Fix tarball naming for CPU-only tarballs#65
Conversation
|
bot: build repo:eessi.io-2023.06-software instance:eessi-bot-mc-aws arch:zen2 |
|
New job on instance
|
|
New job on instance
|
| raise ValueError(f"Invalid environment variable name: {var_name}") | ||
| list_string = os.getenv(var_name, '[]') | ||
|
|
There was a problem hiding this comment.
I guess this is just so there's something to include in generated tarball?
There was a problem hiding this comment.
Yes, remove any changes to eb_hooks.py after testing...
Also, this removal of whitespace is some vim feature that @smoors recommended to me, I think XD
There was a problem hiding this comment.
(there's a reason I opened the PR in draft mode ;-) I'll put it on ready to review after removing this)
|
Closing in favor of #67 |
When introducing the new tarball naming to make sure tarball names were unique per GPU architecture, I did not realize that
${EESSI_ACCELERATOR_TARGET_OVERRIDE//\//-}would simply be completely empty for a CPU build, and thus the string formatting would be short on value. This caused tarball names to be incorrect, as they ended with-0.tar.gz, see e.g. https://github.com/EESSI/staging_bundles/pull/9 . That's problematic, since the bot assumes the item after the last-to be the timestamp. So, with the policy to only upload the last tarball every tarball would be uploaded, since all had identical timestamp (0) according to the bot.Fixed in this PR by making the number of formatting specs conditional on the accelerator override being non-empty.