Closed jschwab closed 3 years ago
Is there a particular commit SHA I can test this out on?
Edit: Per slack conversation, it seems 9a0f870
will do nicely.
Plan is to insert "Unknown" for the various compiler information that would normally be in testhub.yml
. The downside here is that this now couples mesa_test
to knowing what is supposed to be in this file. I don't really see another way about it right now, but it's something we should keep in mind.
I have this working now, and it will submit a compilation failure after attempting to install 9a0f870
. However, it will now happily submit a compilation failure even if you never even tried to run install
. That is, a user could do mesa_test checkout && mesa_test submit --empty
, and it would submit as a compilation failure.
I wonder if we should have the first thing install
does is touch testhub.yml
, and then if that file is found, but empty, submit a compilation failure.
Yes, that's a great idea. Done in bda98f93.
This is now implemented in mesa_test
v1.0.4. If no testhub.yml
file is found, it barks at you and doesn't even try to submit. If one is found, but nothing is in it, it populates the compiler, compiler version, etc. with "Unknown" and carries on with submitting a compilation failure. For example, see Rummy's submission of 9a0f870
, which I artificially made work by creating an empty testhub.yml
file.
Such tricks should no longer be necessary after bda98f93
, which should always succeed in creating testhub.yml
at the beginning of installation.
In extreme cases (i.e., a failure in
const
,utils
, ormath
), the install process may fail to generate the file testhub.yml.mesa_test should know how to handle the absence of this file. It should report a compilation failure.
At present, it dies with