Updated pex to 2.76.1

[sureshjoshi]

Precursor to #22946

Lockfile diff: 3rdparty/python/user_reqs.lock [python-default]
                                                                  
==                    Upgraded dependencies                     ==

  lia-web                        0.2.3        -->   0.3.1
  packaging                      25.0         -->   25.1.dev0
  pex                            2.73.1       -->   2.76.1
  python-multipart               0.0.20       -->   0.0.21
  soupsieve                      2.8          -->   2.8.1
  urllib3                        2.5.0        -->   2.6.2
                                                                  
==                      Added dependencies                      ==

  cross-web                      0.4.0

Also had to re-generate pytest due to packaging test failure:

Lockfile diff: 3rdparty/python/pytest.lock [pytest]
                                                                  
==                    Upgraded dependencies                     ==

  asttokens                      3.0.0        -->   3.0.1
  coverage                       7.10.7       -->   7.13.1
  execnet                        2.1.1        -->   2.1.2
  iniconfig                      2.1.0        -->   2.3.0
  ipython                        9.6.0        -->   9.8.0
  matplotlib-inline              0.1.7        -->   0.2.1
  packaging                      25.0         -->   25.1.dev0

[sureshjoshi]

In spite of 2.73.1 being the update that was referenced by the subsequent two bugs, I've just updated this here - as updating to 2.76.1 should have the same effect.

[sureshjoshi]

Part of the update of strawberry-graphql[fastapi]==0.284.1 and associated deps.

13:11:30.86 [WARN] /Users/sj/.cache/pants/pants_dev_deps/3db7049535f77bbd89235b7d4b164e4b4f0a581f.venv/lib/python3.11/site-packages/strawberry/fastapi/router.py:15: DeprecationWarning: The 'lia' package has been renamed to 'cross_web'. Please update your imports from 'from lia import ...' to 'from cross_web import ...'. The 'lia' package will be removed in a future version.
  from lia import HTTPException, StarletteRequestAdapter

See #22814

[sureshjoshi]

Lol, what a failed test

'foorl@ git+file://localhost
'foorl @ git+file://localhost

[sureshjoshi]

I knew my YOLO wouldn’t work… Will fix this when back at a machine.

[sureshjoshi]

The @ change was made in Requirements in the packaging library last month and has not been released yet:

pypa/packaging#935
pypa/packaging@6da1c50

Pex made this change in 2.74.2 (https://github.com/pex-tool/pex/releases/tag/v2.74.2)

The associated inconsistent code is in pip_requirements.py. We can either wait for the packaging release, or just fix/vendor it ourselves. Or, take a look at Pex's Requirements and see if they're type-compatible.

Just outlining my observations so I don't forget them when I come back to this later today.

[jsirois]

@sureshjoshi am I correct that you're updating packaging just to fix a test that does string comparisons of requirements (and is failing on irrelevant white space differences) instead of comparing parsed requirement objects, for example?

[sureshjoshi]

am I correct that you're updating packaging just to fix a test that does string comparisons of requirements (and is failing on irrelevant white space differences) instead of comparing parsed requirement objects, for example?

No.

[jsirois]

@sureshjoshi Ok, but I can't see why else. I'm sensitive to breaking folks with Pex, but spaces around @ in a direct reference requirement are purely cosmetic; so neither the Pex change nor the packaging change were bugfixes - they were both purely for cosmetics. The fact that Pants tests are sensitive to this is a red flag. IOW I broke Pants tests here, but this is the exceedingly rare case where I don't see Pex as having done something wrong.

There will be another coming up in 2.77.1 (@cburroughs filed #22988). That one I'm still not sure about. Pants is using PEX_PYTHON in a strange way that should never have worked and 2.77.1 will expose that, but I need to think harder / dig harder on that one to see if this should be considered a back-compat break or Pants getting away with murder for a long time.

[sureshjoshi]

Yeah, pex "broke Pants tests", but like... I don't consider that to be a true break either.

The tests were/are brittle. I was investigating what else would break when the next packaging update comes down that contains that change upstream, and why we actually had a set of tests that didn't test the functionality - but did string compares at multiple places in multiple ways. Each change I tested failed a different assertion in that function - all related to the same whitespace issue.

Pants is bogged down with brittle assertions (many of which I'm a contributor to), but the pex-related stuff has always been seamless, so I was documenting (mostly to myself) what I was finding along the way, as between holidays and CES, I knew I'd be swamped and unable to come back to this.

I had some more observations I (apparently) didn't post from later that day, about some tangential investigations. I'll post them if I can dig them up, but in short, that function has a bunch of assertions about internals that I don't think we shouldn't be checking at all.

[sureshjoshi]

Closed in favour of: #23003

Will update the requirements.txt and lockfile in a separate PR.