diffplug / diffplug/dormouse

dor send --stdin interprets backslash escapes in piped bytes

Ouverte
#355 0 commentaires 0 réactions 0 personnes assignées Voir sur GitHub
Langage dominant
TypeScript
Étoiles
5
Forks
0
Merge moyen
13 h 46 min
PR mergées (30 j)
205

Description

## `dor send --stdin` interprets backslash escapes in piped bytes, corrupting the documented use case

`dor send --stdin` reads standard input and forwards it as a **text** input:

```ts
if (flags.stdin === true) {
if (!readStdin) return { ok: false, message: 'stdin is not available' };
return { ok: true, value: [{ kind: 'text', text: await readStdin() }] };
}
```
([send.ts:183-186](https://github.com/diffplug/dormouse/blob/a1782f19b7d8363997eb37c440d1cbc8dfe13829/dor/src/commands/send.ts#L183-L186))

Text inputs then run through `interpretTextEscapes` unless `--raw` is set:

```ts
input += raw ? item.text : interpretTextEscapes(item.text);
```
([send.ts:227-228](https://github.com/diffplug/dormouse/blob/a1782f19b7d8363997eb37c440d1cbc8dfe13829/dor/src/commands/send.ts#L227-L228), converting `\n \r \t \\` at [send.ts:238-253](https://github.com/diffplug/dormouse/blob/a1782f19b7d8363997eb37c440d1cbc8dfe13829/dor/src/commands/send.ts#L238-L253))

### Why this is a footgun for `--stdin`

The flagship stdin example in the help is:

```
cat script.sh | dor send surface:3 --stdin
```
([send.ts:115](https://github.com/diffplug/dormouse/blob/a1782f19b7d8363997eb37c440d1cbc8dfe13829/dor/src/commands/send.ts#L115))

Bytes arriving on stdin are already literal — they are not a shell-authored string where a two-character `\t` stands in for a tab. Shell scripts routinely contain literal backslash sequences (`printf 'a\tb'`, `sed 's/\n/ /'`, `grep -P '\d'`, Windows paths with `\\`). Piping such a file through `--stdin` silently rewrites every `\n`/`\r`/`\t`/`\\`, so the text typed into the target terminal is not the file's contents.

**Repro:** `printf 'printf "a\\tb\\n"\n' | dor send surface:3 --stdin` types a real TAB and newline into the middle of the line instead of the literal `\t`/`\n` the script source contains.

The escape interpretation is desirable for `--text "echo hi\nthere"` (a human types escapes on the command line), but for `--stdin` the input is already-real bytes.

### Design question (why an issue, not a drive-by PR)

The current behavior is *consistent with* the documented contract — the help says "Text input interprets backslash escapes … unless `--raw` is set" ([send.ts:97](https://github.com/diffplug/dormouse/blob/a1782f19b7d8363997eb37c440d1cbc8dfe13829/dor/src/commands/send.ts#L97)), and `--stdin` is documented as text input — so flipping the default is a contract change that needs a maintainer call. Options:

1. **Make `--stdin` raw by default** — treat piped bytes as literal; keep `--text` interpreting escapes. Most aligned with the `cat script.sh | …` example. Would need an opt-in flag if anyone wants escape interpretation on stdin.
2. **Keep current behavior, document it** — call out at the `--stdin` help that it interprets escapes and that `--raw` is needed for literal file contents.

I lean toward option 1, but it changes documented behavior, so I'm leaving the call to a maintainer. Happy to open the PR (including a regression test that pipes a script containing `\t` and asserts the bytes are preserved) once a direction is chosen.

Surfaced by the nightly code-quality survey.

Guide de contribution

Aucun guide de contribution indexé pour ce dépôt

Piste de recherche

Start in dor/src/commands/send.ts at the stdin handling and text escape conversion described in the issue. Confirm the current behavior with the documented piping example, then wait for the maintainer to choose between raw stdin by default and documenting the existing behavior. Done means the chosen contract is implemented or documented, with a regression test preserving the relevant bytes.

Rédigé par le modèle d'indexation à partir du texte de l'issue.

Évaluation

Stack technique
typescript
Domaine
cli
Type d'issue
Bug
Difficulté
3/5
Temps estimé
1-2 jours
Activité
Calme
Clarté
Plutôt claire
Accessibilité débutants
50/100

Recevez les nouvelles issues par e-mail

Un résumé court des issues GitHub adaptées aux débutants.