You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Fix CI diagnostics for failed bazel builds on Windows (#42006)
### What does this PR do?
CI diagnostics for failed `bazel` builds on Windows didn't work as
expected (see #41844 for more details).
The change consists in:
- removing the pseudo-shebang that was preventing from resolving paths
relative to the source script's location,
- making sure the output is UTF-8 (`chcp 65001`) to echo non-ASCII
characters from dumped contents,
- using the right `bazel` startup flag (`--local_startup_timeout_secs`)
to limit the total retry loop to 30s,
while making the retries more frequent (`--connect_timeout_secs`) inside
the loop,
- no longer assuming `$CI_PROJECT_DIR` resolves to the workspace's root,
- payload: fixing/discovering base location for server logs which happen
to be under a subdirectory whose name is a hash (`jgkidtko` on my VM).
### Motivation
Really get actionable outputs for addressing `FATAL: couldn't connect to
server` errors.
### Describe how you validated your changes
Reproduced CI layout locally and fixed the script to make it provide
with available info on `bazel` build failures.
### Additional Notes
This is of course temporary until we figure out what's preventing the
`bazel` CLI from contacting the server.
0 commit comments