Move to igwn-ligolw

[titodalcanton]

Move from python-ligo-lw to the recently released and more modern igwn-ligolw.

Standard information about the request

This is a depencency change.
This change should have no impact on any functionality or result.
This change will drop the dependency on python-ligo-lw and introduce a depencency on igwn-ligolw.

Motivation

igwn-ligolw is more actively maintained and supports modern Python versions and more architectures than python-ligo-lw.

Contents

Mainly replacing imports, but the transition from python-ligo-lw 1.8.3 to igwn-ligolw 2.0.0 has the side effect of introducing a few breaking changes which never made it to a python-ligo-lw release yet, as well as a couple deprecation warnings, so those are handled as well. We can also further simplify tox.ini, consistently with #4857.

Links to any issues or associated PRs

This is essentially a counterpart to #4999. Improves #4857.

Testing performed

Ran tox and a few of the tests manually on a local machine, but I am relying on the CI for now.

Additional notes

I am not sure I want to push this to production codes yet, at least not before more significant testing is done with workflows and GraceDB interactions. In particular, I think we should create a release branch for PyGRB's O4a analysis before merging this (@pannarale). So I am applying a "on hold" label.

• The author of this pull request confirms they will adhere to the code of conduct

[titodalcanton]

Started to do some more thorough tests. Noting down a few more deprecations to address:

UserWarning: igwn_ligolw.utils.process.register_to_xmldoc() is deprecated.  use igwn_ligolw.ligolw.Document.register_process() instead.
UserWarning: igwn_ligolw.lsctables.New() is deprecated.  use igwn_ligolw.Table.new() instead.
UserWarning: igwn_ligolw.lsctables.use_in() is deprecated.  simply importing the lsctables module automatically enables the custom table support features it provides.

Edit: these have been addressed now.

[titodalcanton]

I made a test run of the PyCBC Live example uploading to GraceDB test and I do not see particular issues, so I am gaining confidence about this.

[titodalcanton]

I now repeated the PyCBC Live example test after the latest commits, and I have uploaded two sets of candidates from before and after this PR. I have compared the resulting candidates from two of the injections side-by-side on GraceDB. I have checked both the original and SNR-optimized events. The things that should be identical are identical. So I think this is safe enough to merge, especially if we do not plan to cherry-pick this into the offline analyses yet.

[spxiwh]

This is failing because mac OS + python 3.12 cannot build ligo-segments. I'm guessing that igwn-ligolw is bringing in the old ligo-segments (which can't build, thereby raising the very issue it was intended to solve).

[spxiwh]

Actually, it looks like it's lalsuite that's pulling in ligo-segments ... And we can't yet use the new lalsuite because of the issues around ROM waveforms.

[titodalcanton]

I also think this is due to lalsuite, not igwn-ligolw.

[spxiwh]

Hmm, this may mean that this patch is on hold until a new lalsuite release that does not use ligo-segments?

[titodalcanton]

@spxiwh yes, I think it is reasonable to wait for that, since it has been in progress for a while