emrun --kill_exit 4.0.16 regression

[neondev9]

Hello,

I just updated EmSDK from 4.0.13 to 4.0.18 and noticed that CI tests for WASM in my project failed on timeout.
I have Catch2 based unittests, which passed successfully.
CI failed on timeout.

I launch unittest using following cmdline:
emrun --browser chrome --hostname=localhost --kill_exit --browser_args="--enable-experimental-webassembly-features --headless --remote-debugging-port=0 --disable-gpu --disable-software-rasterizer" <pathtotest.html>

They fail when test code is finished on exit.
If I add --verbose for emrun, then it says:

Web page has quit with a call to exit() with return code 0. Shutting down web server. Pass --serve-after-exit to keep serving even after the page terminates with exit().
Web server loop done.

When I try to interrupt it using Ctrl+C, then:

Traceback (most recent call last):
  File "/Users/dev/Documents/Project/ThirdParty/EmSDK/Source/upstream/emscripten/emrun.py", line 1896, in <module>
    sys.exit(main(sys.argv[1:]))
             ~~~~^^^^^^^^^^^^^^
  File "/Users/dev/Documents/Project/ThirdParty/EmSDK/Source/upstream/emscripten/emrun.py", line 1888, in main
    returncode = run(args)
  File "/Users/dev/Documents/Project/ThirdParty/EmSDK/Source/upstream/emscripten/emrun.py", line 1861, in run
    httpd.server_close()
    ~~~~~~~~~~~~~~~~~~^^
  File "/Users/dev/Documents/Project/ThirdParty/EmSDK/Source/python/3.13.3_64bit/lib/python3.13/socketserver.py", line 715, in server_close
    self._threads.join()
    ~~~~~~~~~~~~~~~~~~^^
  File "/Users/dev/Documents/Project/ThirdParty/EmSDK/Source/python/3.13.3_64bit/lib/python3.13/socketserver.py", line 661, in join
    thread.join()
    ~~~~~~~~~~~^^
  File "/Users/dev/Documents/Project/ThirdParty/EmSDK/Source/python/3.13.3_64bit/lib/python3.13/threading.py", line 1092, in join
    self._handle.join(timeout)

Note: I compared behaviour between different EmSDK versions and this problem didn't exist on 4.0.15 and 4.0.16 introduced it.
I have the same issue across different desktop platforms: Windows 11, Linux (Ubuntu 24), MacOS 26.0.1.

I wonder if it is known issue.
I'm not familiar with Python debugging, but I can test different pre 4.0.16 commits if it necessary.
Let me know if you need more information.

Regards,
Kostiantyn

[sbc100]

Interesting, I'm guess this might not be related to the emrun.py script itself since it saw very minimal changes between 4.0.15 and 4.0.16:

$ git l 4.0.15..4.0.16 emrun.py
0167321e7 Lift the previous 100 max workers limit (#25410) 4 weeks ago [juj]
24fcd2672 Another round of `deadcode` feedback and error suppression. NFC (#25369) 5 weeks ago [Sam Clegg]
1294dd59e Remove more dead python code found using `deadcode` module. NFC (#25359) 5 weeks ago [Sam Clegg]

Non of these changes should have changed the behaviour of emrun at all AFAIKT.

Can you describe a little more exactly what is happening? You say "They fail when test code is finished on exit" but then you also say "When I try to interrupt it using Ctrl+C" which makes it sounds like something it still running.

Can you share the full set of link flags you used to build your tests?

[neondev9]

Something like this

Compiling:
-std=c++20 -g -sMEMORY64=1 -pthread -fno-exceptions -Wall -Wextra -fPIC -g -O0 -Winvalid-pch -fpch-instantiate-templates -Xclang -emit-pch -Xclang -include -Xclang /Users/dev/Documents/Project/Sources/_Projects.CMake/WebAssembly/NeonCommon_UnitTests/CMakeFiles/NeonCommon_UnitTests.dir/Debug/cmake_pch.hxx -x c++-header -MD -MT

Linking:
-std=c++20 -g -sMEMORY64=1 -pthread -fno-exceptions -mwasm64 -sSTACK_SIZE=1048576 -sALLOW_MEMORY_GROWTH=1 -sASYNCIFY=1 -sASYNCIFY_STACK_SIZE=4194304 -sWASM_BIGINT --closure=1 -g --emrun

[neondev9]

emrun/server doesn't close properly when int main() is finished
according to verbose output, that what is happening:
Web page has quit with a call to exit() with return code 0. Shutting down web server.

[sbc100]

So emrun says its going to close down but does not?

If you have time it might be useful to know more precisely which version of emsdk broke this workflow. The bisect instruction are here: https://emscripten.org/docs/contributing/developers_guide.html#bisecting

[sbc100]

It looks like its hanging on this line

emscripten/emrun.py

Line 1861 in 8690df6

httpd.server_close()

. As far as I can tell nothing in this area has changed in a long time.. and this is a call into the python standard library.

[neondev9]

Yeah, sorry, that's what I meant initially.
There is a short mention that tests fail due to timeout in CI setup at the beginning.
No crash.
emrun just doesn't close when app is already finished and --kill_exit is used starting from 4.0.16.

I'll try bisect for commits between 4.0.15 and 4.0.16 in few days and come back.

[neondev9]

I'll try bisect for commits between 4.0.15 and 4.0.16 in few days and come back.

upd: so prebuilt binaries are not available for non-release versions, right?

[sbc100]

Yes, there are prebuilt binaries available for any revision of the emscripten-release repo: https://chromium.googlesource.com/emscripten-releases/

As mentioned in the bisect instruction you can install those binaries using emsdk install [HASH], and thus you can besect between the hack for 4.0.15 and the hash for 4.0.16.

The result will be a specific commit to the emscripten-releases repo which is to blame which should narrow down the number of possible emscripten PRs which are to blame.

[neondev9]

Hi @sbc100,
Didn't have time to try prebuilt binaries for changes between 4.0.15 and 4.0.16 and just tried cherry-pick for emrun.py.
This commit caused this regression: 1294dd5

[sbc100]

Wow, thanks for tracking that down. I guess the protocol_version = 'HTTP/1.1' on the class is not working as expected

What version of python are using to run emrun.py?

[neondev9]

There are logs:
#25654 (comment)
It is python synced by emsdk - 3.13.3.