CodegenSerialize[cg["tagtype"]], f)) "\t\tbreak;"
+		print cg["serialize"] "\tdefault:"
+		print "\t\treturn false;"
+		print "\t}"
+		print "\treturn true;"
+		print "}"
+
+		CodegenSerialize[name] = "\tif (!" funcname "(&%s, w))\n" \
+			"\t\treturn false;\n"
+	}
+	{
+		funcname = PrefixLower cameltosnake(name) "_deserialize"
+		print ""
+		print "static bool\n" \
+			  funcname "(\n\t\t" ctype " *self, struct msg_unpacker *r) {"
+		print sprintf(CodegenDeserialize[cg["tagtype"]], f)
+		print "\tswitch (" f ") {"
+		if (cg["structless"])
+			print cg["structless"] "\t\tbreak;"
+		print cg["deserialize"] "\tdefault:"
+		print "\t\treturn false;"
+		print "\t}"
+		print "\treturn true;"
+		print "}"
+
+		CodegenDeserialize[name] = "\tif (!" funcname "(&%s, r))\n" \
+			"\t\treturn false;\n"
+	}
+
+	CodegenCType[name] = ctype
+	for (i in cg)
+		delete cg[i]
+}

tools/lxdrgen-cpp-posix.cpp

@@ -0,0 +1,67 @@
+// lxdrgen-cpp-posix.cpp: POSIX support code for lxdrgen-cpp.awk.
+//
+// Copyright (c) 2023, Přemysl Eric Janouch <p@janouch.name>
+// SPDX-License-Identifier: 0BSD
+#include <iconv.h>
+
+#include <cstdint>
+#include <string>
+
+// Various BSD derivatives may have a problem here.
+// Linux defines __STDC_ISO_10646__, but also supports "WCHAR_T".
+#ifdef APPLE
+#define ICONV_WCHAR "UTF-32"
+#else
+#define ICONV_WCHAR "WCHAR_T"
+#endif
+
+namespace LibertyXDR {
+
+bool utf8_to_wstring(const uint8_t *utf8, size_t length, std::wstring &wide) {
+	iconv_t conv = iconv_open(ICONV_WCHAR, "UTF-8");
+	if (conv == (iconv_t) -1)
+		return false;
+
+	wchar_t buffer[1024] = {};
+	char *start = (char *) buffer, *out = start, *in = (char *) utf8;
+	size_t available = sizeof buffer;
+	wide.clear();
+	while (iconv(conv, &in, &length, &out, &available) == (size_t) -1) {
+		if (errno != E2BIG) {
+			iconv_close(conv);
+			return false;
+		}
+
+		wide.append(buffer, (out - start) / sizeof *buffer);
+		out = start;
+		available = sizeof buffer;
+	}
+	wide.append(buffer, (out - start) / sizeof *buffer);
+	iconv_close(conv);
+	return true;
+}
+
+bool wstring_to_utf8(const std::wstring &wide, std::string &utf8) {
+	iconv_t conv = iconv_open("UTF-8", ICONV_WCHAR);
+	if (conv == (iconv_t) -1)
+		return false;
+
+	char buffer[1024] = {}, *out = buffer, *in = (char *) wide.data();
+	size_t available = sizeof buffer, length = wide.size() * sizeof wide[0];
+	utf8.clear();
+	while (iconv(conv, &in, &length, &out, &available) == (size_t) -1) {
+		if (errno != E2BIG) {
+			iconv_close(conv);
+			return false;
+		}
+
+		utf8.append(buffer, out - buffer);
+		out = buffer;
+		available = sizeof buffer;
+	}
+	utf8.append(buffer, out - buffer);
+	iconv_close(conv);
+	return true;
+}
+
+} // namespace LibertyXDR

tools/lxdrgen-cpp-win32.cpp

@@ -0,0 +1,47 @@
+// lxdrgen-cpp-win32.cpp: Win32 support code for lxdrgen-cpp.awk.
+//
+// Copyright (c) 2023, Přemysl Eric Janouch <p@janouch.name>
+// SPDX-License-Identifier: 0BSD
+#include <windows.h>
+
+#include <climits>
+#include <cstdint>
+#include <string>
+
+namespace LibertyXDR {
+
+bool utf8_to_wstring(const uint8_t *utf8, size_t length, std::wstring &wide) {
+	wide.clear();
+	if (!length)
+		return true;
+	if (length > INT_MAX)
+		return false;
+
+	int size = MultiByteToWideChar(CP_UTF8, MB_ERR_INVALID_CHARS,
+		(LPCCH) utf8, length, nullptr, 0);
+	if (size <= 0)
+		return false;
+
+	wide.resize(size);
+	return !!MultiByteToWideChar(CP_UTF8, MB_ERR_INVALID_CHARS,
+		(LPCCH) utf8, length, wide.data(), size);
+}
+
+bool wstring_to_utf8(const std::wstring &wide, std::string &utf8) {
+	utf8.clear();
+	if (wide.empty())
+		return true;
+	if (wide.size() > INT_MAX)
+		return false;
+
+	int size = WideCharToMultiByte(CP_UTF8, WC_ERR_INVALID_CHARS,
+		(LPCWCH) wide.data(), wide.size(), nullptr, 0, NULL, NULL);
+	if (size <= 0)
+		return false;
+
+	utf8.resize(size);
+	return !!WideCharToMultiByte(CP_UTF8, WC_ERR_INVALID_CHARS,
+		(LPCWCH) wide.data(), wide.size(), utf8.data(), size, NULL, NULL);
+}
+
+} // namespace LibertyXDR

tools/lxdrgen-cpp.awk

@@ -0,0 +1,350 @@
+# lxdrgen-cpp.awk: C++ backend for lxdrgen.awk.
+#
+# This backend is intended for Windows, it just happens to have a fallback
+# that will probably work on Unices, of which we make use in tests.
+#
+# Copyright (c) 2023, Přemysl Eric Janouch <p@janouch.name>
+# SPDX-License-Identifier: 0BSD
+
+function define_internal(name, ctype) {
+	Types[name] = "internal"
+	CodegenCType[name] = ctype
+	CodegenSerialize[name] = \
+		"\tw.append(%s);\n"
+	CodegenDeserialize[name] = \
+		"\tif (!r.read(%s))\n" \
+		"\t\treturn false;\n"
+}
+
+function define_int(shortname, ctype) {
+	define_internal(shortname, ctype)
+}
+
+function define_sint(size) { define_int("i" size, "int" size "_t") }
+function define_uint(size) { define_int("u" size, "uint" size "_t") }
+
+function codegen_begin() {
+	define_sint("8")
+	define_sint("16")
+	define_sint("32")
+	define_sint("64")
+	define_uint("8")
+	define_uint("16")
+	define_uint("32")
+	define_uint("64")
+
+	define_internal("string", "std::wstring")
+	define_internal("bool", "bool")
+
+	CodegenSerialize["string"] = \
+		"\tif (!w.append(%s))\n" \
+		"\t\treturn false;\n"
+
+	print "// Code generated from " FILENAME ". DO NOT EDIT."
+	print ""
+	print "#include <cstdint>"
+	print "#include <memory>"
+	print "#include <string>"
+	print "#include <vector>"
+	print ""
+	print "namespace LibertyXDR {"
+	print ""
+	print "bool utf8_to_wstring("
+	print "\tconst uint8_t *utf8, size_t length, std::wstring &wide);"
+	print "bool wstring_to_utf8("
+	print "\tconst std::wstring &wide, std::string &utf8);"
+	print ""
+	print "struct Reader {"
+	print "\tconst uint8_t *data = {};"
+	print "\tsize_t length = {};"
+	print ""
+	print "\ttemplate<typename T> bool read(T &number) {"
+	print "\t\tif (length < sizeof number)"
+	print "\t\t\treturn false;"
+	print ""
+	print "\t\tnumber = 0;"
+	print "\t\tfor (size_t i = 0; i < sizeof number; i++) {"
+	print "\t\t\tnumber = number << 8 | *data++;"
+	print "\t\t\tlength--;"
+	print "\t\t}"
+	print "\t\treturn true;"
+	print "\t}"
+	print ""
+	print "\tbool read(bool &boolean) {"
+	print "\t\tuint8_t number = 0;"
+	print "\t\tif (!read(number))"
+	print "\t\t\treturn false;"
+	print ""
+	print "\t\tboolean = number != 0;"
+	print "\t\treturn true;"
+	print "\t}"
+	print ""
+	print "\tbool read(std::wstring &string) {"
+	print "\t\tuint32_t size = 0;"
+	print "\t\tif (!read(size) || size > length)"
+	print "\t\t\treturn false;"
+	print "\t\tif (!utf8_to_wstring(data, size, string))"
+	print "\t\t\treturn false;"
+	print ""
+	print "\t\tdata += size;"
+	print "\t\tlength -= size;"
+	print "\t\treturn true;"
+	print "\t}"
+	print ""
+	print "\tbool read(std::vector<uint8_t> &vector) {"
+	print "\t\tuint32_t size = 0;"
+	print "\t\tif (!read(size) || size > length)"
+	print "\t\t\treturn false;"
+	print "\t\tvector.assign(data, data + size);"
+	print ""
+	print "\t\tdata += size;"
+	print "\t\tlength -= size;"
+	print "\t\treturn true;"
+	print "\t}"
+	print "};"
+	print ""
+	print "struct Writer {"
+	print "\tstd::vector<uint8_t> data;"
+	print ""
+	print "\ttemplate<typename T> bool append(T number) {"
+	print "\t\tuint8_t buffer[sizeof number], *p = buffer + sizeof buffer;"
+	print "\t\twhile (p != buffer) {"
+	print "\t\t\t*--p = number;"
+	print "\t\t\tnumber >>= 8;"
+	print "\t\t}"
+	print "\t\tdata.insert(data.end(), buffer, buffer + sizeof buffer);"
+	print "\t\treturn true;"
+	print "\t}"
+	print ""
+	print "\tbool append(int8_t number) {"
+	print "\t\tdata.push_back(number);"
+	print "\t\treturn true;"
+	print "\t}"
+	print ""
+	print "\tbool append(uint8_t number) {"
+	print "\t\tdata.push_back(number);"
+	print "\t\treturn true;"
+	print "\t}"
+	print ""
+	print "\tbool append(bool boolean) {"
+	print "\t\treturn append(uint8_t(boolean));"
+	print "\t}"
+	print ""
+	print "\tbool append(const std::wstring &string) {"
+	print "\t\tif (string.size() > UINT32_MAX)"
+	print "\t\t\treturn false;"
+	print ""
+	print "\t\tstd::string utf8;"
+	print "\t\tif (!wstring_to_utf8(string, utf8))"
+	print "\t\t\treturn false;"
+	print ""
+	print "\t\tappend<uint32_t>(utf8.size());"
+	print "\t\tdata.insert(data.end(), utf8.begin(), utf8.end());"
+	print "\t\treturn true;"
+	print "\t}"
+	print "};"
+	print ""
+	print "} // namespace LibertyXDR"
+	print "namespace " PrefixCamel " {"
+}
+
+END {
+	print ""
+	print "} // namespace " PrefixCamel
+}
+
+function codegen_constant(name, value) {
+	print ""
+	print "enum { " name " = " value " };"
+}
+
+function codegen_enum_value(name, subname, value, cg) {
+	append(cg, "fields", "\t" subname " = " value ",\n")
+}
+
+function codegen_enum(name, cg) {
+	print ""
+	print "enum struct " name " : int8_t {"
+	print cg["fields"] "};"
+
+	# XXX: This should also check if it isn't out-of-range for any reason,
+	# but our usage of sprintf() stands in the way a bit.
+	CodegenSerialize[name] = \
+		"\tw.append(static_cast<int8_t>(%s));\n"
+	CodegenDeserialize[name] = \
+		"\t{\n" \
+		"\t\tint8_t v = 0;\n" \
+		"\t\tif (!r.read(v) || !v)\n" \
+		"\t\t\treturn false;\n" \
+		"\t\t%s = static_cast<" name ">(v);\n" \
+		"\t}\n"
+
+	CodegenCType[name] = name
+	for (i in cg)
+		delete cg[i]
+}
+
+# Some identifiers do not pose a problem in C, but do in our C++.
+function codegen_struct_sanitize(name) {
+	if (name ~ /^(serialize|deserialize)_*$/ ||
+		name ~ /^(catch|class|delete|except|finally|friend|new|operator)_*$/ ||
+		name ~ /^(private|protected|public|template|this|throw|try|virtual)_*$/)
+		return name "_"
+	return name
+}
+
+function codegen_struct_tag(d, cg,    name, f) {
+	name = codegen_struct_sanitize(d["name"])
+	f = "this->" name
+
+	append(cg, "serialize", sprintf(CodegenSerialize[d["type"]], f))
+	# Do not deserialize here, that would be out of order.
+}
+
+function codegen_struct_field(d, cg,    name, f, serialize, deserialize) {
+	name = codegen_struct_sanitize(d["name"])
+	f = "this->" name
+
+	serialize = CodegenSerialize[d["type"]]
+	deserialize = CodegenDeserialize[d["type"]]
+	if (!d["isarray"]) {
+		append(cg, "fields",
+			"\t" CodegenCType[d["type"]] " " name " = {};\n")
+		append(cg, "serialize", sprintf(serialize, f))
+		append(cg, "deserialize", sprintf(deserialize, f))
+		return
+	}
+
+	append(cg, "fields",
+		"\tstd::vector<" CodegenCType[d["type"]] "> " name ";\n")
+
+	# XXX: We should probably pedantically check for overflows.
+	append(cg, "serialize",
+		sprintf(CodegenSerialize["u32"], "uint32_t(" f ".size())") \
+		"\tfor (const auto &it : " f ")\n" \
+		indent(sprintf(serialize, "it")))
+
+	if (d["type"] == "u8") {
+		append(cg, "deserialize",
+			"\tif (!r.read(" f "))\n" \
+			"\t\treturn false;\n")
+	} else if (deserialize) {
+		append(cg, "deserialize",
+			"\t{\n" \
+			"\t\tuint32_t size = 0;\n" \
+			indent(sprintf(CodegenDeserialize["u32"], "size")) \
+			"\t\t" f ".resize(size);\n" \
+			"\t}\n" \
+			"\tfor (auto &it : " f ")\n" \
+			indent(sprintf(deserialize, "it")))
+	}
+}
+
+function codegen_struct(name, cg) {
+	print ""
+	print "struct " name " {"
+	print cg["fields"]
+	print "\tbool serialize(LibertyXDR::Writer &w) const {"
+	print indent(cg["serialize"]) "\t\treturn true;"
+	print "\t}"
+	print ""
+	print "\tbool deserialize([[maybe_unused]] LibertyXDR::Reader &r) {"
+	print indent(cg["deserialize"]) "\t\treturn true;"
+	print "\t}"
+	print "};"
+
+	CodegenSerialize[name] = "\tif (!%s->serialize(w))\n" \
+		"\t\treturn false;\n"
+	CodegenDeserialize[name] = "\tif (!%s->deserialize(r))\n" \
+		"\t\treturn false;\n"
+
+	CodegenCType[name] = name
+	for (i in cg)
+		delete cg[i]
+}
+
+function codegen_union_tag(name, d, cg,    tagname) {
+	cg["tagtype"] = d["type"]
+	cg["tagname"] = tagname = codegen_struct_sanitize(d["name"])
+
+	print ""
+	print "struct " name " {"
+	print "\t" CodegenCType[d["type"]] " " tagname " = {};"
+	print "\tvirtual ~" name "() = 0;"
+	print "\tvirtual bool serialize(LibertyXDR::Writer &w) const = 0;"
+	print "\tvirtual bool deserialize(LibertyXDR::Reader &r) = 0;"
+	print "};"
+	print ""
+	print name "::~" name "() {}"
+}
+
+function codegen_union_struct(name, casename, cg, scg,     structname) {
+	# And thus not all generated structs are present in Types.
+	structname = name "_" snaketocamel(casename)
+
+	print ""
+	print "struct " structname " : virtual public " name " {"
+	print scg["fields"]
+	print "\t" structname "() {"
+	print "\t\tthis->" cg["tagname"] " = " \
+		CodegenCType[cg["tagtype"]] "::" casename ";"
+	print "\t}"
+	print ""
+	print "\tvirtual bool serialize(LibertyXDR::Writer &w) const {"
+	print indent(scg["serialize"]) "\t\treturn true;"
+	print "\t}"
+	print ""
+	print "\tvirtual bool deserialize([[maybe_unused]] LibertyXDR::Reader &r) {"
+	print indent(scg["deserialize"]) "\t\treturn true;"
+	print "\t}"
+	print "};"
+
+	append(cg, "deserialize",
+		"\tcase " CodegenCType[cg["tagtype"]] "::" casename ":\n" \
+		"\t\treturn new " structname "();\n")
+
+	CodegenSerialize[structname] = "\tif (!%s->serialize(w))\n" \
+		"\t\treturn false;\n"
+	CodegenDeserialize[structname] = "\tif (!%s->deserialize(r))\n" \
+		"\t\treturn false;\n"
+
+	CodegenCType[structname] = structname
+	for (i in scg)
+		delete scg[i]
+}
+
+function codegen_union(name, cg, exhaustive,    ctype) {
+	CodegenSerialize[name] = "\tif (!%s->serialize(w))\n" \
+		"\t\treturn false;\n"
+
+	ctype = "std::unique_ptr<" name ">"
+	if (cg["deserialize"]) {
+		print ""
+		print "static " name " *read" name "(" \
+			CodegenCType[cg["tagtype"]] " " cg["tagname"] ") {"
+		print "\tswitch (" cg["tagname"] ") {"
+		print cg["deserialize"] "\tdefault:"
+		print "\t\treturn nullptr;"
+		print "\t}"
+		print "}"
+		print ""
+		print "static " ctype " read" name "(LibertyXDR::Reader &r) {"
+		print "\tint8_t v = 0;"
+		print "\tif (!r.read(v) || !v)"
+		print "\t\treturn nullptr;"
+		print ""
+		print "\t" ctype " result(read" name "(static_cast<" \
+			CodegenCType[cg["tagtype"]] ">(v)));"
+		print "\tif (!result || !result->deserialize(r))"
+		print "\t\treturn nullptr;"
+		print "\treturn result;"
+		print "}"
+
+		CodegenDeserialize[name] = "\tif (!(%s = read" name "(r)))\n" \
+			"\t\treturn false;\n"
+	}
+
+	CodegenCType[name] = ctype
+	for (i in cg)
+		delete cg[i]
+}

tools/lxdrgen-go.awk

@@ -0,0 +1,540 @@
+# lxdrgen-go.awk: Go backend for lxdrgen.awk.
+#
+# Copyright (c) 2022, Přemysl Eric Janouch <p@janouch.name>
+# SPDX-License-Identifier: 0BSD
+#
+# This backend also enables proxying to other endpoints using JSON.
+
+function define_internal(name, gotype) {
+	Types[name] = "internal"
+	CodegenGoType[name] = gotype
+}
+
+function define_sint(size,    shortname, gotype) {
+	shortname = "i" size
+	gotype = "int" size
+	define_internal(shortname, gotype)
+
+	CodegenAppendJSON[shortname] = \
+		"\tb = strconv.AppendInt(b, int64(%s), 10)\n"
+	if (size == 8) {
+		CodegenSerialize[shortname] = "\tdata = append(data, uint8(%s))\n"
+		CodegenDeserialize[shortname] = \
+			"\tif len(data) >= 1 {\n" \
+			"\t\t%s, data = int8(data[0]), data[1:]\n" \
+			"\t} else {\n" \
+			"\t\treturn nil, false\n" \
+			"\t}\n"
+		return
+	}
+
+	CodegenSerialize[shortname] = \
+		"\tdata = binary.BigEndian.AppendUint" size "(data, uint" size "(%s))\n"
+	CodegenDeserialize[shortname] = \
+		"\tif len(data) >= " (size / 8) " {\n" \
+		"\t\t%s = " gotype "(binary.BigEndian.Uint" size "(data))\n" \
+		"\t\tdata = data[" (size / 8) ":]\n" \
+		"\t} else {\n" \
+		"\t\treturn nil, false\n" \
+		"\t}\n"
+}
+
+function define_uint(size,    shortname, gotype) {
+	# Both []byte and []uint8 luckily marshal as base64-encoded JSON strings,
+	# so there's no need to rename the type as an exception.
+	shortname = "u" size
+	gotype = "uint" size
+	define_internal(shortname, gotype)
+
+	CodegenAppendJSON[shortname] = \
+		"\tb = strconv.AppendUint(b, uint64(%s), 10)\n"
+	if (size == 8) {
+		CodegenSerialize[shortname] = "\tdata = append(data, %s)\n"
+		CodegenDeserialize[shortname] = \
+			"\tif len(data) >= 1 {\n" \
+			"\t\t%s, data = data[0], data[1:]\n" \
+			"\t} else {\n" \
+			"\t\treturn nil, false\n" \
+			"\t}\n"
+		return
+	}
+
+	CodegenSerialize[shortname] = \
+		"\tdata = binary.BigEndian.AppendUint" size "(data, %s)\n"
+	CodegenDeserialize[shortname] = \
+		"\tif len(data) >= " (size / 8) " {\n" \
+		"\t\t%s = binary.BigEndian.Uint" size "(data)\n" \
+		"\t\tdata = data[" (size / 8) ":]\n" \
+		"\t} else {\n" \
+		"\t\treturn nil, false\n" \
+		"\t}\n"
+}
+
+# Currently two outputs cannot coexist within the same package.
+function codegen_private(name) {
+	return "proto" name
+}
+
+function codegen_begin(    funcname) {
+	define_sint("8")
+	define_sint("16")
+	define_sint("32")
+	define_sint("64")
+	define_uint("8")
+	define_uint("16")
+	define_uint("32")
+	define_uint("64")
+	define_internal("bool", "bool")
+	define_internal("string", "string")
+
+	# Cater to "go generate", for what it's worth.
+	CodegenPackage = ENV["GOPACKAGE"]
+	if (!CodegenPackage)
+		CodegenPackage = "main"
+
+	print "// Code generated from " FILENAME ". DO NOT EDIT."
+	print ""
+	print "package " CodegenPackage
+	print ""
+	print "import ("
+	print "\t`encoding/base64`"
+	print "\t`encoding/binary`"
+	print "\t`encoding/json`"
+	print "\t`errors`"
+	print "\t`math`"
+	print "\t`strconv`"
+	print "\t`unicode/utf8`"
+	print ")"
+	print ""
+	print "// This is a hack to always use the base64 import."
+	print "var _ = base64.StdEncoding"
+	print ""
+
+	CodegenAppendJSON["bool"] = \
+		"\tb = strconv.AppendBool(b, %s)\n"
+	CodegenSerialize["bool"] = \
+		"\tif %s {\n" \
+		"\t\tdata = append(data, 1)\n" \
+		"\t} else {\n" \
+		"\t\tdata = append(data, 0)\n" \
+		"\t}\n"
+
+	funcname = codegen_private("ConsumeBoolFrom")
+	print "// " funcname " tries to deserialize a boolean value"
+	print "// from the beginning of a byte stream. When successful,"
+	print "// it returns a subslice with any data that might follow."
+	print "func " funcname "(data []byte, b *bool) ([]byte, bool) {"
+	print "\tif len(data) < 1 {"
+	print "\t\treturn nil, false"
+	print "\t}"
+	print "\tif data[0] != 0 {"
+	print "\t\t*b = true"
+	print "\t} else {"
+	print "\t\t*b = false"
+	print "\t}"
+	print "\treturn data[1:], true"
+	print "}"
+	print ""
+
+	CodegenDeserialize["bool"] = \
+		"\tif data, ok = " funcname "(data, &%s); !ok {\n" \
+		"\t\treturn nil, ok\n" \
+		"\t}\n"
+
+	funcname = codegen_private("AppendStringTo")
+	print "// " funcname " tries to serialize a string value,"
+	print "// appending it to the end of a byte stream."
+	print "func " funcname "(data []byte, s string) ([]byte, bool) {"
+	print "\tif len(s) > math.MaxUint32 {"
+	print "\t\treturn nil, false"
+	print "\t}"
+	print "\tdata = binary.BigEndian.AppendUint32(data, uint32(len(s)))"
+	print "\treturn append(data, s...), true"
+	print "}"
+	print ""
+
+	CodegenSerialize["string"] = \
+		"\tif data, ok = " funcname "(data, %s); !ok {\n" \
+		"\t\treturn nil, ok\n" \
+		"\t}\n"
+
+	funcname = codegen_private("ConsumeStringFrom")
+	print "// " funcname " tries to deserialize a string value"
+	print "// from the beginning of a byte stream. When successful,"
+	print "// it returns a subslice with any data that might follow."
+	print "func " funcname "(data []byte, s *string) ([]byte, bool) {"
+	print "\tif len(data) < 4 {"
+	print "\t\treturn nil, false"
+	print "\t}"
+	print "\tlength := binary.BigEndian.Uint32(data)"
+	print "\tif data = data[4:]; uint64(len(data)) < uint64(length) {"
+	print "\t\treturn nil, false"
+	print "\t}"
+	print "\t*s = string(data[:length])"
+	print "\tif !utf8.ValidString(*s) {"
+	print "\t\treturn nil, false"
+	print "\t}"
+	print "\treturn data[length:], true"
+	print "}"
+	print ""
+
+	CodegenDeserialize["string"] = \
+		"\tif data, ok = " funcname "(data, &%s); !ok {\n" \
+		"\t\treturn nil, ok\n" \
+		"\t}\n"
+
+	funcname = codegen_private("UnmarshalEnumJSON")
+	print "// " funcname " converts a JSON fragment to an integer,"
+	print "// ensuring that it's within the expected range of enum values."
+	print "func " funcname "(data []byte) (int64, error) {"
+	print "\tvar n int64"
+	print "\tif err := json.Unmarshal(data, &n); err != nil {"
+	print "\t\treturn 0, err"
+	print "\t} else if n > math.MaxInt8 || n < math.MinInt8 {"
+	print "\t\treturn 0, errors.New(`integer out of range`)"
+	print "\t} else {"
+	print "\t\treturn n, nil"
+	print "\t}"
+	print "}"
+	print ""
+}
+
+function codegen_constant(name, value) {
+	print "const " PrefixCamel snaketocamel(name) " = " value
+	print ""
+}
+
+function codegen_enum_value(name, subname, value, cg,    goname) {
+	goname = PrefixCamel name snaketocamel(subname)
+	append(cg, "fields",
+		"\t" goname " = " value "\n")
+	append(cg, "stringer",
+		"\tcase " goname ":\n" \
+		"\t\treturn `" snaketocamel(subname) "`\n")
+	append(cg, "marshal",
+		goname ",\n")
+	append(cg, "unmarshal",
+		"\tcase `" snaketocamel(subname) "`:\n" \
+		"\t\t*v = " goname "\n")
+}
+
+function codegen_enum(name, cg,    gotype, fields, funcname) {
+	gotype = PrefixCamel name
+	print "type " gotype " int8"
+	print ""
+
+	print "const ("
+	print cg["fields"] ")"
+	print ""
+
+	print "func (v " gotype ") String() string {"
+	print "\tswitch v {"
+	print cg["stringer"] "\tdefault:"
+	print "\t\treturn strconv.Itoa(int(v))"
+	print "\t}"
+	print "}"
+	print ""
+
+	CodegenIsMarshaler[name] = 1
+	fields = cg["marshal"]
+	sub(/,\n$/, ":", fields)
+	gsub(/\n/, "\n\t", fields)
+	print "func (v " gotype ") MarshalJSON() ([]byte, error) {"
+	print "\tswitch v {"
+	print indent("case " fields)
+	print "\t\treturn []byte(`\"` + v.String() + `\"`), nil"
+	print "\t}"
+	print "\treturn json.Marshal(int(v))"
+	print "}"
+	print ""
+
+	funcname = codegen_private("UnmarshalEnumJSON")
+	print "func (v *" gotype ") UnmarshalJSON(data []byte) error {"
+	print "\tvar s string"
+	print "\tif json.Unmarshal(data, &s) == nil {"
+	print "\t\t// Handled below."
+	print "\t} else if n, err := " funcname "(data); err != nil {"
+	print "\t\treturn err"
+	print "\t} else {"
+	print "\t\t*v = " gotype "(n)"
+	print "\t\treturn nil"
+	print "\t}"
+	print ""
+	print "\tswitch s {"
+	print cg["unmarshal"] "\tdefault:"
+	print "\t\treturn errors.New(`unrecognized value: ` + s)"
+	print "\t}"
+	print "\treturn nil"
+	print "}"
+	print ""
+
+	# XXX: This should also check if it isn't out-of-range for any reason,
+	# but our usage of sprintf() stands in the way a bit.
+	CodegenSerialize[name] = "\tdata = append(data, uint8(%s))\n"
+	CodegenDeserialize[name] = \
+		"\tif len(data) >= 1 {\n" \
+		"\t\t%s, data = " gotype "(data[0]), data[1:]\n" \
+		"\t} else {\n" \
+		"\t\treturn nil, false\n" \
+		"\t}\n"
+
+	CodegenGoType[name] = gotype
+	for (i in cg)
+		delete cg[i]
+}
+
+function codegen_marshal(type, f,    marshal) {
+	if (CodegenAppendJSON[type])
+		return sprintf(CodegenAppendJSON[type], f)
+
+	# Complex types are json.Marshalers, there's no need to json.Marshal(&f).
+	if (CodegenIsMarshaler[type])
+		marshal = f ".MarshalJSON()"
+	else
+		marshal = "json.Marshal(" f ")"
+
+	return \
+		"\tif j, err := " marshal "; err != nil {\n" \
+		"\t\treturn nil, err\n" \
+		"\t} else {\n" \
+		"\t\tb = append(b, j...)\n" \
+		"\t}\n"
+}
+
+function codegen_struct_field_marshal(d, cg,    camel, f, marshal) {
+	camel = snaketocamel(d["name"])
+	f = "s." camel
+	if (!d["isarray"]) {
+		append(cg, "marshal",
+			"\tb = append(b, `,\"" decapitalize(camel) "\":`...)\n" \
+			codegen_marshal(d["type"], f))
+		return
+	}
+
+	# Note that we do not produce `null` for nil slices, unlike encoding/json.
+	# And arrays never get deserialized as such.
+	if (d["type"] == "u8") {
+		append(cg, "marshal",
+			"\tb = append(b, `,\"" decapitalize(camel) "\":\"`...)\n" \
+			"\tb = append(b, base64.StdEncoding.EncodeToString(" f ")...)\n" \
+			"\tb = append(b, '\"')\n")
+		return
+	}
+
+	append(cg, "marshal",
+		"\tb = append(b, `,\"" decapitalize(camel) "\":[`...)\n" \
+		"\tfor i := 0; i < len(" f "); i++ {\n" \
+		"\t\tif i > 0 {\n" \
+		"\t\t\tb = append(b, ',')\n" \
+		"\t\t}\n" \
+		indent(codegen_marshal(d["type"], f "[i]")) \
+		"\t}\n" \
+		"\tb = append(b, ']')\n")
+}
+
+function codegen_struct_field(d, cg,    camel, f, serialize, deserialize) {
+	codegen_struct_field_marshal(d, cg)
+
+	camel = snaketocamel(d["name"])
+	f = "s." camel
+	serialize = CodegenSerialize[d["type"]]
+	deserialize = CodegenDeserialize[d["type"]]
+	if (!d["isarray"]) {
+		append(cg, "fields", "\t" camel " " CodegenGoType[d["type"]] \
+			" `json:\"" decapitalize(camel) "\"`\n")
+		append(cg, "serialize", sprintf(serialize, f))
+		append(cg, "deserialize", sprintf(deserialize, f))
+		return
+	}
+
+	append(cg, "fields", "\t" camel " []" CodegenGoType[d["type"]] \
+		" `json:\"" decapitalize(camel) "\"`\n")
+
+	# XXX: This should also check if it isn't out-of-range for any reason.
+	append(cg, "serialize",
+		sprintf(CodegenSerialize["u32"], "uint32(len(" f "))"))
+	if (d["type"] == "u8") {
+		append(cg, "serialize",
+			"\tdata = append(data, " f "...)\n")
+	} else {
+		append(cg, "serialize",
+			"\tfor i := 0; i < len(" f "); i++ {\n" \
+			indent(sprintf(serialize, f "[i]")) \
+			"\t}\n")
+	}
+
+	append(cg, "deserialize",
+		"\t{\n" \
+		"\t\tvar length uint32\n" \
+		indent(sprintf(CodegenDeserialize["u32"], "length")))
+	if (d["type"] == "u8") {
+		append(cg, "deserialize",
+			"\t\tif uint64(len(data)) < uint64(length) {\n" \
+			"\t\t\treturn nil, false\n" \
+			"\t\t}\n" \
+			"\t\t" f ", data = data[:length], data[length:]\n" \
+			"\t}\n")
+	} else {
+		append(cg, "deserialize",
+			"\t\t" f " = make([]" CodegenGoType[d["type"]] ", length)\n" \
+			"\t}\n" \
+			"\tfor i := 0; i < len(" f "); i++ {\n" \
+			indent(sprintf(deserialize, f "[i]")) \
+			"\t}\n")
+	}
+}
+
+function codegen_struct_tag(d, cg,    camel, f) {
+	codegen_struct_field_marshal(d, cg)
+
+	camel = snaketocamel(d["name"])
+	f = "s." camel
+	append(cg, "fields", "\t" camel " " CodegenGoType[d["type"]] \
+		" `json:\"" decapitalize(camel) "\"`\n")
+	append(cg, "serialize", sprintf(CodegenSerialize[d["type"]], f))
+	# Do not deserialize here, that is already done by the containing union.
+}
+
+function codegen_struct(name, cg,    gotype) {
+	gotype = PrefixCamel name
+	print "type " gotype " struct {\n" cg["fields"] "}\n"
+
+	if (cg["marshal"]) {
+		CodegenIsMarshaler[name] = 1
+		print "func (s *" gotype ") MarshalJSON() ([]byte, error) {"
+		print "\tb := []byte{}"
+		print cg["marshal"] "\tb[0] = '{'"
+		print "\treturn append(b, '}'), nil"
+		print "}"
+		print ""
+	}
+
+	if (cg["serialize"]) {
+		print "func (s *" gotype ") AppendTo(data []byte) ([]byte, bool) {"
+		print "\tok := true"
+		print cg["serialize"] "\treturn data, ok"
+		print "}"
+		print ""
+
+		CodegenSerialize[name] = \
+			"\tif data, ok = %s.AppendTo(data); !ok {\n" \
+			"\t\treturn nil, ok\n" \
+			"\t}\n"
+	}
+	if (cg["deserialize"]) {
+		print "func (s *" gotype ") ConsumeFrom(data []byte) ([]byte, bool) {"
+		print "\tok := true"
+		print cg["deserialize"] "\treturn data, ok"
+		print "}"
+		print ""
+
+		CodegenDeserialize[name] = \
+			"\tif data, ok = %s.ConsumeFrom(data); !ok {\n" \
+			"\t\treturn nil, ok\n" \
+			"\t}\n"
+	}
+
+	CodegenGoType[name] = gotype
+	for (i in cg)
+		delete cg[i]
+}
+
+function codegen_union_tag(name, d, cg) {
+	cg["tagtype"] = d["type"]
+	cg["tagname"] = snaketocamel(d["name"])
+	# The tag is implied from the type of struct stored in the interface.
+}
+
+function codegen_union_struct(name, casename, cg, scg,     structname, init) {
+	# And thus not all generated structs are present in Types.
+	structname = name snaketocamel(casename)
+	codegen_struct(structname, scg)
+
+	init = CodegenGoType[structname] "{" cg["tagname"] \
+		": " decapitalize(cg["tagname"]) "}"
+	append(cg, "unmarshal",
+		"\tcase " CodegenGoType[cg["tagtype"]] snaketocamel(casename) ":\n" \
+		"\t\ts := " init "\n" \
+		"\t\terr = json.Unmarshal(data, &s)\n" \
+		"\t\tu.Interface = &s\n")
+	append(cg, "serialize",
+		"\tcase *" CodegenGoType[structname] ":\n" \
+		indent(sprintf(CodegenSerialize[structname], "union")))
+	append(cg, "deserialize",
+		"\tcase " CodegenGoType[cg["tagtype"]] snaketocamel(casename) ":\n" \
+		"\t\ts := " init "\n" \
+		indent(sprintf(CodegenDeserialize[structname], "s")) \
+		"\t\tu.Interface = &s\n")
+}
+
+function codegen_union(name, cg, exhaustive,    gotype, tagvar) {
+	gotype = PrefixCamel name
+	print "type " gotype " struct {"
+	print "\tInterface any"
+	print "}"
+	print ""
+
+	# This cannot be a pointer method, it wouldn't work recursively.
+	CodegenIsMarshaler[name] = 1
+	print "func (u " gotype ") MarshalJSON() ([]byte, error) {"
+	print "\treturn u.Interface.(json.Marshaler).MarshalJSON()"
+	print "}"
+	print ""
+
+	tagvar = decapitalize(cg["tagname"])
+	print "func (u *" gotype ") UnmarshalJSON(data []byte) (err error) {"
+	print "\tvar t struct {"
+	print "\t\t" cg["tagname"] " " CodegenGoType[cg["tagtype"]] \
+		" `json:\"" tagvar "\"`"
+	print "\t}"
+	print "\tif err := json.Unmarshal(data, &t); err != nil {"
+	print "\t\treturn err"
+	print "\t}"
+	print ""
+	print "\tswitch " tagvar " := t." cg["tagname"] "; " tagvar " {"
+	print cg["unmarshal"] "\tdefault:"
+	print "\t\terr = errors.New(`unsupported value: ` + " tagvar ".String())"
+	print "\t}"
+	print "\treturn err"
+	print "}"
+	print ""
+
+	# XXX: Consider changing the interface into an AppendTo/ConsumeFrom one,
+	# that would eliminate these type case switches entirely.
+	# On the other hand, it would make it possible to send unsuitable structs.
+	print "func (u *" gotype ") AppendTo(data []byte) ([]byte, bool) {"
+	print "\tok := true"
+	print "\tswitch union := u.Interface.(type) {"
+	print cg["serialize"] "\tdefault:"
+	print "\t\treturn nil, false"
+	print "\t}"
+	print "\treturn data, ok"
+	print "}"
+	print ""
+
+	CodegenSerialize[name] = \
+		"\tif data, ok = %s.AppendTo(data); !ok {\n" \
+		"\t\treturn nil, ok\n" \
+		"\t}\n"
+
+	print "func (u *" gotype ") ConsumeFrom(data []byte) ([]byte, bool) {"
+	print "\tok := true"
+	print "\tvar " tagvar " " CodegenGoType[cg["tagtype"]]
+	print sprintf(CodegenDeserialize[cg["tagtype"]], tagvar)
+	print "\tswitch " tagvar " {"
+	print cg["deserialize"] "\tdefault:"
+	print "\t\treturn nil, false"
+	print "\t}"
+	print "\treturn data, ok"
+	print "}"
+	print ""
+
+	CodegenDeserialize[name] = \
+		"\tif data, ok = %s.ConsumeFrom(data); !ok {\n" \
+		"\t\treturn nil, ok\n" \
+		"\t}\n"
+
+	CodegenGoType[name] = gotype
+	for (i in cg)
+		delete cg[i]
+}

tools/lxdrgen-mjs.awk

@@ -0,0 +1,226 @@
+# lxdrgen-mjs.awk: Javascript backend for lxdrgen.awk.
+#
+# Copyright (c) 2022, Přemysl Eric Janouch <p@janouch.name>
+# SPDX-License-Identifier: 0BSD
+#
+# This backend is currently for decoding the binary format only.
+# (JSON is way too expensive to process and transfer.)
+#
+# Import the resulting script as a Javascript module.
+# Identifiers intentionally aren't prefixed.
+
+function define_internal(name) {
+	Types[name] = "internal"
+}
+
+function define_sint(size,    shortname) {
+	shortname = "i" size
+	define_internal(shortname)
+	CodegenDeserialize[shortname] = "\t%s = r." shortname "()\n"
+
+	print ""
+	print "\t" shortname "() {"
+	if (size == "64") {
+		# XXX: 2^53 - 1 must be enough for anyone.  BigInts are a PITA.
+		print "\t\tconst " shortname \
+			" = Number(this.getBigInt" size "(this.offset))"
+	} else {
+		print "\t\tconst " shortname " = this.getInt" size "(this.offset)"
+	}
+	print "\t\tthis.offset += " (size / 8)
+	print "\t\treturn " shortname
+	print "\t}"
+}
+
+function define_uint(size,    shortname) {
+	shortname = "u" size
+	define_internal(shortname)
+	CodegenDeserialize[shortname] = "\t%s = r." shortname "()\n"
+
+	print ""
+	print "\t" shortname "() {"
+	if (size == "64") {
+		# XXX: 2^53 - 1 must be enough for anyone.  BigInts are a PITA.
+		print "\t\tconst " shortname \
+			" = Number(this.getBigUint" size "(this.offset))"
+	} else {
+		print "\t\tconst " shortname " = this.getUint" size "(this.offset)"
+	}
+	print "\t\tthis.offset += " (size / 8)
+	print "\t\treturn " shortname
+	print "\t}"
+}
+
+function codegen_begin() {
+	print "// Code generated from " FILENAME ". DO NOT EDIT."
+	print ""
+	print "export class Reader extends DataView {"
+	print "\tconstructor() {"
+	print "\t\tsuper(...arguments)"
+	print "\t\tthis.offset = 0"
+	print "\t\tthis.decoder = new TextDecoder('utf-8', {fatal: true})"
+	print "\t}"
+	print ""
+	print "\tget empty() {"
+	print "\t\treturn this.byteLength <= this.offset"
+	print "\t}"
+	print ""
+	print "\trequire(len) {"
+	print "\t\tif (this.byteLength - this.offset < len)"
+	print "\t\t\tthrow `Premature end of data`"
+	print "\t\treturn this.byteOffset + this.offset"
+	print "\t}"
+
+	define_internal("string")
+	CodegenDeserialize["string"] = "\t%s = r.string()\n"
+
+	print ""
+	print "\tstring() {"
+	print "\t\tconst len = this.getUint32(this.offset)"
+	print "\t\tthis.offset += 4"
+	print "\t\tconst array = new Uint8Array("
+	print "\t\t\tthis.buffer, this.require(len), len)"
+	print "\t\tthis.offset += len"
+	print "\t\treturn this.decoder.decode(array)"
+	print "\t}"
+
+	define_internal("bool")
+	CodegenDeserialize["bool"] = "\t%s = r.bool()\n"
+
+	print ""
+	print "\tbool() {"
+	print "\t\tconst u8 = this.getUint8(this.offset)"
+	print "\t\tthis.offset += 1"
+	print "\t\treturn u8 != 0"
+	print "\t}"
+
+	define_sint("8")
+	define_sint("16")
+	define_sint("32")
+	define_sint("64")
+	define_uint("8")
+	define_uint("16")
+	define_uint("32")
+	define_uint("64")
+
+	print "}"
+}
+
+function codegen_constant(name, value) {
+	print ""
+	print "export const " decapitalize(snaketocamel(name)) " = " value
+}
+
+function codegen_enum_value(name, subname, value, cg) {
+	append(cg, "fields", "\t" snaketocamel(subname) ": " value ",\n")
+}
+
+function codegen_enum(name, cg) {
+	print ""
+	print "export const " name " = Object.freeze({"
+	print cg["fields"] "})"
+
+	CodegenDeserialize[name] = "\t%s = r.i8()\n"
+	for (i in cg)
+		delete cg[i]
+}
+
+function codegen_struct_field(d, cg,    camel, f, deserialize) {
+	camel = decapitalize(snaketocamel(d["name"]))
+	f = "s." camel
+	append(cg, "fields", "\t" camel "\n")
+
+	deserialize = CodegenDeserialize[d["type"]]
+	if (!d["isarray"]) {
+		append(cg, "deserialize", sprintf(deserialize, f))
+		return
+	}
+
+	append(cg, "deserialize",
+		"\t{\n" \
+		indent(sprintf(CodegenDeserialize["u32"], "const len")))
+	if (d["type"] == "u8") {
+		append(cg, "deserialize",
+			"\t\t" f " = new Uint8Array(\n" \
+			"\t\t\tr.buffer, r.require(len), len)\n" \
+			"\t\tr.offset += len\n" \
+			"\t}\n")
+		return
+	}
+	if (d["type"] == "i8") {
+		append(cg, "deserialize",
+			"\t\t" f " = new Int8Array(\n" \
+			"\t\t\tr.buffer, r.require(len), len)\n" \
+			"\t\tr.offset += len\n" \
+			"\t}\n")
+		return
+	}
+
+	append(cg, "deserialize",
+		"\t\t" f " = new Array(len)\n" \
+		"\t}\n" \
+		"\tfor (let i = 0; i < " f ".length; i++)\n" \
+		indent(sprintf(deserialize, f "[i]")))
+}
+
+function codegen_struct_tag(d, cg) {
+	append(cg, "fields", "\t" decapitalize(snaketocamel(d["name"])) "\n")
+	# Do not deserialize here, that is already done by the containing union.
+}
+
+function codegen_struct(name, cg) {
+	print ""
+	print "export class " name " {"
+	print cg["fields"] cg["methods"]
+	print "\tstatic deserialize(r) {"
+	print "\t\tconst s = new " name "()"
+	print indent(cg["deserialize"]) "\t\treturn s"
+	print "\t}"
+	print "}"
+
+	CodegenDeserialize[name] = "\t%s = " name ".deserialize(r)\n"
+	for (i in cg)
+		delete cg[i]
+}
+
+function codegen_union_tag(name, d, cg) {
+	cg["tagtype"] = d["type"]
+	cg["tagname"] = decapitalize(snaketocamel(d["name"]))
+}
+
+function codegen_union_struct(name, casename, cg, scg,     structname) {
+	append(scg, "methods",
+		"\n" \
+		"\tconstructor() {\n" \
+		"\t\tthis." cg["tagname"] \
+			" = " cg["tagtype"] "." snaketocamel(casename) "\n" \
+		"\t}\n")
+
+	# And thus not all generated structs are present in Types.
+	structname = name snaketocamel(casename)
+	codegen_struct(structname, scg)
+
+	append(cg, "deserialize",
+		"\tcase " cg["tagtype"] "." snaketocamel(casename) ":\n" \
+		"\t{\n" \
+		indent(sprintf(CodegenDeserialize[structname], "const s")) \
+		"\t\treturn s\n" \
+		"\t}\n")
+}
+
+function codegen_union(name, cg, exhaustive,    tagvar) {
+	tagvar = cg["tagname"]
+
+	print ""
+	print "export function deserialize" name "(r) {"
+	print sprintf(CodegenDeserialize[cg["tagtype"]], "const " tagvar) \
+		"\tswitch (" tagvar ") {"
+	print cg["deserialize"] "\tdefault:"
+	print "\t\tthrow `Unknown " cg["tagtype"] " (${tagvar})`"
+	print "\t}"
+	print "}"
+
+	CodegenDeserialize[name] = "\t%s = deserialize" name "(r)\n"
+	for (i in cg)
+		delete cg[i]
+}

tools/lxdrgen-swift.awk

@@ -0,0 +1,277 @@
+# lxdrgen-swift.awk: Swift backend for lxdrgen.awk.
+#
+# Copyright (c) 2023, Přemysl Eric Janouch <p@janouch.name>
+# SPDX-License-Identifier: 0BSD
+
+function define_internal(name, swifttype) {
+	Types[name] = "internal"
+	CodegenSwiftType[name] = swifttype
+	CodegenDeserialize[name] = "%s.read()"
+}
+
+function define_sint(size,    shortname, swifttype) {
+	shortname = "i" size
+	swifttype = "Int" size
+	define_internal(shortname, swifttype)
+}
+
+function define_uint(size,    shortname, swifttype) {
+	shortname = "u" size
+	swifttype = "UInt" size
+	define_internal(shortname, swifttype)
+}
+
+function codegen_begin() {
+	define_sint("8")
+	define_sint("16")
+	define_sint("32")
+	define_sint("64")
+	define_uint("8")
+	define_uint("16")
+	define_uint("32")
+	define_uint("64")
+	define_internal("bool", "Bool")
+	define_internal("string", "String")
+
+	print "// Code generated from " FILENAME ". DO NOT EDIT."
+	print "import Foundation"
+	print ""
+	print "public struct " PrefixCamel "Reader {"
+	print "\tpublic var data: Data"
+	print ""
+	print "\tpublic enum ReadError: Error {"
+	print "\t\tcase unexpectedEOF"
+	print "\t\tcase invalidEncoding"
+	print "\t\tcase overflow"
+	print "\t\tcase unexpectedValue"
+	print "\t}"
+	print ""
+	print "\tpublic mutating func read<T: FixedWidthInteger>() throws -> T {"
+	print "\t\tlet size = MemoryLayout<T>.size"
+	print "\t\tguard data.count >= size else {"
+	print "\t\t\tthrow ReadError.unexpectedEOF"
+	print "\t\t}"
+	print "\t\tvar acc: T = 0"
+	print "\t\tdata.prefix(size).forEach { acc = acc << 8 | T($0) }"
+	print "\t\tdata = data.dropFirst(size)"
+	print "\t\treturn acc"
+	print "\t}"
+	print ""
+	print "\tpublic mutating func read() throws -> Bool {"
+	print "\t\ttry read() != UInt8(0)"
+	print "\t}"
+	print ""
+	print "\tpublic mutating func read() throws -> String {"
+	print "\t\tlet size: UInt32 = try self.read()"
+	print "\t\tguard let count = Int(exactly: size) else {"
+	print "\t\t\tthrow ReadError.overflow"
+	print "\t\t}"
+	print "\t\tguard data.count >= count else {"
+	print "\t\t\tthrow ReadError.unexpectedEOF"
+	print "\t\t}"
+	print "\t\tdefer {"
+	print "\t\t\tdata = data.dropFirst(count)"
+	print "\t\t}"
+	print "\t\tif let s = String(data: data.prefix(count), encoding: .utf8) {"
+	print "\t\t\treturn s"
+	print "\t\t} else {"
+	print "\t\t\tthrow ReadError.invalidEncoding"
+	print "\t\t}"
+	print "\t}"
+	print ""
+	print "\tpublic mutating func read<" \
+		"T: RawRepresentable<Int8>>() throws -> T {"
+	print "\t\tguard let value = T(rawValue: try read()) else {"
+	print "\t\t\tthrow ReadError.unexpectedValue"
+	print "\t\t}"
+	print "\t\treturn value"
+	print "\t}"
+	print ""
+	print "\tpublic mutating func read<T>("
+	print "\t\t\t_ read: (inout Self) throws -> T) throws -> [T] {"
+	print "\t\tlet size: UInt32 = try self.read()"
+	print "\t\tguard let count = Int(exactly: size) else {"
+	print "\t\t\tthrow ReadError.overflow"
+	print "\t\t}"
+	print "\t\tvar array = [T]()"
+	print "\t\tarray.reserveCapacity(count)"
+	print "\t\tfor _ in 0..<count {"
+	print "\t\t\tarray.append(try read(&self))"
+	print "\t\t}"
+	print "\t\treturn array"
+	print "\t}"
+	print "}"
+	print ""
+	print "public struct " PrefixCamel "Writer {"
+	print "\tpublic var data = Data()"
+	print ""
+	print "\tpublic mutating func append<T: FixedWidthInteger>(_ number: T) {"
+	print "\t\tvar n = number.byteSwapped"
+	print "\t\tfor _ in 0..<MemoryLayout<T>.size {"
+	print "\t\t\tdata.append(UInt8(truncatingIfNeeded: n))"
+	print "\t\t\tn >>= 8"
+	print "\t\t}"
+	print "\t}"
+	print ""
+	print "\tpublic mutating func append(_ bool: Bool) {"
+	print "\t\tappend(UInt8(bool ? 1 : 0))"
+	print "\t}"
+	print ""
+	print "\tpublic mutating func append(_ string: String) {"
+	print "\t\tlet bytes = string.data(using: .utf8)!"
+	print "\t\tappend(UInt32(bytes.count))"
+	print "\t\tdata.append(bytes)"
+	print "\t}"
+	print ""
+	print "\tpublic mutating func append<T: " \
+		"RawRepresentable<Int8>>(_ value: T) {"
+	print "\t\tappend(value.rawValue)"
+	print "\t}"
+	print ""
+	print "\tpublic mutating func append<T>("
+	print "\t\t\t_ array: Array<T>, _ write: (inout Self, T) -> ()) {"
+	print "\t\tappend(UInt32(array.count))"
+	print "\t\tfor i in 0..<array.count {"
+	print "\t\t\twrite(&self, array[i])"
+	print "\t\t}"
+	print "\t}"
+	print ""
+	print "\tpublic mutating func append<T: " \
+		PrefixCamel "Encodable>(_ value: T) {"
+	print "\t\tvalue.encode(to: &self)"
+	print "\t}"
+	print "}"
+	print ""
+	print "public protocol " PrefixCamel "Encodable { " \
+		"func encode(to: inout " PrefixCamel "Writer) }"
+}
+
+function codegen_constant(name, value) {
+	print ""
+	print "public let " decapitalize(PrefixCamel snaketocamel(name)) " = " value
+}
+
+function codegen_enum_value(name, subname, value, cg) {
+	append(cg, "fields",
+		"\tcase " decapitalize(snaketocamel(subname)) " = " value "\n")
+}
+
+function codegen_enum(name, cg,    swifttype) {
+	swifttype = PrefixCamel name
+	print ""
+	print "public enum " swifttype ": Int8 {"
+	print cg["fields"] "}"
+
+	CodegenSwiftType[name] = swifttype
+	CodegenDeserialize[name] = "%s.read()"
+	for (i in cg)
+		delete cg[i]
+}
+
+function codegen_struct_field(d, cg,    camel) {
+	camel = decapitalize(snaketocamel(d["name"]))
+	if (!d["isarray"]) {
+		append(cg, "fields",
+			"\tpublic var " camel ": " CodegenSwiftType[d["type"]] "\n")
+		append(cg, "deserialize",
+			"\t\tself." camel " = try " \
+				sprintf(CodegenDeserialize[d["type"]], "from") "\n")
+		append(cg, "serialize",
+			"\t\tto.append(self." camel ")\n")
+		return
+	}
+
+	append(cg, "fields",
+		"\tpublic var " camel ": [" CodegenSwiftType[d["type"]] "]\n")
+	append(cg, "deserialize",
+		"\t\tself." camel " = try from.read() { r in try " \
+			sprintf(CodegenDeserialize[d["type"]], "r") " }\n")
+	append(cg, "serialize",
+		"\t\tto.append(self." camel ") { (w, value) in w.append(value) }\n")
+}
+
+function codegen_struct_tag(d, cg,    camel) {
+	camel = decapitalize(snaketocamel(d["name"]))
+	append(cg, "serialize",
+		"\t\tto.append(self." camel ")\n")
+}
+
+function codegen_struct(name, cg,    swifttype) {
+	swifttype = PrefixCamel name
+	print ""
+	print "public struct " swifttype " {\n" cg["fields"] "}"
+	print ""
+	print "extension " swifttype ": " PrefixCamel "Encodable {"
+	print "\tpublic init(from: inout " PrefixCamel "Reader) throws {"
+	print cg["deserialize"] "\t}"
+	print ""
+	print "\tpublic func encode(to: inout " PrefixCamel "Writer) {"
+	print cg["serialize"] "\t}"
+	print "}"
+
+	CodegenSwiftType[name] = swifttype
+	CodegenDeserialize[name] = "%s.read()"
+	for (i in cg)
+		delete cg[i]
+}
+
+function codegen_union_tag(name, d, cg) {
+	cg["tagtype"] = d["type"]
+	cg["tagname"] = decapitalize(snaketocamel(d["name"]))
+}
+
+function codegen_union_struct(name, casename, cg, scg,     swifttype) {
+	# And thus not all generated structs are present in Types.
+	swifttype = PrefixCamel name snaketocamel(casename)
+	casename = decapitalize(snaketocamel(casename))
+	print ""
+	print "public struct " swifttype ": " PrefixCamel name " {"
+	print "\tpublic var " cg["tagname"] \
+		": " CodegenSwiftType[cg["tagtype"]] " { ." casename " }"
+	print scg["fields"] "}"
+	print ""
+	print "extension " swifttype ": " PrefixCamel "Encodable {"
+	print "\tfileprivate init(from: inout " PrefixCamel "Reader) throws {"
+	print scg["deserialize"] "\t}"
+	print ""
+	print "\tpublic func encode(to: inout " PrefixCamel "Writer) {"
+	print scg["serialize"] "\t}"
+	print "}"
+
+	append(cg, "cases", "\tcase ." casename ":\n" \
+		"\t\treturn try " swifttype "(from: &from)\n")
+
+	CodegenSwiftType[name] = swifttype
+	CodegenDeserialize[name] = "%s.read()"
+	for (i in scg)
+		delete scg[i]
+}
+
+function codegen_union(name, cg, exhaustive,    swifttype, init) {
+	# Classes don't have automatic member-wise initializers,
+	# thus using structs and protocols.
+	swifttype = PrefixCamel name
+	print ""
+	print "public protocol " swifttype ": " PrefixCamel "Encodable {"
+	print "\tvar " cg["tagname"] ": " CodegenSwiftType[cg["tagtype"]] " { get }"
+	print "}"
+
+	if (!exhaustive)
+		append(cg, "cases", "\tdefault:\n" \
+			"\t\tthrow " PrefixCamel "Reader.ReadError.unexpectedValue\n")
+
+	init = decapitalize(swifttype)
+	print ""
+	print "public func " init \
+		"(from: inout " PrefixCamel "Reader) throws -> " swifttype " {"
+	print "\tlet " cg["tagname"] ": " CodegenSwiftType[cg["tagtype"]] \
+		" = try from.read()"
+	print "\tswitch " cg["tagname"] " {"
+	print cg["cases"] "\t}"
+	print "}"
+
+	CodegenSwiftType[name] = swifttype
+	CodegenDeserialize[name] = init "(from: &%s)"
+	for (i in cg)
+		delete cg[i]
+}

tools/lxdrgen.awk

@@ -0,0 +1,298 @@
+# lxdrgen.awk: an XDR-derived code generator for network protocols.
+#
+# Copyright (c) 2022 - 2023, Přemysl Eric Janouch <p@janouch.name>
+# SPDX-License-Identifier: 0BSD
+#
+# Usage: env LC_ALL=C awk -f lxdrgen.awk -f lxdrgen-{c,go,mjs}.awk \
+#  -v PrefixCamel=Foo foo.lxdr > foo.{c,go,mjs} | {clang-format,gofmt,...}
+
+# --- Utilities ----------------------------------------------------------------
+
+function cameltosnake(s) {
+	while (match(s, /[[:lower:]][[:upper:]]/)) {
+		s = substr(s, 1, RSTART) "_" \
+			tolower(substr(s, RSTART + 1, RLENGTH - 1)) \
+			substr(s, RSTART + RLENGTH)
+	}
+	return tolower(s)
+}
+
+function snaketocamel(s) {
+	s = toupper(substr(s, 1, 1)) tolower(substr(s, 2))
+	while (match(s, /_[[:alnum:]]/)) {
+		s = substr(s, 1, RSTART - 1) \
+			toupper(substr(s, RSTART + 1, RLENGTH - 1)) \
+			substr(s, RSTART + RLENGTH)
+	}
+	return s
+}
+
+function decapitalize(s) {
+	if (match(s, /^[[:upper:]][[:lower:]]/))
+		return tolower(substr(s, 1, 1)) substr(s, 2)
+	if (match(s, /^[[:upper:]]$/))
+		return tolower(s)
+	return s
+}
+
+function indent(s) {
+	if (!s)
+		return s
+
+	gsub(/\n/, "\n\t", s)
+	sub(/\t*$/, "", s)
+	return "\t" s
+}
+
+function append(a, key, value) {
+	a[key] = a[key] value
+}
+
+# --- Parsing ------------------------------------------------------------------
+
+function fatal(message) {
+	print "// " FILENAME ":" FNR ": fatal error: " message
+	print FILENAME ":" FNR ": fatal error: " message > "/dev/stderr"
+	exit 1
+}
+
+function skipcomment() {
+	do {
+		if (match($0, /[*]\//)) {
+			$0 = substr($0, RSTART + RLENGTH)
+			return
+		}
+	} while (getline > 0)
+	fatal("unterminated block comment")
+}
+
+function nexttoken() {
+	do {
+		if (match($0, /^[[:space:]]+/)) {
+			$0 = substr($0, RLENGTH + 1)
+		} else if (match($0, /^\/\/.*/)) {
+			$0 = ""
+		} else if (match($0, /^\/[*]/)) {
+			$0 = substr($0, RLENGTH + 1)
+			skipcomment()
+		} else if (match($0, /^[[:alpha:]][[:alnum:]_]*/)) {
+			Token = substr($0, 1, RLENGTH)
+			$0 = substr($0, RLENGTH + 1)
+			return Token
+		# AWK implementations rarely support non-decimal notations
+		# in their implicit string-to-number conversions.
+		} else if (match($0, /^(0|-?[1-9][0-9]*)/)) {
+			Token = substr($0, 1, RLENGTH)
+			$0 = substr($0, RLENGTH + 1)
+			return Token
+		} else if ($0) {
+			Token = substr($0, 1, 1)
+			$0 = substr($0, 2)
+			return Token
+		}
+	} while ($0 || getline > 0)
+	Token = ""
+	return Token
+}
+
+function expect(v) {
+	if (!v)
+		fatal("broken expectations at `" Token "' before `" $0 "'")
+	return v
+}
+
+function accept(what) {
+	if (Token != what)
+		return 0
+	nexttoken()
+	return 1
+}
+
+function identifier(    v) {
+	if (Token !~ /^[[:alpha:]]/)
+		return 0
+	v = Token
+	nexttoken()
+	return v
+}
+
+function number(    v) {
+	if (Token !~ /^(0|-?[1-9])/)
+		return 0
+	v = Token
+	nexttoken()
+	return v
+}
+
+function readnumber(    ident) {
+	ident = identifier()
+	if (!ident)
+		return expect(number())
+	if (!(ident in Consts))
+		fatal("unknown constant: " ident)
+	return Consts[ident]
+}
+
+function defconst(    ident, num) {
+	if (!accept("const"))
+		return 0
+
+	ident = expect(identifier())
+	expect(accept("="))
+	num = readnumber()
+	if (ident in Consts)
+		fatal("constant redefined: " ident)
+
+	Consts[ident] = num
+	codegen_constant(ident, num)
+	return 1
+}
+
+function readtype(    ident) {
+	ident = deftype()
+	if (ident)
+		return ident
+
+	ident = identifier()
+	if (!ident)
+		return 0
+
+	if (!(ident in Types))
+		fatal("unknown type: " ident)
+	return ident
+}
+
+function defenum(    name, ident, value, cg) {
+	delete cg[0]
+
+	name = expect(identifier())
+	expect(accept("{"))
+	while (!accept("}")) {
+		ident = expect(identifier())
+		value = value + 1
+		if (accept("="))
+			value = readnumber() + 0
+		if (!value)
+			fatal("enumeration values cannot be zero")
+		if (value < -128 || value > 127)
+			fatal("enumeration value out of range")
+		expect(accept(","))
+		append(EnumValues, name, SUBSEP ident)
+		if (EnumValues[name, ident]++)
+			fatal("duplicate enum value: " ident)
+		codegen_enum_value(name, ident, value, cg)
+	}
+
+	Types[name] = "enum"
+	codegen_enum(name, cg)
+	return name
+}
+
+function readfield(out,    nonvoid) {
+	nonvoid = !accept("void")
+	if (nonvoid) {
+		out["type"] = expect(readtype())
+		out["name"] = expect(identifier())
+		# TODO: Consider supporting XDR's VLA length limits here.
+		# TODO: Consider supporting XDR's fixed-length syntax for string limits.
+		out["isarray"] = accept("<") && expect(accept(">"))
+	}
+	expect(accept(";"))
+	return nonvoid
+}
+
+function defstruct(    name, d, cg) {
+	delete d[0]
+	delete cg[0]
+
+	name = expect(identifier())
+	expect(accept("{"))
+	while (!accept("}")) {
+		if (readfield(d))
+			codegen_struct_field(d, cg)
+	}
+
+	Types[name] = "struct"
+	codegen_struct(name, cg)
+	return name
+}
+
+function defunion(    name, tag, tagtype, tagvalue, cg, scg, d, a, i,
+		unseen, exhaustive) {
+	delete cg[0]
+	delete scg[0]
+	delete d[0]
+
+	name = expect(identifier())
+	expect(accept("switch"))
+	expect(accept("("))
+	tag["type"] = tagtype = expect(readtype())
+	tag["name"] = expect(identifier())
+	expect(accept(")"))
+
+	if (Types[tagtype] != "enum")
+		fatal("not an enum type: " tagtype)
+	codegen_union_tag(name, tag, cg)
+
+	split(EnumValues[tagtype], a, SUBSEP)
+	for (i in a)
+		unseen[a[i]]++
+
+	expect(accept("{"))
+	while (!accept("}")) {
+		if (accept("case")) {
+			if (tagvalue)
+				codegen_union_struct(name, tagvalue, cg, scg)
+
+			tagvalue = expect(identifier())
+			expect(accept(":"))
+			if (!unseen[tagvalue]--)
+				fatal("no such value or duplicate case: " tagtype "." tagvalue)
+			codegen_struct_tag(tag, scg)
+		} else if (tagvalue) {
+			if (readfield(d))
+				codegen_struct_field(d, scg)
+		} else {
+			fatal("union fields must fall under a case")
+		}
+	}
+	if (tagvalue)
+		codegen_union_struct(name, tagvalue, cg, scg)
+
+	# Unseen cases are simply not recognized/allowed.
+	exhaustive = 1
+	for (i in unseen)
+		if (i && unseen[i])
+			exhaustive = 0
+
+	Types[name] = "union"
+	codegen_union(name, cg, exhaustive)
+	return name
+}
+
+function deftype() {
+	if (accept("enum"))
+		return defenum()
+	if (accept("struct"))
+		return defstruct()
+	if (accept("union"))
+		return defunion()
+	return 0
+}
+
+{
+	if (PrefixCamel) {
+		PrefixLower = tolower(cameltosnake(PrefixCamel)) "_"
+		PrefixUpper = toupper(cameltosnake(PrefixCamel)) "_"
+	}
+
+	# This is not in a BEGIN clause (even though it consumes all input),
+	# so that the code generator can insert the first FILENAME.
+	codegen_begin()
+
+	nexttoken()
+	while (Token != "") {
+		expect(defconst() || deftype())
+		expect(accept(";"))
+	}
+}