ctest -C Debug fails on vs2022 (miniscript

maflcko commented at 5:25 pm on April 24, 2025: member

cmd: ctest --output-on-failure --stop-on-failure -j $NUMBER_OF_PROCESSORS -C Debug

The full log and config is in the comment below: #32341 (comment). Which was motivated by #32339 attempting to use a windows debug build.

maflcko added the label Windows on Apr 24, 2025

maflcko added the label Tests on Apr 24, 2025

maflcko renamed this:
~~ctest -C Debug fails on vs2022 with test_bitcoin-qt (Exit code 0xc0000135)~~
ctest -C Debug fails on vs2022 (Exit code 0xc0000135)
on Apr 24, 2025

maflcko closed this on Apr 24, 2025

hebasto commented at 9:05 pm on April 24, 2025: member

FWIW, here is the demo branch where the native Windows CI job has been switched to the “Debug” configuration.

CI logs: https://github.com/hebasto/bitcoin/actions/runs/14650416249.

maflcko commented at 5:54 am on April 25, 2025: member

Thx. Though, now it says:

0The following tests FAILED:
1	 60 - miniscript_tests (SEGFAULT)
2Errors while running CTest

This raises two questions:

I’d say for the CI we want to reset /Ob0 /Od back to /O2 /Ob1 to avoid slowness.
I guess this is not really a segfault/bug in the code, but rather a stack overflow, due to the debug config eating more memory?

hebasto commented at 6:40 am on April 25, 2025: member

Thx. Though, now it says:
0The following tests FAILED:
1	 60 - miniscript_tests (SEGFAULT)
2Errors while running CTest
I guess this is not really a segfault/bug in the code, but rather a stack overflow, due to the debug config eating more memory?

I’m not sure. I can reproduce the issue locally and it refers to an exception:

0    Start 58: miniscript_tests
11/1 Test [#58](/bitcoin-bitcoin/58/): miniscript_tests .................***Exception: SegFault 51.83 sec

Will try to narrow it down.

maflcko commented at 7:51 am on April 25, 2025: member

A third question would be, why the fuzz output is https://github.com/hebasto/bitcoin/actions/runs/14650416249/job/41114604643#step:14:60: subprocess timed out: Currently only libFuzzer is supported. A timeout should not happen for this process call, and the expected outcome should be an immediate stderr of test/fuzz/fuzz.cpp:269 main: Assertion 'read_file(input_path, buffer)' failed. Error processing input "-help=1" and an ignored exit code.

maflcko renamed this:
~~ctest -C Debug fails on vs2022 (Exit code 0xc0000135)~~
ctest -C Debug fails on vs2022 (miniscript_tests (SEGFAULT))
on Apr 25, 2025

fanquake commented at 8:46 am on April 25, 2025: member

Re-opening, given it looks like there are at least two cmake/windows bugs here?

fanquake reopened this on Apr 25, 2025

hebasto commented at 12:43 pm on April 25, 2025: member

Thx. Though, now it says:
0The following tests FAILED:
1	 60 - miniscript_tests (SEGFAULT)
2Errors while running CTest
I guess this is not really a segfault/bug in the code, but rather a stack overflow, due to the debug config eating more memory?

Fixed in #32349.

l0rinc referenced this in commit bd89f7eff0 on Apr 25, 2025

hodlinator commented at 10:05 pm on April 26, 2025: contributor

Was curious how this issue was introduced. Seems like it’s existed for some time but was found by #32339 momentarily switching some Windows CI to use Debug configs. Would it be worth adding to the description instead of just “ref: #32339”?

l0rinc referenced this in commit 28f7f3f58d on Apr 27, 2025

fanquake commented at 9:26 am on April 28, 2025: member

but was found by #32339 momentarily

It was actually first pointed out here, #31367 (comment), by @dergoegge.

maflcko commented at 9:35 am on April 28, 2025: member

I’d say for the CI we want to reset /Ob0 /Od back to /O2 /Ob1 to avoid slowness. @hebasto If you don’t mind, could you update your CI branch with this. If not, that’s fine also and I’ll try it myself.

hebasto commented at 9:45 am on April 28, 2025: member

I’d say for the CI we want to reset /Ob0 /Od back to /O2 /Ob1 to avoid slowness.

@hebasto If you don’t mind, could you update your CI branch with this. If not, that’s fine also and I’ll try it myself.

Will do.

hebasto commented at 10:41 am on April 28, 2025: member

I’d say for the CI we want to reset /Ob0 /Od back to /O2 /Ob1 to avoid slowness.

@hebasto If you don’t mind, could you update your CI branch with this. If not, that’s fine also and I’ll try it myself.

~~https://github.com/hebasto/bitcoin-core-nightly/actions/runs/14705754388/job/41265724818~~ https://github.com/hebasto/bitcoin-core-nightly/actions/runs/14707592382/job/41271542519

l0rinc referenced this in commit b80d0bdee4 on Apr 28, 2025

achow101 referenced this in commit 2d5b424414 on Apr 29, 2025

achow101 closed this on Apr 29, 2025

maflcko commented at 5:17 pm on April 30, 2025: member

A third question would be, why the fuzz output is https://github.com/hebasto/bitcoin/actions/runs/14650416249/job/41114604643#step:14:60: subprocess timed out: Currently only libFuzzer is supported. A timeout should not happen for this process call, and the expected outcome should be an immediate stderr of test/fuzz/fuzz.cpp:269 main: Assertion 'read_file(input_path, buffer)' failed. Error processing input "-help=1" and an ignored exit code.

Looks like the assertions fails, but the program never prints anything, or aborts:

https://github.com/maflcko/bitcoin-core-with-ci/actions/runs/14759346479/job/41435857332#step:13:58

0
1Run "$BITCOINFUZZ" -help=1
2Wed, 30 Apr 2025 16:36:03 GMT D:\a\bitcoin-core-with-ci\bitcoin-core-with-ci\src\test\fuzz\fuzz.cpp:269 main: Assertion `read_file(input_path, buffer)' failed.
3Wed, 30 Apr 2025 17:02:52 GMT Error: The operation was canceled.

I guess this means the std::signal(SIGABRT, signal_handler) does something different.

Possibly a debugger is automatically attached?

AlnetharyT referenced this in commit 5a1669f648 on May 2, 2025

hebasto commented at 6:43 pm on May 2, 2025: member

A third question would be, why the fuzz output is https://github.com/hebasto/bitcoin/actions/runs/14650416249/job/41114604643#step:14:60: subprocess timed out: Currently only libFuzzer is supported. A timeout should not happen for this process call, and the expected outcome should be an immediate stderr of test/fuzz/fuzz.cpp:269 main: Assertion 'read_file(input_path, buffer)' failed. Error processing input "-help=1" and an ignored exit code.

Looks like the assertions fails, but the program never prints anything, or aborts:

https://github.com/maflcko/bitcoin-core-with-ci/actions/runs/14759346479/job/41435857332#step:13:58
0
1Run "$BITCOINFUZZ" -help=1
2Wed, 30 Apr 2025 16:36:03 GMT D:\a\bitcoin-core-with-ci\bitcoin-core-with-ci\src\test\fuzz\fuzz.cpp:269 main: Assertion `read_file(input_path, buffer)' failed.
3Wed, 30 Apr 2025 17:02:52 GMT Error: The operation was canceled.
I guess this means the std::signal(SIGABRT, signal_handler) does something different.

Possibly a debugger is automatically attached?

It is actually a message box:

The following patch helps:

0--- a/src/test/fuzz/fuzz.cpp
1+++ b/src/test/fuzz/fuzz.cpp
2@@ -245,6 +245,7 @@ int main(int argc, char** argv)
3         test_one_input(buffer);
4         return 0;
5     }
6+    _set_abort_behavior(0, _WRITE_ABORT_MSG);
7     std::signal(SIGABRT, signal_handler);
8     const auto start_time{Now<SteadySeconds>()};
9     int tested = 0;

For more details, see Microsoft docs:

ctest -C Debug fails on vs2022 (miniscript_tests (SEGFAULT)) #32341