emrun-related JS code is broken

[oitel]

emrun cannot get output from a browser (tested for Firefox 135 and Chromium 133) for the apps compiled with emsdk v4.0.3 and v4.0.4:

Web server root directory: /tmp/em_test/v403
Starting web server: http://0.0.0.0:6931/
Starting browser: /usr/bin/xdg-open http://localhost:6931/main.html
/usr/bin/xdg-open
Searching for processes by full path name "/usr/bin/xdg-open".. found 0 entries
Launched browser process with pid=200059
Now listening at http://0.0.0.0:6931/
Entering web server loop.
First navigation occurred. Identifying currently running browser processes
Searching for processes by full path name "/usr/bin/xdg-open".. found 0 entries
Was unable to detect the browser process that was spawned by emrun. This may occur if the target page was opened in a tab on a browser process that already existed before emrun started up.
The html page you are running is not emrun-capable. Stdout, stderr and exit(returncode) capture will not work. Recompile the application with the --emrun linker flag to enable this, or pass --no-emrun-detect to emrun to hide this check.

For the apps compiled with emsdk v4.0.2, emrun works successfully:

Web server root directory: /tmp/em_test/v402
Starting web server: http://0.0.0.0:6931/
Starting browser: /usr/bin/xdg-open http://localhost:6931/main.html
/usr/bin/xdg-open
Searching for processes by full path name "/usr/bin/xdg-open".. found 0 entries
Launched browser process with pid=200392
Now listening at http://0.0.0.0:6931/
Entering web server loop.
First navigation occurred. Identifying currently running browser processes
Searching for processes by full path name "/usr/bin/xdg-open".. found 0 entries
Was unable to detect the browser process that was spawned by emrun. This may occur if the target page was opened in a tab on a browser process that already existed before emrun started up.
POST: "/stdio.html" (path: "/stdio.html", query: "")
POST: "/stdio.html" (path: "/stdio.html", query: "")
Hello, world!
POST: "/stdio.html" (path: "/stdio.html", query: "")
Shutting down because browser is no longer alive
Browser process has shut down, quitting web server.
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.
Closed web server.
emrun quitting with process exit code 0

Seems like it's caused by calling a non-existing addOnExit function from src/emrun_postjs.js. The function was moved in a series of commits:
13947f4 Unify more of the minimal runtime and normal runtime initialization code. (#23488)
a4f64a3 Move all addOn functions into a JS library. (#23565)

Version of emscripten/emsdk:

emcc (Emscripten gcc/clang-like replacement + linker emulating GNU ld) 4.0.3 (a9651ff57165f5710bb09a5fe52590fd6ddb72df)
clang version 21.0.0git (https:/github.com/llvm/llvm-project 6dc41a639334b913e762f65410fcd14a722b137f)
Target: wasm32-unknown-emscripten
Thread model: posix

Failing command line in full:

emcc -s ENVIRONMENT=web --emrun main.cpp -o main.html && emrun --verbose main.html

main.cpp:

#include <iostream>

int main()
{
    std::cout << "Hello, world!" << std::endl;
    return 0;
}

Update 1: added info about v4.0.4.
Update 2: added info about the addOnExit function.
Update 3: updated the title.

[oitel]

Managed to fix it locally by patching the link tool:

diff --git a/tools/link.py b/tools/link.py
index 9440e8908..2e82c0cf4 100644
--- a/tools/link.py
+++ b/tools/link.py
@@ -674,6 +674,8 @@ def phase_linker_setup(options, linker_args):  # noqa: C901, PLR0912, PLR0915
     if user_settings.get('EXIT_RUNTIME') == '0':
       exit_with_error('--emrun is not compatible with EXIT_RUNTIME=0')
     settings.EXIT_RUNTIME = 1
+    # emrun_postjs.js needs this library function.
+    settings.DEFAULT_LIBRARY_FUNCS_TO_INCLUDE += ['$addOnExit']
 
   if options.cpu_profiler:
     options.post_js.append(utils.path_from_root('src/cpuprofiler.js'))

[sbc100]

Thanks so much the fix! Would you like to send a PR. I'll take a look at why that doesn't seem to be covered by our test.

[oitel]

Sure thing, I'll prepare a PR a bit later.

[esotericpig]

Bump. I also get this same error in v4.0.4 of Emscripten with a simple main.cpp:

Uncaught ReferenceError: addOnExit is not defined

[sbc100]

Are you using --emrun @esotericpig?

If not can you see where the undefined reference to addOnExit is coming from? i.e. in devtools can you see a backtrace?

[sbc100]

I believe this specific issue was closes by #23830

[esotericpig]

Yes, I have --emrun.

I searched here and then compared it to my grep locally:

$ grep --exclude-dir=.git --exclude-dir=cache -ir 'addOnExit'
upstream/emscripten/test/test_other.py:                      '-sEXPORTED_RUNTIME_METHODS=baz,addOnExit',
upstream/emscripten/test/test_other.py:      import init, { _foo as foo, _bar as bar, baz, qux, addOnExit, HEAP32 } from "./modularize_instance.mjs";
upstream/emscripten/test/test_other.py:      assert(typeof addOnExit === 'function'); // exported runtime function with EXPORTED_RUNTIME_METHODS
upstream/emscripten/test/test_other.py:addOnExit(() => console.log("addOnExit"));
upstream/emscripten/test/test_other.py:      ['addOnExit', True],
upstream/emscripten/test/test_other.py:      '-sDEFAULT_LIBRARY_FUNCS_TO_INCLUDE=$addOnPreRun,$addOnInit,$addOnPostCtor,$addOnPreMain,$addOnExit,$addOnPostRun',
upstream/emscripten/src/threadprofiler.js:    addOnExit(() => clearInterval(i));
upstream/emscripten/src/threadprofiler.js:      addOnExit(() => clearInterval(i));
upstream/emscripten/src/lib/libglut.js:  glutInit__deps: ['$Browser', '$addOnExit'],
upstream/emscripten/src/lib/libglut.js:    addOnExit(() => {
upstream/emscripten/src/lib/libhtml5.js:    '$addOnExit',
upstream/emscripten/src/lib/libhtml5.js:        addOnExit(JSEvents.removeAllEventListeners);
upstream/emscripten/src/lib/libcore.js:  $addOnExit__deps: ['$onExits'],
upstream/emscripten/src/lib/libcore.js:  $addOnExit: (cb) => onExits.unshift(cb),
upstream/emscripten/src/emrun_postjs.js:      addOnExit(() => {
upstream/emscripten/tools/link.py:    settings.DEFAULT_LIBRARY_FUNCS_TO_INCLUDE += ['$addOnInit', '$addOnExit']
upstream/emscripten/ChangeLog.md:   addOnPreRun, addOnInit, addOnPreMain, addOnExit, addOnPostRun,

After comparing the two, I found that link.py only has one instance of 'addOnExit' locally.

My emsdk repo is on branch main. I did git fetch && git pull. I also tried deleting upstream. Then I ran all of the normal stuff:

source emsdk_env.sh
emsdk install latest
emsdk activate latest
source emsdk_env.sh

I did grep again and still not in link.py. Maybe I should delete cache too? Not sure.

Anyway, I manually edited link.py to add this line, and it now works without any errors.

[sbc100]

If you want to try out as-yet-unreleased changes you need to emsdk install tot

[esotericpig]

If you want to try out as-yet-unreleased changes you need to emsdk install tot

Okay, but using emsdk install latest && emsdk activate latest and building a project causes this error, so I guess that's by design? I thought this would use the latest stable version, but guess it's using the main branch instead?

[sbc100]

If you want to try changes out that have not yet made it into a release you can use emsdk install tot (for Tip-Of-Tree)