v0.19: Fix some bugs in std/os signals #1492

jay · 2026-09-08T06:33:07Z

jay commented

2026-09-08 06:33:07 +00:00

Fixes portability and correctness bugs introduced in :std/os signal handling by #1484:

kqueue FFI declarations used invalid C type identifiers.
Oversized kill arguments could truncate and select unintended targets.
Concurrent initialization could route Linux signals to an unused dispatcher.

Focused Linux tests pass. Paired benchmarks found no statistically significant registration or callback-latency regression. Restored kill checks add approximately 13 ns in a zero-signal probe; synthetic burst throughput was approximately 1% lower.

Native BSD/macOS kqueue smoke testing remains pending.

Warning

LLM-assisted with gpt-6-astra and gerbil-mcp.

Fixes portability and correctness bugs introduced in `:std/os` signal handling by #1484: - `kqueue` FFI declarations used invalid C type identifiers. - Oversized `kill` arguments could truncate and select unintended targets. - Concurrent initialization could route Linux signals to an unused dispatcher. Focused Linux tests pass. Paired benchmarks found no statistically significant registration or callback-latency regression. Restored `kill` checks add approximately 13 ns in a zero-signal probe; synthetic burst throughput was approximately 1% lower. Native BSD/macOS `kqueue` smoke testing remains pending. > [!WARNING] > LLM-assisted with `gpt-6-astra` and `gerbil-mcp`.

jay added 4 commits

2026-09-08 06:33:07 +00:00

fix(kqueue): use valid C type identifiers dba66398a5

fix(signals): check kill integer conversions 4eb566461f

Keep errno handling while rejecting C-int overflow before kill can select an unintended target. Exercise the production binding only through a C test double.

fix(signals): initialize the dispatcher atomically 8f54e7b5ae

fix(signals): receive inherited blocked signals 1cac361603

Preserve caller-owned thread masks. Let signalfd consume pending blocked signals alongside sigaction delivery, forwarding both through the existing dispatcher. Serialize descriptor updates and consumption with registration and removal.

vyzo reviewed

2026-09-10 17:15:08 +00:00

vyzo left a comment

jay is this still WIP?

jay commented

2026-09-10 18:21:30 +00:00

@vyzo hey, yes, I had astra go to town on this and I have not had time yet to validate its work or do any performance benchmarking. This week and weekend is super busy for me. You are welcome to take this over if you would like, or scrap it, otherwise I will have some more time next week!

vyzo commented

2026-09-10 18:39:46 +00:00

it's ok we can wait for you.

jay added 1 commit

2026-09-11 00:48:25 +00:00

revert(signals): defer blocked-signal reception 5411e041f3

Keep the three independent safety fixes without the receiver cost.
The smallest tested receiver still slows registration and callbacks
after activation. Preserve it on v0.19-blocked-signals-deferred.

This reverts commit 1cac361603.

jay added 1 commit

2026-09-11 00:55:05 +00:00

Merge branch 'v0.19-staging' into v0.19-fix-signals f57c997548

jay changed title from ~~WIP: v0.19: Fix some bugs in std/os signals~~ to v0.19: Fix some bugs in std/os signals

2026-09-11 00:55:06 +00:00

jay requested reviews from vyzo, fare

2026-09-11 00:55:46 +00:00

vyzo reviewed

2026-09-11 07:28:13 +00:00

vyzo left a comment

generally looks good, but i left a couple of comments that need addressing.

i would strongly prefer to put the ffi test support stuffs in a separate signal-test-support module that gets compiled with the stdlib.

generally looks good, but i left a couple of comments that need addressing. i would strongly prefer to put the ffi test support stuffs in a separate signal-test-support module that gets compiled with the stdlib.

src/std/os/kqueue.ss Outdated

					
				@ -55,2 +55,2 @@

				(def-C-type/pointer kevent "struct kevent" release: "ffi_free")

				(def-C-type/pointer timespec "struct timespec" release: "ffi_free")

				(C-declare "typedef struct kevent ffi_kevent;")

				(C-declare "typedef struct timespec ffi_timespec;")

note that we already have timespec as def-C-struct, you might want to reuse that. it also avoids mallocs.

src/std/os/signal-test.ss Outdated

					
				@ -0,0 +14,4 @@

				    (lambda (output)

				      (display #<<END-SCHEME

				(import :std/ffi)

				(C-declare #<<END-C

why not create a separate signal-test-support module that is compiled and contains the ffi code.

this is the pattern we use for tests that require native code.

why not create a separate signal-test-support module that is compiled and contains the ffi code. this is the pattern we use for tests that require native code.

jay added 1 commit

2026-09-11 16:18:56 +00:00

reuse timespec and compile test helpers 8805340a50

jay commented

2026-09-11 16:20:09 +00:00

@vyzo addressed your feedback

vyzo approved these changes

2026-09-11 19:07:19 +00:00

vyzo left a comment

lgtm. @fare you want to do a clanker assisted pass?

fare commented

2026-09-12 18:15:11 +00:00

The clanker (Sol High) is satisfied—except for the part where we assume that pid_t is int, and don't check. But I believe that is a Gambit FFI issue. Maybe we need to improve the Gambit FFI at some point?

Clanky also asks for test runs on BSD, but oh well.

The clanker (Sol High) is satisfied—except for the part where we assume that pid_t is int, and don't check. But I believe that is a Gambit FFI issue. Maybe we need to improve the Gambit FFI at some point? Clanky also asks for test runs on BSD, but oh well.

fare force-pushed v0.19-fix-signals from 8805340a50 to 9d0639d043

2026-09-12 18:15:22 +00:00

Compare

fare merged commit 69d96e16a0 into v0.19-staging

2026-09-12 18:23:56 +00:00

fare referenced this pull request from a commit

2026-09-12 18:23:58 +00:00

v0.19: Fix some bugs in std/os signals (#1492)

Sign in to join this conversation.

No reviewers