[Bug]: Non-ASCII input inside an unfinished CSI loses visible recovery text from history
Los mantenedores suelen responder en 1 día
Nadie ha tomado este issue todavía.
Evaluación
- Dificultad
- 3/5
- Tiempo estimado
- 1-2 días
- Aptitud para principiantes
- 45/100
- Tipo de issue
- Error
- Claridad
- Bastante claro
- Estado de actividad
- Activo
- Stack tecnológico
- typescript, wasm
- Área
- backend
Línea de trabajo
Start in apps/server/src/terminal/Manager.ts, at sanitizeTerminalHistoryChunk and isCsiFinalByte, which decide what history keeps after an unfinished CSI. The attached repro-csi-unicode.mjs script shows the failing case: the input A, ESC, [, U+754C, n, B keeps only AB in history while the live engine renders A�nB. Done means replaying that input matches the live output, and the valid A CSI 31m B control still keeps its bytes.
Escrito por el modelo de indexación a partir del texto del issue.
Descripción
Summary
The history scanner keeps treating non-ASCII input as part of an unfinished control sequence after the bundled terminal has already recovered and resumed printing text. A later status-report final byte makes the scanner discard that visible text. A deterministic scanner and engine probe reproduced output present live but absent on replay.
Before submitting
- Related issues: searched open and closed issues and PRs; related history is listed below, with no exact current duplicate found.
- Removed credentials, tokens, private paths, and sensitive data.
Area
apps/server
Observed behavior
For A ESC [ 界 n B (UTF-8 hex 411b5be7958c6e42), the bundled Ghostty engine renders A�nB and generates no reply. The history scanner instead keeps AB, so replay loses both the recovery character and the ordinary n that the live engine displayed. The valid A CSI 31m B control retains its bytes and renders AB both live and on replay.
Reproduction
- Use a checkout of the baseline below. Save the script as
repro-csi-unicode.mjsoutside the checkout and runnode repro-csi-unicode.mjs /path/to/checkout. - The script runs an unfinished CSI followed by U+754C and printable sentinels, plus a valid color-sequence control through the unchanged production definitions extracted into an in-memory module, then compares live bytes and replayed history in the bundled Ghostty WASM. It needs Node with
node:module.stripTypeScriptTypes, Git, and the checked-in WASM artifacts; installed application dependencies are unnecessary.
The result is deterministic. Every two-chunk scanner split and every raw-byte split of the live engine input produced the same result, including all three bytes of U+754C. The script creates a disposable directory, points HOME, TMPDIR, and T3CODE_HOME into it, removes it on exit, and leaves the checkout untouched. Exit 1 means the defect was reproduced, exit 0 means expected behavior, and exit 2 means setup or a control failed. The recorded run exited 1 after all controls passed. This is a unit-style scanner/engine reproduction, not a running application or native-device test.
Reproduction script: repro-csi-unicode.mjs
import fs from 'node:fs';
import os from 'node:os';
import {execFileSync} from 'node:child_process';
import path from 'node:path';
import { stripTypeScriptTypes } from 'node:module';
const scratch=fs.mkdtempSync(path.join(os.tmpdir(),'terminal-history-repro-'));
let status=2;
try {
for(const name of ['HOME','TMPDIR','T3CODE_HOME']){const dir=path.join(scratch,name.toLowerCase());fs.mkdirSync(dir);process.env[name]=dir;}
if(!process.argv[2])throw Error('Pass the checkout path as the only argument');
const root = path.resolve(process.argv[2]);
console.log('Git', execFileSync('git',['--version'],{encoding:'utf8'}).trim());
console.log('T3 commit',execFileSync('git',['-C',root,'rev-parse','HEAD'],{encoding:'utf8'}).trim());
console.log('Node',process.version);
const source=fs.readFileSync(path.join(root,'apps/server/src/terminal/Manager.ts'),'utf8');
const scanSource=source.slice(source.indexOf('function isCsiFinalByte('),source.indexOf('\nfunction ',source.indexOf('function sanitizeTerminalHistoryChunk(')+10));
const boundedSource=source.slice(source.indexOf('export class BoundedTerminalHistory'),source.indexOf('// Pop beyond Ghostty'));
const contract=fs.readFileSync(path.join(root,'packages/contracts/src/terminal.ts'),'utf8');
const tailSource=contract.slice(contract.indexOf('const historyEncoder'),contract.indexOf('const TrimmedNonEmptyStringSchema'));
const clientSource=fs.readFileSync(path.join(root,'packages/client-runtime/src/state/terminalOutput.ts'),'utf8').replace('import { retainedTerminalHistory } from "@t3tools/contracts";','');
const extracted=stripTypeScriptTypes('const DEFAULT_HISTORY_BYTE_LIMIT=8*1024*1024; const MAX_HISTORY_CHUNK_LENGTH=16*1024;\n'+boundedSource+tailSource+scanSource+'\n'+clientSource+'\nexport {sanitizeTerminalHistoryChunk};',{mode:'strip'});
const api=await import('data:text/javascript;base64,'+Buffer.from(extracted).toString('base64'));
const scan=(chunks)=>{let kept='',pending=''; for(const s of chunks){const r=api.sanitizeTerminalHistoryChunk(pending,s);kept+=r.visibleText;pending=r.pendingControlSequence;}return {kept,pending};};
const wasm=await WebAssembly.instantiate(fs.readFileSync(path.join(root,'apps/web/src/terminal/ghostty/vendor/ghostty-vt.wasm')),{env:{log:()=>{}}});
const e=wasm.instance.exports;
const call=(n,...a)=>e[n](...a);
const mem=e.memory;
const alloc=(n)=>call('ghostty_wasm_alloc_u8_array',n);
const free=(p,n)=>call('ghostty_wasm_free_u8_array',p,n);
const view=()=>new DataView(mem.buffer);
const bytes=(p,n)=>new Uint8Array(mem.buffer,p,n);
const decode=(p,n)=>new TextDecoder().decode(bytes(p,n));
const check=(r)=>{if(r!==0)throw Error('Ghostty result '+r);};
let replies='';
const trampoline=await WebAssembly.instantiate(fs.readFileSync(path.join(root,'apps/web/src/terminal/ghostty/vendor/ghostty-write-pty.wasm')),{env:{t3_write_pty:(_t,_u,p,n)=>{replies+=decode(p,n);}}});
const table=e.__indirect_function_table, callback=table.length;
table.grow(1,trampoline.instance.exports.ghostty_write_pty);
const revSlot=alloc(8);check(call('ghostty_build_info',10,revSlot));console.log('Ghostty',decode(view().getUint32(revSlot,true),view().getUint32(revSlot+4,true)));free(revSlot,8);
function terminal(chunks){
const opts=alloc(8),slot=alloc(4);bytes(opts,8).fill(0);view().setUint16(opts,80,true);view().setUint16(opts+2,24,true);check(call('ghostty_terminal_new',0,slot,opts));const t=view().getUint32(slot,true);free(opts,8);free(slot,4);
call('ghostty_terminal_set',t,0,1);call('ghostty_terminal_set',t,1,callback);replies='';
for(const s of chunks){const b=typeof s==='string'?Buffer.from(s):s;if(!b.length)continue;const p=alloc(b.length);bytes(p,b.length).set(b);call('ghostty_terminal_vt_write',t,p,b.length);free(p,b.length);}
const pos=alloc(8);check(call('ghostty_terminal_get',t,3,pos));const x=view().getUint16(pos,true);check(call('ghostty_terminal_get',t,4,pos));const y=view().getUint16(pos,true);check(call('ghostty_terminal_get',t,12,pos));const title=decode(view().getUint32(pos,true),view().getUint32(pos+4,true));free(pos,8);
// WASM C ABI passes the formatter options struct by reference.
const fopts=alloc(24),fslot=alloc(4),output=alloc(8192),written=alloc(4);bytes(fopts,24).fill(0);view().setUint32(fopts,24,true);bytes(fopts+8,2).fill(1);check(call('ghostty_formatter_terminal_new',0,fslot,t,fopts));const f=view().getUint32(fslot,true);check(call('ghostty_formatter_format_buf',f,output,8192,written));const text=decode(output,view().getUint32(written,true));call('ghostty_formatter_free',f);free(fopts,24);free(fslot,4);free(output,8192);free(written,4);
const result={reply:replies,text,title,x,y};call('ghostty_terminal_free',t);return result;
}
const assert=(condition,message)=>{if(!condition)throw Error('Setup/control failed: '+message);};
const defects=[];
function allSplits(input){
const expected=scan([input]),baseline=terminal([input]);
for(let i=0;i<=input.length;i++)assert(JSON.stringify(scan([input.slice(0,i),input.slice(i)]))===JSON.stringify(expected),'scanner split '+i);
const raw=Buffer.from(input);
for(let i=0;i<=raw.length;i++)assert(JSON.stringify(terminal([raw.subarray(0,i),raw.subarray(i)]))===JSON.stringify(baseline),'Ghostty byte split '+i);
return {kept:expected.kept,pending:expected.pending,live:baseline,replay:terminal([expected.kept])};
}
const control=allSplits('A\x1b[31mB');
assert(control.kept==='A\x1b[31mB'&&control.live.text==='AB'&&control.replay.text==='AB','valid CSI style');
const input='A\x1b[界nB',result=allSplits(input);
assert(result.live.text==='A\ufffdnB'&&result.live.reply===''&&!result.pending,'live parser recovery');
console.log(JSON.stringify({input,utf8Hex:Buffer.from(input).toString('hex'),expectedReplayText:result.live.text,...result}));
if(result.replay.text!==result.live.text)defects.push('visible recovery text lost');
status=defects.length?1:0;
}catch(error){console.log('Setup error:',error.message.replaceAll(scratch,'<scratch>'));status=2;}
finally{fs.rmSync(scratch,{recursive:true,force:true});}
console.log(status===0?'RESULT: expected behavior':status===1?'RESULT: defect reproduced':'RESULT: inconclusive');
process.exit(status);
Expected behavior
Filtering history should preserve the visible output resulting from the bundled parser’s recovery, rather than swallowing printable text as a status query that the engine never dispatched. ECMA-48 section 5.4 gives separate byte ranges for CSI parameters, intermediates and final bytes; a Unicode code point outside those ranges cannot be blindly treated as another parameter. Error recovery itself is implementation-dependent, so the expected display here is established by the pinned Ghostty engine, not a claim that the standard prescribes its replacement character.
Impact and scope
Impact: P2 - A process producing a truncated CSI followed by non-ASCII text can leave saved scrollback missing text visible live. This is bounded to malformed or interrupted control streams. The report does not ask the terminal to preserve the original CJK character beyond what its live parser displays.
Form impact: Minor bug or occasional failure. Workaround: NOT_ENOUGH_INFO — no general end-user workaround was demonstrated for arbitrary process output.
Version
T3 Code commit 59e64eef881632a23f89c4138bcf37f23d92bf93.
Runtime and platform
Node v26.12.1; Git 2.56.0; macOS arm64 (Darwin 27.2.0). Vendored libghostty-vt WASM build info reports 9f62873bf195e4d8a762d768a1405a5f2f7b1697, matching native/libghostty-vt/VERSION. No browser, application server, real PTY, or native mobile device was started.
Interface
Library: server terminal-history scanner and bundled Ghostty C ABI.
Earliest failure stage
Input parsing or validation.
Relevant output or evidence
Actual behavior
The following display uses named markers for ESC, NUL, BEL, DEL, and U+009C; reproduction inputs in the script above retain executable escape notation.
Git git version 2.56.0
T3 commit 59e64eef881632a23f89c4138bcf37f23d92bf93
Node v26.12.1
Ghostty 9f62873bf195e4d8a762d768a1405a5f2f7b1697
{"input":"A<ESC>[界nB","utf8Hex":"411b5be7958c6e42","expectedReplayText":"A�nB","kept":"AB","pending":"","live":{"reply":"","text":"A�nB","title":"","x":4,"y":0},"replay":{"reply":"","text":"AB","title":"","x":2,"y":0}}
RESULT: defect reproduced
Evidence
- Expected source: the documented restoration/retention contract and pinned parser behavior described above.
- Failure source: the script output above, running the unchanged production definitions and checked-in WASM.
- Evidence provenance: observed.
- Local verification: reproduced.
- Reproduction completeness: complete for the scanner/retention and engine boundary; no full-client behavior is claimed.
apps/server/src/terminal/Manager.ts#L1144-L1328 contains the scanner exercised by the probe. apps/server/src/terminal/Manager.ts#L2321-L2331 passes each PTY output chunk to it and appends its kept bytes to session history.
Restoration check
The script should exit 0: replay preserves the same visible recovery text as live parsing, and the valid color-sequence control continues to retain its bytes and display correctly.
Additional context
References
- https://github.com/pingdotgg/t3code/pull/29604 — fixed ESC recovery inside control strings; this report is a CSI interrupted by non-ASCII input, without an OSC or DCS.
- https://github.com/pingdotgg/t3code/pull/29649 — fixed ground-state C1 Unicode introducers; this example uses U+754C inside an already-started CSI.
AI-assisted report; verified with the script and tools listed above. No source changes, fixes, or PRs are included.
- Lenguaje dominante
- TypeScript
- Estrellas
- 24.8k
- Forks
- 6.4k
- Merge medio
- 59 min
- PR fusionados (30 d)
- 418
Preparar el entorno
Inicia el contenedor de desarrollo del proyecto en tu navegador, con tu propia cuenta de GitHub.
- Sin Dockerfile ni archivo de Docker Compose
- Tiene una plantilla de pull request
- Leer la guía de contribución
Primeros pasos
- Lee el issue completo y luego la guía de contribución del proyecto.
- Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
- Haz un fork del repositorio y trabaja en una rama.
- Abre un pull request que haga referencia al número del issue.
Más de pingdotgg/t3code
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 66/100
Los mantenedores suelen responder en 1 día
-
via-triage
Dificultad 2/5 1-3 horas Aptitud para principiantes 82/100
Los mantenedores suelen responder en 1 día
-
[Bug]: MCP App inline frames never go below 80px, so compact apps get blank space under themAbierto
Dificultad 2/5 Medio día Aptitud para principiantes 68/100
Los mantenedores suelen responder en 1 día
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 76/100
Los mantenedores suelen responder en 1 día
-
[Bug]: Settings inheritance popover shows raw slugs for branch naming, default permissions and response streamingPosiblemente ocupada @juliusmarminge la tomó hace 1 día. Abiertovouch:denounced
Dificultad 2/5 1-3 horas Aptitud para principiantes 84/100
Los mantenedores suelen responder en 1 día
Todos los issues de pingdotgg/t3code
Issues similares
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 72/100
nikoheikkila/nikoheikkila.fi#1131 ·
Los mantenedores suelen responder en 1 día
-
Resources: Update work of urumqiPosiblemente ocupada Un pull request vinculado a esta issue está abierto o ya se fusionó. Abierto
Dificultad 2/5 1-3 horas Aptitud para principiantes 65/100
railmapgen/rmp-gallery#4176 ·
Los mantenedores suelen responder en 2 días
-
GraphQL custom headers duplicate differently cased default headersPosiblemente ocupada @yxr-2025 la tomó hoy. Abierto
Dificultad 2/5 1-3 horas Aptitud para principiantes 70/100
elie222/rakazo#4625 · 1 comentario ·
Los mantenedores suelen responder en 1 día
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 76/100
Los mantenedores suelen responder en 1 día
-
[New Project]: PacaAbierto
Dificultad 2/5 1-3 horas Aptitud para principiantes 65/100
github/forgoodfirstissue#574 ·