Port LeakSanitizer to emscripten #8711
Conversation
quantum5
force-pushed
the
lsan
branch
5 times, most recently
from
99e81fe to
af770e1
Compare
quantum5
force-pushed
the
lsan
branch
3 times, most recently
from
8a396b5 to
280803d
Compare
tests/test_other.py
Outdated
| def do_smart_test(self, source, name='test.cpp', literals=[], regexes=[], | ||
| emcc_args=[], run_args=[], assert_returncode=0): | ||
| shutil.copyfile(source, name) | ||
| run_process([PYTHON, EMCC, name] + emcc_args) |
There was a problem hiding this comment.
self.get_emcc_args() is better I think
There was a problem hiding this comment.
I don't think anything in test_other.py uses self.emcc_args.
There was a problem hiding this comment.
I recall that @kripken wants no tests in test_other to use self.emcc_args and the mechanism behind that.
There was a problem hiding this comment.
I believe that is incorrect. For self.set_setting() and self.emcc_args shouldn't work for all test subclasses. Thats the point of the common code in runner.py. @kripken can you confirm?
The set_setting/get_setting/emcc_args/build stuff in RunnerCore is all generic right?
|
|
||
|
|
||
| class libubsan_rt_wasm(SanitizerLibrary): | ||
| name = 'libubsan_rt_wasm' |
There was a problem hiding this comment.
Do these libraries need to end in _wasm? I think think that was only need when we have certain parts a library that were wasm only. In these cases I don't think its needed.
There was a problem hiding this comment.
In this case, the entire library is wasm only. I think we can wait until we get rid of fastcomp to get rid of the _wasm suffix.
There was a problem hiding this comment.
Also note that this is already in the code. I don't think we should change the name of all sanitizer libraries in this change.
There was a problem hiding this comment.
OK, lets do that as a followup perhaps.
|
|
||
|
|
||
| class libubsan_rt_wasm(SanitizerLibrary): | ||
| name = 'libubsan_rt_wasm' |
There was a problem hiding this comment.
OK, lets do that as a followup perhaps.
| } | ||
|
|
||
| void DoStopTheWorld(StopTheWorldCallback callback, void *argument) { | ||
| StopTheWorld(callback, argument); |
There was a problem hiding this comment.
Comment that this is a nop please!
4 participants
This seems sufficiently stable for an initial version.
The bullet, poppler, sqlite3 tests all run fine under LSan, without false positives. Leaks were introduced into sqlite3 and it was caught by LSan.
Limitations: