| Issue |
152246
|
| Summary |
CI logs do not obviously point to error message
|
| Labels |
infrastructure
|
| Assignees |
|
| Reporter |
arsenm
|
The current CI system has poor UX for finding the point of failure. Yesterday I had a PR that was failing the clang build. I spent 15 minutes staring at the raw log output, and never found it.
First I click the "CI Checks / Build and Test xxx" links under "N failing checks"
<img width="908" height="314" alt="Image" src="" />
Then I get to this list of steps, of which only "Build and Test" is ever relevant.
<img width="1092" height="935" alt="Image" src="" />
This log output is still CI implementation detail boilerplate, the real failure will be under cmake or most likely the last ninja step. The last step simply has too much output, and the failure isn't at the bottom.
<img width="1082" height="536" alt="Image" src="" />
I usually see `ninja: build stopped: cannot make progress due to previous errors.`, but the actual failure is nowhere to be seen. I end up downloading the raw logs, and Ctrl-Fing for anything that looks like an error message (which is quite difficult, we have a lot of files with error and fail in the name)
Sometimes if I click the copilot link to "Explain Error", it finds the point in the giant log but most of the time that just produces an error. If I scroll to the bottom of the log, it looks like this:
<img width="1097" height="916" alt="Image" src="" />
It looks like the step that failed is `python generate_test_report_github.py`. A bit above you can see the `ninja: build stopped: cannot make progress due to previous errors.`, but then you have to really search to find wherever that happened.
In short, I should hardly ever need to download the raw logs to figure out what the error was
![]()
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
