gh-148260: Use at least 1 MiB stack size on musl by vstinner · Pull Request #149993 · python/cpython · GitHub

On Linux when Python is linked to the musl C library, use a thread stack size of at least 1 MiB instead of musl default which is 128 kiB.

Issue: test_threading.test_recursion_limit() does crash (segfault) on Alpine Linux (musl) #148260

On Linux when Python is linked to the musl C library, use a thread stack size of at least 1 MiB instead of musl default which is 128 kiB.

On "AMD64 Alpine Linux NoGIL 3.x" buildbot, currently test_threading.test_recursion_limit() does crash with a stack overflow: issue gh-148260.

With this change, Python is built with -Wl,-z,stack-size=1048576 linker flag and so uses a thread stack size of 1 MiB by default, instead of musl default which is 128 kiB.

With this change, test_threading.test_recursion_limit() does pass. I tested Python built in debug mode (gcc -O3) and release mode (gcc -Og), with and without the GIL (--disable-gil). The buildbot worker uses ./configure --with-pydebug --disable-gil.

@corona10 @erlend-aasland @picnixz: I'm looking for autotools experts to review my configure.ac change. I'm not comfortable with changing the build system.

Currently, the ./configure script doesn't run a built C program, it only uses AC_LINK_IFELSE or AC_COMPILE_IFELSE to check if a C program can be built or linked.

It seems like running a program is the only way to check if Python is linked to the musl library. My change is not specific to the musl library, but instead read the default thread stack size to decide if -Wl,-z,stack-size=1048576 linker flag should be used.

My configure.ac is inspired by Python 3.12 ./configure script which runs $CPP $CPPFLAGS conftest.c >conftest.out to read the output of the C preprocessor. But my change is different since it compiles a C program and then runs the C program. The code is skipped if Python is cross-compiled.

I chose to use $CC -pthread to make sure that the pthread API can be used in the C program. We check if -pthread is supported above, near line 3064, which sets CC="$CC -pthread" if the flag is supported. I should maybe omit -pthread and move my code after the -pthread check.

I think using -pthread explicitly is fine, it doesn't matter if it's passed to the compiler twice.

!buildbot Alpine

🤖 New build scheduled with the buildbot fleet by @zware for commit fd078ac 🤖

Results will be shown at:

https://buildbot.python.org/all/#/grid?branch=refs%2Fpull%2F149993%2Fmerge

The command will test the builders whose names match following regular expression: Alpine

The builders matched are:

AMD64 Alpine Linux PR
AMD64 Alpine Linux NoGIL PR

Cool, the test suite passed on Alpine Linux buildbots:

buildbot/AMD64 Alpine Linux NoGIL PR — Build done.
buildbot/AMD64 Alpine Linux PR — Build done.

I think using -pthread explicitly is fine, it doesn't matter if it's passed to the compiler twice.

For debugging it would probably be nice if the output said whether the stack size is good enough, needs overriding or just unknown (return code 2)

The possible exit codes are:

0: the thread stack is at least 1 MiB
1: the thread stack is less than 1 MiB
2: an error occurred

The two possible outputs are:

checking for thread stack size... default if the stack size is already at least 1 MiB
checking for thread stack size... 1048576 if the stack size must be set to 1 MiB with LDFLAGS

Do you mean that the error case should be rendered differently, such as "error"? Usually, it seems like configure ignores errors silently.

I mean I would like the output to be "checking for the thread stack size... unknown". It can keep behaving like 'default', but at least from the configure output it'll be more obvious it wasn't able to detect the actual thread size. But it's not a big deal if it's too tricky to get right.

Ok, I modified the code to display "error" if gcc/clang failed to build the C program, or if the C program failed (exit code differently than 0 and 1).

!buildbot Alpine

🤖 New build scheduled with the buildbot fleet by @vstinner for commit 86027a6 🤖

Results will be shown at:

https://buildbot.python.org/all/#/grid?branch=refs%2Fpull%2F149993%2Fmerge

The command will test the builders whose names match following regular expression: Alpine

The builders matched are:

AMD64 Alpine Linux PR
AMD64 Alpine Linux NoGIL PR

buildbot Alpine

The worker is down at the moment; I had a storm-related power outage this morning and my UPS has not yet recharged enough to turn the host back on. It will be back by this afternoon.

buildbot/AMD64 Alpine Linux NoGIL PR
buildbot/AMD64 Alpine Linux PR

It works as expected:

configure log: checking for thread stack size... 1048576.
compile log: gcc -shared -Wl,-z,stack-size=1048576 Modules/arraymodule.o -o Modules/array.cpython-316td-x86_64-linux-musl.so
tests: test_threading passed

On "AMD64 Alpine Linux NoGIL PR", test_threading crashed :-( But it passed when rerun...

0:03:47 load avg: 6.33 [255/505/1] test_threading worker non-zero exit code (Exit code -11 (SIGSEGV)) -- running (1): test_compileall (30.8 sec)

(...)
test_start_thread_again (test.test_threading.ThreadingExceptionTests.test_start_thread_again) ... ok
test_concurrent_tee_negative_n (test.test_threading.ThreadingIteratorToolsTests.test_concurrent_tee_negative_n) ... ok
test_concurrent_tee_supports_concurrent_consumers (test.test_threading.ThreadingIteratorToolsTests.test_concurrent_tee_supports_concurrent_consumers) ...
(...)
0:08:02 load avg: 3.48 Run 1 test in parallel using 1 worker process (timeout: 30 min, worker timeout: 35 min)
0:08:26 load avg: 2.66 [1/1] test_threading passed
(...)

Thanks @vstinner for the PR 🌮🎉.. I'm working now to backport this PR to: 3.15.
🐍🍒⛏🤖

GH-150150 is a backport of this pull request to the 3.15 branch.

…150150) gh-148260: Use at least 1 MiB stack size on musl (GH-149993) On Linux when Python is linked to the musl C library, use a thread stack size of at least 1 MiB instead of musl default which is 128 kiB. (cherry picked from commit df6c157) Co-authored-by: Victor Stinner <vstinner@python.org>

Thanks @vstinner for the PR 🌮🎉.. I'm working now to backport this PR to: 3.14.
🐍🍒⛏🤖

GH-151586 is a backport of this pull request to the 3.14 branch.

…151586) gh-148260: Use at least 1 MiB stack size on musl (GH-149993) On Linux when Python is linked to the musl C library, use a thread stack size of at least 1 MiB instead of musl default which is 128 kiB. (cherry picked from commit df6c157) Co-authored-by: Victor Stinner <vstinner@python.org>

pythongh-148260 : Use at least 1 MiB stack size on musl

fd078ac

On Linux when Python is linked to the musl C library, use a thread stack size of at least 1 MiB instead of musl default which is 128 kiB.

vstinner requested review from AA-Turner, corona10, emmatyping and erlend-aasland as code owners May 18, 2026 14:02

bedevere-app Bot mentioned this pull request May 18, 2026

test_threading.test_recursion_limit() does crash (segfault) on Alpine Linux (musl) #148260

Closed

bedevere-app Bot added the awaiting core review label May 18, 2026

vstinner commented May 18, 2026

View reviewed changes

Yhg1s approved these changes May 18, 2026

View reviewed changes

bedevere-app Bot added awaiting merge and removed awaiting core review labels May 18, 2026

Display "unknown" on error

86027a6

vstinner merged commit df6c157 into python:main May 20, 2026
52 checks passed

vstinner deleted the musl_stack_size branch May 20, 2026 11:22

bedevere-app Bot removed the awaiting merge label May 20, 2026

vstinner added the needs backport to 3.15 pre-release feature fixes, bugs and security fixes label May 20, 2026

bedevere-app Bot removed the needs backport to 3.15 pre-release feature fixes, bugs and security fixes label May 20, 2026

markshannon mentioned this pull request May 21, 2026

GH-126910: Make _Py_get_machine_stack_pointer return the actual stack pointer #149103

Merged

vstinner added the needs backport to 3.14 bugs and security fixes label Jun 17, 2026

bedevere-app Bot removed the needs backport to 3.14 bugs and security fixes label Jun 17, 2026

+                  ac_cv_thread_stack_size="default"
+                  if $CC -pthread $CFLAGS conftest.c -o conftest &>/dev/null; then
+                    ./conftest &>/dev/null
+                    if test $? -eq 1; then

gh-148260: Use at least 1 MiB stack size on musl by vstinner · Pull Request #149993 · python/cpython · GitHub

Uh oh!

Uh oh!

gh-148260: Use at least 1 MiB stack size on musl - #149993

gh-148260: Use at least 1 MiB stack size on musl#149993vstinner merged 2 commits intopython:mainpython/cpython:mainfrom vstinner:musl_stack_sizevstinner/cpython:musl_stack_sizeCopy head branch name to clipboard

Conversation

vstinner commented May 18, 2026 • edited by bedevere-app Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

vstinner commented May 18, 2026

Uh oh!

vstinner commented May 18, 2026

Uh oh!

vstinner May 18, 2026

Choose a reason for hiding this comment

Uh oh!

Yhg1s May 18, 2026

Choose a reason for hiding this comment

Uh oh!

zware commented May 18, 2026

Uh oh!

bedevere-bot commented May 18, 2026

Uh oh!

vstinner commented May 18, 2026

Uh oh!

Yhg1s May 18, 2026

Choose a reason for hiding this comment

Uh oh!

Yhg1s May 18, 2026

Choose a reason for hiding this comment

Uh oh!

vstinner May 18, 2026

Choose a reason for hiding this comment

Uh oh!

Yhg1s May 18, 2026

Choose a reason for hiding this comment

Uh oh!

vstinner May 18, 2026

Choose a reason for hiding this comment

Uh oh!

vstinner commented May 19, 2026

Uh oh!

bedevere-bot commented May 19, 2026

Uh oh!

zware commented May 19, 2026

Uh oh!

vstinner commented May 20, 2026

Uh oh!

Uh oh!

miss-islington-app Bot commented May 20, 2026

Uh oh!

bedevere-app Bot commented May 20, 2026

Uh oh!

miss-islington-app Bot commented Jun 17, 2026

Uh oh!

bedevere-app Bot commented Jun 17, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

gh-148260: Use at least 1 MiB stack size on musl#149993
vstinner merged 2 commits into
python:mainfrom
vstinner:musl_stack_size

vstinner commented May 18, 2026 •
edited by bedevere-app Bot
Loading