fix: retry waitpid on EINTR; reap untracked children (stopped-job job-control loss) #20

Merged

ober merged 6 commits from fix/job-control-waitpid-eintr into main 2026-08-25 22:50:53 -04:00

Conversation 0 Commits 6 Files changed 3 +37 -4

ober commented 2026-08-10 22:35:51 -04:00

Owner

Copy link

Problem

A stopped foreground process group could escape job control when a signal such as SIGWINCH interrupted the blocking foreground wait. The shell treated the interrupted wait as a successful exit, so jobs and fg lost the process group even though it remained alive. Unreaped auxiliary children could also accumulate as zombies.

Root cause

ffi_do_waitpid called waitpid once without retrying EINTR. wait-for-foreground-process-raw then mapped the failed wait to an apparent exit.

Changes

• ffi-shim.c: retry waitpid when it returns EINTR.
• jobs.ss: reap terminated children with waitpid(-1, WNOHANG), update tracked processes by PID, and continue polling stopped processes.
• Refresh the branch onto current main and resolve the build/workflow conflicts.
• Advance VERSION from 0.5.0 to 0.5.1.

Verification

• make test: 691 jsh unit tests, 25 search UI tests, and 13 secure-history PTY checks passed.
• make binary: passed on macOS; jsh-macos -c "echo pr20-refresh-smoke" passed.
• make pre-pr-macos-linux-amd64-all: passed, producing the static stripped jsh-linux-amd64 all-features binary.
• jobs.ss balance and changed-line security checks passed.

Forgejo CI is re-running against the refreshed branch.

## Problem A stopped foreground process group could escape job control when a signal such as SIGWINCH interrupted the blocking foreground wait. The shell treated the interrupted wait as a successful exit, so `jobs` and `fg` lost the process group even though it remained alive. Unreaped auxiliary children could also accumulate as zombies. ## Root cause `ffi_do_waitpid` called `waitpid` once without retrying `EINTR`. `wait-for-foreground-process-raw` then mapped the failed wait to an apparent exit. ## Changes - `ffi-shim.c`: retry `waitpid` when it returns `EINTR`. - `jobs.ss`: reap terminated children with `waitpid(-1, WNOHANG)`, update tracked processes by PID, and continue polling stopped processes. - Refresh the branch onto current `main` and resolve the build/workflow conflicts. - Advance `VERSION` from 0.5.0 to 0.5.1. ## Verification - `make test`: 691 jsh unit tests, 25 search UI tests, and 13 secure-history PTY checks passed. - `make binary`: passed on macOS; `jsh-macos -c "echo pr20-refresh-smoke"` passed. - `make pre-pr-macos-linux-amd64-all`: passed, producing the static stripped `jsh-linux-amd64` all-features binary. - `jobs.ss` balance and changed-line security checks passed. Forgejo CI is re-running against the refreshed branch.

ober added 1 commit 2026-08-10 22:35:51 -04:00

fix: retry waitpid on EINTR; reap untracked children in job-update-status! ... 07fcaab79e

ffi_do_waitpid returned -1 on EINTR, and wait-for-foreground-process-raw
mapped that to (values 0 #f) = clean exit. A signal delivered to the shell
during the blocking foreground wait (e.g. SIGWINCH from a ,mux pane resize)
therefore made jsh abandon a live child: the process group kept running,
stopped on its next tty access (SIGTTIN), and wedged in T state invisible
to jobs/fg while holding resources (observed with codex's writer lock).

- ffi-shim.c: retry waitpid on EINTR in ffi_do_waitpid.
- jobs.ss: job-update-status! now reaps all terminated children via a
  waitpid(-1, WNOHANG) sweep (kills the zombie accumulation), marks
  tracked processes it reaps by pid, and also polls 'stopped processes
  so a process killed while stopped is still reaped.

Verified on a pty: before, one SIGWINCH during a foreground wait wedged
/bin/sh + reader child in T state with jobs empty and 'fg: %%: no such
job'; after, the wait survives and the job stays managed. Zombie count no
longer grows per command (was +1 per command).

ober scheduled this pull request to auto merge when all checks succeed 2026-08-10 22:56:54 -04:00

ober added 1 commit 2026-08-10 23:07:15 -04:00

fix: use system cc for ffi shims 6690062813

ober added 1 commit 2026-08-10 23:10:06 -04:00

fix: prepare jsh libraries before tests bdf4481318

ober added 1 commit 2026-08-10 23:14:02 -04:00

fix: make jerbuild script readable by runner toolchain 0e878339c6

ober added 1 commit 2026-08-10 23:18:32 -04:00

fix: avoid list-sort import exclusion a11d376a75

ober added 1 commit 2026-08-25 22:41:42 -04:00

Merge main into fix/job-control-waitpid-eintr 60ee1a1633

ober scheduled this pull request to auto merge when all checks succeed 2026-08-25 22:50:38 -04:00

ober merged commit 740c21d010 into main 2026-08-25 22:50:53 -04:00

ober referenced this pull request from a commit 2026-08-25 22:50:53 -04:00

Merge pull request #20

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

ober/jerboa-shell-extras!20

No description provided.