Přemysl Eric Janouch
fb291b6def
All checks were successful
Alpine 3.20 Success
The new filter comes with these enhancements: - Processing is rune-wise rather than byte-wise; it assumes UTF-8 input and single-cell wide characters, but this condition should be /usually/ satisfied. - Unprocessed control characters are escaped, `cat -v` style. - A lot of escape sequences is at least recognised, if not processed. - Rudimentary preparation for efficient dynamic updates of task views, through Javascript. We make terminal resets and screen clearing commands flush all output and assume that the terminal has a new origin for any later positioning commands. This appears to work well enough with GRUB, at least. The filter is now exposed through a command line option.
370 lines
8.2 KiB
Go
370 lines
8.2 KiB
Go
package main
|
|
|
|
import (
|
|
"bytes"
|
|
"io"
|
|
"log"
|
|
"os"
|
|
"strconv"
|
|
"strings"
|
|
"sync"
|
|
"unicode/utf8"
|
|
)
|
|
|
|
type terminalLine struct {
|
|
// For simplicity, we assume that all runes take up one cell,
|
|
// including TAB and non-spacing ones.
|
|
// The next step would be grouping non-spacing characters,
|
|
// in particular Unicode modifier letters, with their base.
|
|
columns []rune
|
|
|
|
// updateGroup is the topmost line that has changed since this line
|
|
// has appeared, for the purpose of update tracking.
|
|
updateGroup int
|
|
}
|
|
|
|
// terminalWriter does a best-effort approximation of an infinite-size
|
|
// virtual terminal.
|
|
type terminalWriter struct {
|
|
sync.Mutex
|
|
Tee io.WriteCloser
|
|
lines []terminalLine
|
|
|
|
// Zero-based coordinates within lines.
|
|
column, line int
|
|
|
|
// lineTop is used as the base for positioning commands.
|
|
lineTop int
|
|
|
|
written int
|
|
byteBuffer []byte
|
|
runeBuffer []rune
|
|
}
|
|
|
|
func (tw *terminalWriter) log(format string, v ...interface{}) {
|
|
if os.Getenv("ACID_TERMINAL_DEBUG") != "" {
|
|
log.Printf("terminal: "+format+"\n", v...)
|
|
}
|
|
}
|
|
|
|
func (tw *terminalWriter) Serialize(top int) []byte {
|
|
var b bytes.Buffer
|
|
for i := top; i < len(tw.lines); i++ {
|
|
b.WriteString(string(tw.lines[i].columns))
|
|
b.WriteByte('\n')
|
|
}
|
|
return b.Bytes()
|
|
}
|
|
|
|
func (tw *terminalWriter) Write(p []byte) (written int, err error) {
|
|
tw.Lock()
|
|
defer tw.Unlock()
|
|
|
|
// TODO(p): Rather use io.MultiWriter?
|
|
// Though I'm not sure what to do about closing (FD leaks).
|
|
// Eventually, any handles would be garbage collected in any case.
|
|
if tw.Tee != nil {
|
|
tw.Tee.Write(p)
|
|
}
|
|
|
|
// Enough is enough, writing too much is highly suspicious.
|
|
ok, remaining := true, 64<<20-tw.written
|
|
if remaining < 0 {
|
|
ok, p = false, nil
|
|
} else if remaining < len(p) {
|
|
ok, p = false, p[:remaining]
|
|
}
|
|
tw.written += len(p)
|
|
|
|
// By now, more or less everything should run in UTF-8.
|
|
//
|
|
// This might have better performance with a ring buffer,
|
|
// so as to avoid reallocations.
|
|
b := append(tw.byteBuffer, p...)
|
|
if !ok {
|
|
b = append(b, "\nToo much terminal output\n"...)
|
|
}
|
|
for utf8.FullRune(b) {
|
|
r, len := utf8.DecodeRune(b)
|
|
b, tw.runeBuffer = b[len:], append(tw.runeBuffer, r)
|
|
}
|
|
tw.byteBuffer = b
|
|
for tw.processRunes() {
|
|
}
|
|
return len(p), nil
|
|
}
|
|
|
|
func (tw *terminalWriter) processPrint(r rune) {
|
|
// Extend the buffer vertically.
|
|
for len(tw.lines) <= tw.line {
|
|
tw.lines = append(tw.lines,
|
|
terminalLine{updateGroup: len(tw.lines)})
|
|
}
|
|
|
|
// Refresh update trackers, if necessary.
|
|
if tw.lines[len(tw.lines)-1].updateGroup > tw.line {
|
|
for i := tw.line; i < len(tw.lines); i++ {
|
|
tw.lines[i].updateGroup = tw.line
|
|
}
|
|
}
|
|
|
|
// Emulate `cat -v` for C0 characters.
|
|
seq := make([]rune, 0, 2)
|
|
if r < 32 && r != '\t' {
|
|
seq = append(seq, '^', 64+r)
|
|
} else {
|
|
seq = append(seq, r)
|
|
}
|
|
|
|
// Extend the line horizontally and write the rune.
|
|
for _, r := range seq {
|
|
line := &tw.lines[tw.line]
|
|
for len(line.columns) <= tw.column {
|
|
line.columns = append(line.columns, ' ')
|
|
}
|
|
|
|
line.columns[tw.column] = r
|
|
tw.column++
|
|
}
|
|
}
|
|
|
|
func (tw *terminalWriter) processFlush() {
|
|
tw.column = 0
|
|
tw.line = len(tw.lines)
|
|
tw.lineTop = tw.line
|
|
}
|
|
|
|
func (tw *terminalWriter) processParsedCSI(
|
|
private rune, param, intermediate []rune, final rune) bool {
|
|
var params []int
|
|
if len(param) > 0 {
|
|
for _, p := range strings.Split(string(param), ";") {
|
|
i, _ := strconv.Atoi(p)
|
|
params = append(params, i)
|
|
}
|
|
}
|
|
|
|
if private == '?' && len(intermediate) == 0 &&
|
|
(final == 'h' || final == 'l') {
|
|
for _, p := range params {
|
|
// 25 (DECTCEM): There is no cursor to show or hide.
|
|
// 7 (DECAWM): We cannot wrap, we're infinite.
|
|
if !(p == 25 || (p == 7 && final == 'l')) {
|
|
return false
|
|
}
|
|
}
|
|
return true
|
|
}
|
|
if private != 0 || len(intermediate) > 0 {
|
|
return false
|
|
}
|
|
|
|
switch {
|
|
case final == 'C': // Cursor Forward
|
|
if len(params) == 0 {
|
|
tw.column++
|
|
} else if len(params) >= 1 {
|
|
tw.column += params[0]
|
|
}
|
|
return true
|
|
case final == 'D': // Cursor Backward
|
|
if len(params) == 0 {
|
|
tw.column--
|
|
} else if len(params) >= 1 {
|
|
tw.column -= params[0]
|
|
}
|
|
if tw.column < 0 {
|
|
tw.column = 0
|
|
}
|
|
return true
|
|
case final == 'E': // Cursor Next Line
|
|
if len(params) == 0 {
|
|
tw.line++
|
|
} else if len(params) >= 1 {
|
|
tw.line += params[0]
|
|
}
|
|
tw.column = 0
|
|
return true
|
|
case final == 'F': // Cursor Preceding Line
|
|
if len(params) == 0 {
|
|
tw.line--
|
|
} else if len(params) >= 1 {
|
|
tw.line -= params[0]
|
|
}
|
|
if tw.line < tw.lineTop {
|
|
tw.line = tw.lineTop
|
|
}
|
|
tw.column = 0
|
|
return true
|
|
case final == 'H': // Cursor Position
|
|
if len(params) == 0 {
|
|
tw.line = tw.lineTop
|
|
tw.column = 0
|
|
} else if len(params) >= 2 && params[0] != 0 && params[1] != 0 {
|
|
tw.line = tw.lineTop + params[0] - 1
|
|
tw.column = params[1] - 1
|
|
} else {
|
|
return false
|
|
}
|
|
return true
|
|
case final == 'J': // Erase in Display
|
|
if len(params) == 0 || params[0] == 0 || params[0] == 2 {
|
|
// We're not going to erase anything, thank you very much.
|
|
tw.processFlush()
|
|
} else {
|
|
return false
|
|
}
|
|
return true
|
|
case final == 'K': // Erase in Line
|
|
if tw.line >= len(tw.lines) {
|
|
return true
|
|
}
|
|
line := &tw.lines[tw.line]
|
|
if len(params) == 0 || params[0] == 0 {
|
|
if len(line.columns) > tw.column {
|
|
line.columns = line.columns[:tw.column]
|
|
}
|
|
} else if params[0] == 1 {
|
|
for i := 0; i < tw.column; i++ {
|
|
line.columns[i] = ' '
|
|
}
|
|
} else if params[0] == 2 {
|
|
line.columns = nil
|
|
} else {
|
|
return false
|
|
}
|
|
return true
|
|
case final == 'm':
|
|
// Straight up ignoring all attributes, at least for now.
|
|
return true
|
|
}
|
|
return false
|
|
}
|
|
|
|
func (tw *terminalWriter) processCSI(rb []rune) ([]rune, bool) {
|
|
if len(rb) < 3 {
|
|
return nil, true
|
|
}
|
|
|
|
i, private, param, intermediate := 2, rune(0), []rune{}, []rune{}
|
|
if rb[i] >= 0x3C && rb[i] <= 0x3F {
|
|
private = rb[i]
|
|
i++
|
|
}
|
|
for i < len(rb) && ((rb[i] >= '0' && rb[i] <= '9') || rb[i] == ';') {
|
|
param = append(param, rb[i])
|
|
i++
|
|
}
|
|
for i < len(rb) && rb[i] >= 0x20 && rb[i] <= 0x2F {
|
|
intermediate = append(intermediate, rb[i])
|
|
i++
|
|
}
|
|
if i == len(rb) {
|
|
return nil, true
|
|
}
|
|
if rb[i] < 0x40 || rb[i] > 0x7E {
|
|
return rb, false
|
|
}
|
|
if !tw.processParsedCSI(private, param, intermediate, rb[i]) {
|
|
tw.log("unhandled CSI %s", string(rb[2:i+1]))
|
|
return rb, false
|
|
}
|
|
return rb[i+1:], true
|
|
}
|
|
|
|
func (tw *terminalWriter) processEscape(rb []rune) ([]rune, bool) {
|
|
if len(rb) < 2 {
|
|
return nil, true
|
|
}
|
|
|
|
// Very roughly following https://vt100.net/emu/dec_ansi_parser
|
|
// but being a bit stricter.
|
|
switch r := rb[1]; {
|
|
case r == '[':
|
|
return tw.processCSI(rb)
|
|
case r == ']':
|
|
// TODO(p): Skip this properly, once we actually hit it.
|
|
tw.log("unhandled OSC")
|
|
return rb, false
|
|
case r == 'P':
|
|
// TODO(p): Skip this properly, once we actually hit it.
|
|
tw.log("unhandled DCS")
|
|
return rb, false
|
|
|
|
// Only handling sequences we've seen bother us in real life.
|
|
case r == 'c':
|
|
// Full reset, use this to flush all output.
|
|
tw.processFlush()
|
|
return rb[2:], true
|
|
case r == 'M':
|
|
tw.line--
|
|
return rb[2:], true
|
|
|
|
case (r >= 0x30 && r <= 0x4F) || (r >= 0x51 && r <= 0x57) ||
|
|
r == 0x59 || r == 0x5A || r == 0x5C || (r >= 0x60 && r <= 0x7E):
|
|
// → esc_dispatch
|
|
tw.log("unhandled ESC %c", r)
|
|
return rb, false
|
|
//return rb[2:], true
|
|
case r >= 0x20 && r <= 0x2F:
|
|
// → escape intermediate
|
|
i := 2
|
|
for i < len(rb) && rb[i] >= 0x20 && rb[i] <= 0x2F {
|
|
i++
|
|
}
|
|
if i == len(rb) {
|
|
return nil, true
|
|
}
|
|
if rb[i] < 0x30 || rb[i] > 0x7E {
|
|
return rb, false
|
|
}
|
|
// → esc_dispatch
|
|
tw.log("unhandled ESC %s", string(rb[1:i+1]))
|
|
return rb, false
|
|
//return rb[i+1:], true
|
|
default:
|
|
// Note that Debian 12 has been seen to produce ESC<U+2026>
|
|
// and such due to some very blind string processing.
|
|
return rb, false
|
|
}
|
|
}
|
|
|
|
func (tw *terminalWriter) processRunes() bool {
|
|
rb := tw.runeBuffer
|
|
if len(rb) == 0 {
|
|
return false
|
|
}
|
|
|
|
switch rb[0] {
|
|
case '\a':
|
|
// Ding dong!
|
|
case '\b':
|
|
if tw.column > 0 {
|
|
tw.column--
|
|
}
|
|
case '\n', '\v':
|
|
tw.line++
|
|
|
|
// Forced ONLCR flag, because that's what most shell output expects.
|
|
fallthrough
|
|
case '\r':
|
|
tw.column = 0
|
|
|
|
case '\x1b':
|
|
var ok bool
|
|
if rb, ok = tw.processEscape(rb); rb == nil {
|
|
return false
|
|
} else if ok {
|
|
tw.runeBuffer = rb
|
|
return true
|
|
}
|
|
|
|
// Unsuccessful parses get printed for later inspection.
|
|
fallthrough
|
|
default:
|
|
tw.processPrint(rb[0])
|
|
}
|
|
tw.runeBuffer = rb[1:]
|
|
return true
|
|
}
|