-
Notifications
You must be signed in to change notification settings - Fork 91
fix: skip stdin read when positional message provided #935
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Open
sahrizvi
wants to merge
2
commits into
main
Choose a base branch
from
fix/run-stdin-wedge-934
base: main
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
Open
Changes from all commits
Commits
Show all changes
2 commits
Select commit
Hold shift + click to select a range
File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,137 @@ | ||
| import fs from "fs" | ||
| import type { Stats } from "fs" | ||
|
|
||
| const FIRST_BYTE_TIMEOUT_MS = 100 | ||
|
|
||
| type Stat = Pick<Stats, "isFIFO" | "isFile" | "isSocket"> | ||
|
|
||
| export interface ReadStdinDeps { | ||
| isTTY?: boolean | ||
| fstat?: () => Stat | ||
| // Returns "" if no first byte arrives within timeoutMs; otherwise drains | ||
| // stdin to EOF and returns the full content. Default implementation uses | ||
| // process.stdin events so that a wedged-after-first-byte case still has a | ||
| // well-defined behavior (waits for `end`); callers can inject a faster | ||
| // implementation in tests. | ||
| readStdin?: (timeoutMs: number) => Promise<string> | ||
| timeoutMs?: number | ||
| } | ||
|
|
||
| // Read piped/redirected stdin without wedging on an inherited-but-idle fd. | ||
| // | ||
| // The failure mode this guards against: subprocess callers (Claude Code's | ||
| // Bash tool, Python `subprocess.run(..., stdin=None)`, CI, plugin hosts) | ||
| // leave stdin attached to a parent pipe that is never written to and never | ||
| // closed. A blind `Bun.stdin.text()` waits forever for an EOF that never | ||
| // arrives. | ||
| // | ||
| // Strategy — two gates: | ||
| // | ||
| // 1. fstat gate: only FIFOs (pipes), regular files (redirects), and | ||
| // sockets (process supervisors, socket activation, `nc -l`) can carry | ||
| // real input. TTYs and character devices (e.g. `< /dev/null`) skip. | ||
| // | ||
| // 2. First-byte timeout: instead of bounding the whole-stream drain, we | ||
| // wait up to `timeoutMs` for the first readable byte. If no byte | ||
| // arrives in that window, we treat stdin as inherited-idle and skip. | ||
| // If a byte arrives, we drain to EOF without further deadline — so a | ||
| // slow producer that takes >100ms total but flushes its first chunk | ||
| // within the window is not truncated. This avoids the two pitfalls of | ||
| // a whole-stream race: (a) the orphaned `Bun.stdin.text()` continuing | ||
| // to hold fd 0 open after the loser is abandoned, and (b) silent | ||
| // mid-stream truncation of legitimate slow / large producers. | ||
| export async function readStdinIfAvailable(deps: ReadStdinDeps = {}): Promise<string> { | ||
| const isTTY = deps.isTTY ?? Boolean(process.stdin.isTTY) | ||
| const fstat = deps.fstat ?? (() => fs.fstatSync(0) as Stat) | ||
| const readStdin = deps.readStdin ?? defaultReadStdin | ||
| const timeoutMs = deps.timeoutMs ?? FIRST_BYTE_TIMEOUT_MS | ||
|
|
||
| if (isTTY) return "" | ||
|
|
||
| try { | ||
| const stat = fstat() | ||
| if (!stat.isFIFO() && !stat.isFile() && !stat.isSocket()) return "" | ||
| } catch { | ||
| return "" | ||
| } | ||
|
|
||
| return readStdin(timeoutMs) | ||
| } | ||
|
|
||
| // Compose the final prompt from a positional message and stdin input. | ||
| // Extracted as a pure function so the regression case from PR #935 | ||
| // (`echo ctx | run "prompt"` must concatenate, not silently drop ctx) can | ||
| // be unit-tested without spawning the full run command. | ||
| export function assembleStdinMessage(positional: string, stdinInput: string): string { | ||
| if (stdinInput.trim().length === 0) return positional | ||
| if (positional.length === 0) return stdinInput | ||
| return positional + "\n" + stdinInput | ||
| } | ||
|
|
||
| // Default implementation of the first-byte race over `process.stdin`. | ||
| // | ||
| // Why not `Bun.stdin.text()`: that reads the entire stream as a single | ||
| // uncancellable Promise. If we race it against a timer and the timer wins, | ||
| // the read still holds fd 0 open until the producer eventually closes, | ||
| // blocking process exit (the original wedge moved to teardown). | ||
| // | ||
| // Using `process.stdin` event listeners lets us: | ||
| // - bind the timeout to "first byte" rather than "full drain", so slow | ||
| // producers and large payloads aren't truncated; | ||
| // - cleanly remove our listeners and `unref` the stream on the no-data | ||
| // path, so an inherited-open fd doesn't pin the event loop. | ||
| function defaultReadStdin(timeoutMs: number): Promise<string> { | ||
| return new Promise<string>((resolve) => { | ||
| const stdin = process.stdin | ||
| const chunks: Buffer[] = [] | ||
| let firstByteReceived = false | ||
| let firstByteTimer: ReturnType<typeof setTimeout> | undefined | ||
| let settled = false | ||
|
|
||
| const cleanup = () => { | ||
| stdin.off("data", onData) | ||
| stdin.off("end", onEnd) | ||
| stdin.off("error", onError) | ||
| if (firstByteTimer) clearTimeout(firstByteTimer) | ||
| try { | ||
| stdin.pause() | ||
| } catch {} | ||
| try { | ||
| stdin.unref?.() | ||
| } catch {} | ||
| } | ||
|
|
||
| const settle = (result: string) => { | ||
| if (settled) return | ||
| settled = true | ||
| cleanup() | ||
| resolve(result) | ||
| } | ||
|
|
||
| const onData = (chunk: Buffer) => { | ||
| if (!firstByteReceived) { | ||
| firstByteReceived = true | ||
| if (firstByteTimer) { | ||
| clearTimeout(firstByteTimer) | ||
| firstByteTimer = undefined | ||
| } | ||
| } | ||
| chunks.push(chunk) | ||
| } | ||
| const onEnd = () => settle(Buffer.concat(chunks).toString("utf8")) | ||
| const onError = () => settle(Buffer.concat(chunks).toString("utf8")) | ||
|
|
||
| firstByteTimer = setTimeout(() => { | ||
| if (!firstByteReceived) settle("") | ||
| }, timeoutMs) | ||
|
|
||
| stdin.on("data", onData) | ||
| stdin.on("end", onEnd) | ||
| stdin.on("error", onError) | ||
| try { | ||
| stdin.resume() | ||
| } catch { | ||
| settle("") | ||
| } | ||
| }) | ||
| } | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,129 @@ | ||
| import { describe, expect, test } from "bun:test" | ||
| import path from "path" | ||
|
|
||
| // The fixture imports the real helper and prints { result, elapsed } as JSON. | ||
| // Spawning it lets us exercise the actual `process.stdin` event path against | ||
| // real fd 0 conditions — something dependency injection can't cover. | ||
| const FIXTURE = path.join(__dirname, "stdin-fixture.ts") | ||
|
|
||
| type FileSink = { write(chunk: string | Uint8Array): number; end(): Promise<number> | number } | ||
|
|
||
| async function runFixture(opts: { | ||
| writeStdin?: (sink: FileSink) => Promise<void> | ||
| killAfterMs?: number | ||
| }): Promise<{ code: number | null; result?: string; elapsed?: number; stdout: string; stderr: string }> { | ||
| const proc = Bun.spawn(["bun", "run", FIXTURE], { | ||
| stdin: "pipe", | ||
| stdout: "pipe", | ||
| stderr: "pipe", | ||
| }) | ||
|
|
||
| if (opts.writeStdin) { | ||
| await opts.writeStdin(proc.stdin as unknown as FileSink) | ||
| } | ||
|
|
||
| let killTimer: ReturnType<typeof setTimeout> | undefined | ||
| if (opts.killAfterMs) { | ||
| killTimer = setTimeout(() => proc.kill(), opts.killAfterMs) | ||
| } | ||
| const code = await proc.exited | ||
| if (killTimer) clearTimeout(killTimer) | ||
|
|
||
| const stdout = await new Response(proc.stdout).text() | ||
| const stderr = await new Response(proc.stderr).text() | ||
|
|
||
| try { | ||
| const parsed = JSON.parse(stdout) | ||
| return { code, result: parsed.result, elapsed: parsed.elapsed, stdout, stderr } | ||
| } catch { | ||
| return { code, stdout, stderr } | ||
| } | ||
| } | ||
|
|
||
| describe("readStdinIfAvailable (spawned subprocess)", () => { | ||
| // M1-regression — the canonical wedge: an inherited pipe that's never | ||
| // written to and never closed. Pre-fix this hung forever; with the | ||
| // previous Promise.race fix, the await released at 100ms but fd 0 stayed | ||
| // open until the parent closed. With the first-byte event race + unref, | ||
| // the child exits promptly. | ||
| test( | ||
| "exits promptly with empty result when stdin is an inherited-but-idle pipe", | ||
| async () => { | ||
| const { code, result, elapsed } = await runFixture({ | ||
| // Don't write — leave the pipe open and silent. Close after 1s so | ||
| // the parent's writer isn't garbage-collected; by then the child | ||
| // should have already exited via the first-byte timeout. | ||
| writeStdin: async (sink) => { | ||
| await new Promise((r) => setTimeout(r, 1000)) | ||
| try { | ||
| await sink.end() | ||
| } catch {} | ||
| }, | ||
| killAfterMs: 5000, | ||
| }) | ||
| expect(code).toBe(0) | ||
| expect(result).toBe("") | ||
| expect(elapsed).toBeLessThan(500) | ||
| }, | ||
| 10000, | ||
| ) | ||
|
|
||
| // MAJOR-regression from PR #935: `echo ctx | run "prompt"` must still | ||
| // deliver "ctx". Verified here by writing data + closing the pipe. | ||
| test( | ||
| "returns piped data when producer writes and closes", | ||
| async () => { | ||
| const { code, result } = await runFixture({ | ||
| writeStdin: async (sink) => { | ||
| sink.write("context data") | ||
| await sink.end() | ||
| }, | ||
| }) | ||
| expect(code).toBe(0) | ||
| expect(result).toBe("context data") | ||
| }, | ||
| 10000, | ||
| ) | ||
|
|
||
| // M2-regression: a producer that takes >100ms to flush first byte was | ||
| // truncated by the old Promise.race timeout. The first-byte gate must | ||
| // accept the byte once it arrives and then drain the rest without a | ||
| // deadline. | ||
| test( | ||
| "preserves data from slow producer (first byte arrives just before timeout)", | ||
| async () => { | ||
| const { code, result } = await runFixture({ | ||
| writeStdin: async (sink) => { | ||
| // Sleep close to but under the 100ms first-byte budget, then | ||
| // flush. The whole-stream-race fix would have returned "". | ||
| await new Promise((r) => setTimeout(r, 60)) | ||
| sink.write("slow ctx") | ||
| await sink.end() | ||
| }, | ||
| }) | ||
| expect(code).toBe(0) | ||
| expect(result).toBe("slow ctx") | ||
| }, | ||
| 10000, | ||
| ) | ||
|
|
||
| // Negative control: a producer slow enough to miss the first-byte window | ||
| // should yield "" (intentional cutoff, no truncation of in-flight data). | ||
| test( | ||
| "returns empty when first byte arrives after the first-byte timeout", | ||
| async () => { | ||
| const { code, result } = await runFixture({ | ||
| writeStdin: async (sink) => { | ||
| await new Promise((r) => setTimeout(r, 400)) | ||
| try { | ||
| sink.write("too late") | ||
| await sink.end() | ||
| } catch {} | ||
| }, | ||
| }) | ||
| expect(code).toBe(0) | ||
| expect(result).toBe("") | ||
| }, | ||
| 10000, | ||
| ) | ||
| }) |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,9 @@ | ||
| // Subprocess fixture used by stdin-e2e.test.ts. | ||
| // Imports the real helper, invokes it against real fd 0, and prints the | ||
| // result as JSON so the test can assert on it. | ||
| import { readStdinIfAvailable } from "../../src/util/stdin" | ||
|
|
||
| const start = Date.now() | ||
| const result = await readStdinIfAvailable() | ||
| const elapsed = Date.now() - start | ||
| process.stdout.write(JSON.stringify({ result, elapsed })) |
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P2: Fixed 100ms first-byte timeout can discard legitimate piped stdin from slow producers. This changes CLI behavior from “wait for pipeline input” to “treat as no stdin” based only on startup timing.
Prompt for AI agents
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks — this is the right concern to raise. It's a deliberate tradeoff, not an oversight, and worth documenting why we picked this point:
The 100ms applies only to first byte, not full drain. Once a single byte arrives, the read runs to EOF with no deadline. So
git diff | run "review"where the diff takes 5 seconds to stream is fine — the timeout is irrelevant after byte 1.The narrow case this affects is a producer that does >100ms of CPU work before emitting any output. Realistic examples:
slow-tool | run "msg"whereslow-toolcomputes for 200ms before flushing. The failure mode is benign —stdinInputbecomes"", the positional message proceeds alone, no silent data corruption.The alternatives we considered (and rejected):
subprocess.run, CI) hangs at 0% CPU forever.-/--stdin) — breaks theecho ctx | run "p"ergonomics that PR fix: skip stdin read when positional message provided #935's reviewer specifically asked us to restore.Empirically, every kernel-buffered producer (shell pipes, file redirects,
git diff,cat,echo) flushes its first byte in <10ms because the data is already in the pipe buffer before the child spawns. 100ms is ~10× the realistic worst case for those.If a real report comes in about a slow-flush producer getting cut, we can either bump the timeout, or add a
--stdin-timeoutflag. Until then, the current value is the right point on the wedge-vs-truncation curve.There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the feedback! I've saved this as a new learning to improve future reviews.