libc++: std/utilities/charconv/charconv.msvc/test.pass.cpp fails after #91651 #113265
Comments
github-actions
bot
added
the
libc++
|
Ack, will look into it when I get a chance. |
|
Do you have any more information on how to replicate this (e.g. cmake arguments?) I'm not very familiar with the libcxx tests. |
|
https://github.com/llvm/llvm-project/actions/runs/11923013907/job/33283101979?pr=116149 is another CI run with the same error. |
|
Those are tests timing out. I disabled them under msan since they're seemingly just too slow to run under that sanitizer. |
|
This test remains flaky in Android. Note that we are seeing a failure, not a test timeout. |
And where would you be able to look that that failure? |
The failure message is in #1. If you want to reproduce the ToT Android LLVM toolchain build, you will need to first install the Android repo tool, then: $ mkdir llvm-toolchain && cd llvm-toolchain
$ repo init -u https://android.googlesource.com/platform/manifest -b llvm-toolchain
$ repo sync -c
$ toolchain/llvm_android/build.py --build-llvm-next --llvm-rev=main --no-build=windowsTowards the end of the build, libcxx tests will be ran, and there is an around 50% chance that the test would fail. |
|
Is there a way to skip the unrelated tests? Running the full suite with the provided command takes ~20 minutes on my workstation and I've yet to get the error. Being able to rebuild faster would greatly help in debugging. |
|
For the llvm-toolchain repro above, once cmake for stage2 is complete, you can ctrl-c out of the stage2 build and run |
|
Alright, I have managed to replicate. The randomized test gave me a different failing test case so I'll talk them through separately. Also: The only part that's important is the "round-trip failed for
{
const char* s = "1.247130438231756e+30";
const char* last = s + std::strlen(s) - 1;
double value = 0.25;
std::from_chars_result result = std::from_chars(s, last, value);
assert(result.ec == std::errc{});
assert(result.ptr == last);
assert(value == 1.247130438231756e+30);
} |
|
Some of the other failing cases we observed: 0x3F92EF16EEE175FB |
|
Update: I was wrong about that first case being On the more positive side, I can consistently replicate the error by manually setting |
|
Current Status: Details: I have some other work with more pressing deadlines, so this is going on the backburner for a bit, but if you want to try yourself, here's the diff I used for testing: The result I get on i386 when running The correct values are: (The last bit of the expanded float mantissa is lower) |
After #91651, the above test fails in an internal Android builder when targeting i386-linux-gnu. Tagging @michaelrj-google.
The text was updated successfully, but these errors were encountered: