Update to SuiteSparse:GraphBLAS 8.2.1

[eriknw]

Also, hopefully fix #109 by changing release trigger to "published", not "created".

[alugowski]

Good catch! So the workflow was being triggered on release "create" but the upload was only enabled for "publish".

[eriknw]

Yup! I think "published" is the right one to use, which you reasonably did in the if condition:
https://docs.github.com/en/webhooks/webhook-events-and-payloads?actionType=published#release

New release coming up 🤞

[eriknw]

Why? 😭

 build/temp.macosx-11.0-arm64-cpython-38/suitesparse_graphblas._graphblas.c:570:10: fatal error: 'GraphBLAS.h' file not found

https://github.com/GraphBLAS/python-suitesparse-graphblas/actions/runs/6513170637/job/17692251398?pr=110#step:8:1615

[eriknw]

I see, the brew fetch failed in add_arm_to_libomp_dylib.sh. Retry should hopefully fix.

[eriknw]

@alugowski do you know how to add retries to brew commands? A quick search left me even more confused. Network in CI can be flaky, so it would be nice to make our commands more resilient.

[eriknw]

Even though it's a little slower and adds some noise, I really like building wheels for PRs!

[alugowski]

@alugowski do you know how to add retries to brew commands? A quick search left me even more confused. Network in CI can be flaky, so it would be nice to make our commands more resilient.

Looks like the --retry switch does that:

       --retry
              Retry if downloading fails or re-download if the checksum of a previously cached version no longer matches. Tries at most 5 times with exponential backoff.

We can also add a "|| exit 1" on the script invocation to fail fast. I can make a PR.

[alugowski]

Even though it's a little slower and adds some noise, I really like building wheels for PRs!

Agreed!

[alugowski]

I recently discovered you can add input options to manual workflow dispatch. So I experimented with such an input to control PyPI upload:

Would you like it to be added here? That way if a workflow run fails for an ephemeral reason then it can be rerun manually.

[alugowski]

@alugowski do you know how to add retries to brew commands? A quick search left me even more confused. Network in CI can be flaky, so it would be nice to make our commands more resilient.

#111

[alugowski]

I recently discovered you can add input options to manual workflow dispatch. So I experimented with such an input to control PyPI upload:

Would you like it to be added here? That way if a workflow run fails for an ephemeral reason then it can be rerun manually.

#112