Fix running emrun tests on Linux against the Linux system browser by juj · Pull Request #26810 · emscripten-core/emscripten

juj · 2026-04-28T16:15:04Z

Fix running emrun tests on Linux against the Linux system browser. This has a few fixes in one:

Fix emrun to work around Firefox https://bugzil.la/1996614 by moving -url to last on cmdline.

Move emrun suite to its own file test_emrun, so that it can be migrated to run single-threaded. This is because emrun may tear down the test browser in a non-parallel-friendly manner:

emscripten/emrun.py

Lines 423 to 443 in 92ef17c

    
           def detect_browser_processes(): 
        
             if not browser_exe: 
        
               return # Running with --no-browser, we are not binding to a spawned browser. 
        
             global current_browser_processes 
        
             logv('First navigation occurred. Identifying currently running browser processes') 
        
             running_browser_processes = list_processes_by_name(browser_exe) 
        
             def pid_existed(pid): 
        
               for proc in previous_browser_processes: 
        
                 if proc['pid'] == pid: 
        
                   return True 
        
               return False 
        
             for p in running_browser_processes: 
        
               logv(f'Detected running browser process id: {p["pid"]}, existed already at emrun startup? {pid_existed(p["pid"])}') 
        
             current_browser_processes = [p for p in running_browser_processes if not pid_existed(p['pid'])] 
        
             if len(current_browser_processes) == 0: 
        
               logv('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.')

Adjust the platform.ini lookup of system Firefox on Linux to find it in /usr/lib/, or fall back to assuming oldest supported browser if not found.

…x emrun to work around Firefox https://bugzil.la/1996614

sbc100 · 2026-04-28T17:36:58Z

+  if 'firefox' in browser_exe:
+    browser += ['-url', url]
+  else:
+    browser += [url]


With the old code the url was only added in the non-android case.

Thanks, good catch..

sbc100 · 2026-04-28T17:38:41Z

+        if WINDOWS:
+          exe_file += '.exe'
+        if os.path.isfile(exe_file):
+          return exe_file


Can you just use shutil.which here?

If not, can you move this utility out the top level?

Thanks, I wasn't aware of that - that's much better.

…ved to its own file.

juj · 2026-04-28T19:43:52Z

+
+from tools.shared import EMCC, PIPE
+
+


The emrun class below is moved without changes.

sbc100 · 2026-04-28T20:32:36Z


  if not EMTEST_BROWSER:
-    EMTEST_BROWSER = 'google-chrome'
+    EMTEST_BROWSER = shutil.which('google-chrome')


Is this needed? IIRC this works as expected on my desktop without this change.

Yes, because if I don't do that search, I don't know if I should search for firefox. I imagine we do want to favor google-chrome as the first default, if both are installed on the system?

I think just having single default is fine. We have --browser=xxxx in addition to EMTEST_BROWSER=xxx that folks can use to specifiy non-default values.

It would be convenient to have this automatically look up like this. This way I can run e.g. test/runner emrun on Raspberry Pi, finding Chrome as default, and the same command on Debian Linux, finding firefox as default.

This makes it simpler for me to not need to encode on the CI runner master script side the knowledge of "what browser do you have". I'd imagine this can help other users as well, who have a certain browser on their system. (and not have a FTUE that errors)

juj · 2026-04-28T20:32:42Z

This is now ready to land, though there is an odd error

======================================================================
FAIL [0.001s]: test_fp16 (test_other.other)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/root/project/test/common.py", line 381, in resulting_test
    return func(self, *args)
  File "/root/project/test/decorators.py", line 183, in decorated
    return func(self, *args, **kwargs)
  File "/root/project/test/test_other.py", line 14948, in test_fp16
    self.do_runf('test_fp16.c', cflags=['-msimd128', '-mfp16', '-mrelaxed-simd', '-sENVIRONMENT=shell'] + args)
  File "/root/project/test/common.py", line 1393, in do_runf
    return self._build_and_run(filename, expected_output, **kwargs)
  File "/root/project/test/common.py", line 1441, in _build_and_run
    js_output = self.run_js(js_file, engine, args,
  File "/root/project/test/common.py", line 1025, in run_js
    self.fail('JS subprocess failed (%s): %s (expected=%s).  Output:\n%s' % (error.cmd, error.returncode, assert_returncode, ret))
AssertionError: JS subprocess failed (/root/.jsvu/bin/v8 --wasm-staging --experimental-wasm-fp16 /tmp/emtest_cmsh_y2k/emscripten_test_other__ga4wtwi/test_fp16.js --): 1 (expected=0).  Output:
V8 is running with experimental features enabled. Stability and security will suffer.
console.error: failed to asynchronously prepare wasm: CompileError: WebAssembly.instantiate(): Compiling function #12 failed: invalid simd opcode @+8623
console.error: Aborted(CompileError: WebAssembly.instantiate(): Compiling function #12 failed: invalid simd opcode @+8623)
/tmp/emtest_cmsh_y2k/emscripten_test_other__ga4wtwi/test_fp16.js:520: RuntimeError: Aborted(CompileError: WebAssembly.instantiate(): Compiling function #12 failed: invalid simd opcode @+8623)
  var e = new WebAssembly.RuntimeError(what);
          ^
RuntimeError: Aborted(CompileError: WebAssembly.instantiate(): Compiling function #12 failed: invalid simd opcode @+8623)
    at abort (/tmp/emtest_cmsh_y2k/emscripten_test_other__ga4wtwi/test_fp16.js:520:11)
    at instantiateArrayBuffer (/tmp/emtest_cmsh_y2k/emscripten_test_other__ga4wtwi/test_fp16.js:602:5)
    at async createWasm (/tmp/emtest_cmsh_y2k/emscripten_test_other__ga4wtwi/test_fp16.js:691:16)

1 pending unhandled Promise rejection(s) detected.

from a fp16 test, that should not be related to this PR. I wonder how that failed.

sbc100 · 2026-04-28T20:32:56Z

+    if not EMTEST_BROWSER:
+      EMTEST_BROWSER = shutil.which('firefox')
+    if not EMTEST_BROWSER:
+      EMTEST_BROWSER = 'default-browser-not-found'


How about making this into a hard error?

This code is run to detect a browser in the general test runner config, even if a browser is not actually run. So it was more straightforward to keep it as-is.

I think we should fix that at some point. We shouldn't need to go through browser config unless we are running browser tests. I know there are some subtle issues with multiprocessing and startup here, so its maybe non-trival though.

Ok, about about:

EMTEST_BROWSER = 'google-chrome' if not shutil.which(EMTEST_BROWSER): EMTEST_BROWSER = 'firefox' if not shutil.which(EMTEST_BROWSER): # FIXME: This should really be and error, but this code currently also runs for non-browser tests. EMTEST_BROWSER = 'default-browser-not-found'

This way the value of EMTEST_BROWSER doesn't become and absolute path as part of this change.

Yeah, that works

sbc100 · 2026-04-28T20:34:33Z

+      self.assertContained('Hello, world!', stdout)
+      self.assertContained('Testing ASCII characters: !"$%&\'()*+,-./:;<=>?@[\\]^_`{|}~', stdout)
+      self.assertContained('Testing char sequences: %20%21 &auml;', stdout)
+      self.assertContained('hello, error stream!', stderr)


Can you make the "move emrun into its own file" into separate PR?

(Seems reasonable to combine with adding it to list of non-parall suites, since that is the motivation for moving it).

I'd prefer not, that would cause more churn at this point.

sbc100 · 2026-04-28T20:42:21Z

+    ini_path = find_system_firefox_platform_ini()
+    if not ini_path:
+      logger.warning(f'Firefox browser detected in {EMTEST_BROWSER}, but could not find Firefox platform.ini to detect Firefox version. Assuming OLDEST_SUPPORTED_FIREFOX={OLDEST_SUPPORTED_FIREFOX}')
+      return OLDEST_SUPPORTED_FIREFOX


Have you ever run into this issue?

Maybe we should just assert os.path.exists(init_path), 'unable to find firefox ini file' below before reading it?

Yes, the CircleCI build-linux bot just ran into it.. that is why I added that.

It has /usr/bin/firefox but no platform.ini anywhere in sight.

On my own systems I do have platform.ini in /usr/lib/firefox-esr/platform.ini or /usr/lib/firefox/platform.ini

I debugged CircleCI with a search from /usr/ to find if there's a platform.ini somewhere, but that turned up nothing.

8b43aae#diff-eb5e6400ad6f92fec8556d07ec1d8d14915c79814e87f354c5ab65e6163f6ad1R176-R177

The firefox we use in circleci installed from tar archive in EMTEST_BROWSER="$HOME/firefox/firefox"? Is that the one you mean?

This step: https://app.circleci.com/pipelines/github/emscripten-core/emscripten/50907/workflows/6696a39d-0750-4b09-9165-349d61aa099b/jobs/1184191

has /usr/bin/firefox (shutil.which('firefox') returned /usr/bin/firefox), but does not have /usr/.../platform.ini anywhere.

EMTEST_BROWSER: /usr/bin/firefox -new-instance -wait-for-browser Searching for platform.ini /usr/local/lib/python3.10/dist-packages/numpy/typing/tests/data/mypy.ini /usr/local/lib/python3.10/dist-packages/numpy/_core/lib/npy-pkg-config/npymath.ini /usr/local/lib/python3.10/dist-packages/numpy/_core/lib/npy-pkg-config/mlib.ini Searching for platform.ini over Traceback (most recent call last): File "/root/project/./test/runner.py", line 775, in <module> sys.exit(main()) File "/root/project/./test/runner.py", line 732, in main log_test_environment() File "/root/project/./test/runner.py", line 655, in log_test_environment print(f'Firefox version: {browser_common.get_firefox_version()}') File "/root/project/test/browser_common.py", line 189, in get_firefox_version m = re.search(r"^Milestone=(.*)$", read_file(ini_path), re.MULTILINE) File "/root/project/tools/utils.py", line 157, in read_file with open(file_path, encoding='utf-8') as fh: FileNotFoundError: [Errno 2] No such file or directory: '/usr/bin/platform.ini'

Now I see.. this task gives a clue: https://app.circleci.com/pipelines/github/emscripten-core/emscripten/50916/workflows/c9a01534-7402-4de2-aac7-028abe35de8e/jobs/1184421

It prints

Command '/usr/bin/firefox' requires the firefox snap to be installed. Please install it with: snap install firefox

so I wonder if there's a stub executable in /usr/bin/firefox present that prints out that message. So it is expected that there wouldn't exist any platform.ini on the system.

~~But now I'm puzzled a bit as to why this is detecting /usr/bin/firefox in the first place...~~

Oh that's because I wrote a silly typo.. fixed now.

That CI job should not have EMTEST_BROWSER set at all.. its not trying to run any browser tests.

I guess your new code is causing that error because its now finding the /usr/bin/firefox even though it doesn't needed to for this CI step.

My guess is that /usr/bin/firefox is a symlink to somewhere like /snap/..., so maybe following the symlinks would fix it.

Ideally we could fix that TODO since then it would be moot point, but I know it wasn't easy last time I tried to move that code around.

I'm happy to iterate on this stuff in followups if you want to land this change now.

Thanks, yeah hopefully after this lands, the instability from emrun suite will be gone.

juj added 2 commits April 28, 2026 19:14

Fix running emrun tests on Linux against the Linux system browser. Fi…

a21cc1c

…x emrun to work around Firefox https://bugzil.la/1996614

Debug

c6ddb7f

sbc100 reviewed Apr 28, 2026

View reviewed changes

juj added 9 commits April 28, 2026 20:38

Default to a nonexisting browser name if not found

d5cb0db

Use shutil.which()

e0df2e0

Add not options.android

0c632a1

XXXX DEBUG

8b43aae

Assume oldest supported Firefox if Firefox platform.ini is not found.

f92bc4a

ruff

8604995

Fix browser_exe might be null, if not launching browser.

4f6d116

Run test_emrun suite sequentially. In order to do that, it must be mo…

e7794a5

…ved to its own file.

ruff

2dca116

juj commented Apr 28, 2026

View reviewed changes

sbc100 reviewed Apr 28, 2026

View reviewed changes

juj added 2 commits April 28, 2026 23:49

Adjust autolookup for EMTEST_BROWSER to avoid PATH

7abb9b5

shutil.which() in Firefox lookup.

4aed2bb

sbc100 approved these changes Apr 28, 2026

View reviewed changes

Fix EMTEST_BROWSER lookup.

61da1f9

juj enabled auto-merge (squash) April 28, 2026 21:14

juj disabled auto-merge April 28, 2026 21:47

juj merged commit a046aaf into emscripten-core:main Apr 28, 2026
28 of 30 checks passed

dschuff mentioned this pull request May 5, 2026

Add a WebGPU page to porting with a link to a tutorial #26862

Merged

	def detect_browser_processes():
	if not browser_exe:
	return # Running with --no-browser, we are not binding to a spawned browser.

	global current_browser_processes
	logv('First navigation occurred. Identifying currently running browser processes')
	running_browser_processes = list_processes_by_name(browser_exe)

	def pid_existed(pid):
	for proc in previous_browser_processes:
	if proc['pid'] == pid:
	return True
	return False

	for p in running_browser_processes:
	logv(f'Detected running browser process id: {p["pid"]}, existed already at emrun startup? {pid_existed(p["pid"])}')

	current_browser_processes = [p for p in running_browser_processes if not pid_existed(p['pid'])]

	if len(current_browser_processes) == 0:
	logv('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.')

Conversation

juj commented Apr 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sbc100 Apr 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

juj commented Apr 28, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

juj Apr 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

juj Apr 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

juj Apr 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

juj Apr 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

juj commented Apr 28, 2026 •

edited

Loading

sbc100 Apr 28, 2026 •

edited

Loading

juj Apr 28, 2026 •

edited

Loading

juj Apr 28, 2026 •

edited

Loading

juj Apr 28, 2026 •

edited

Loading

juj Apr 28, 2026 •

edited

Loading