Skip to content

⬆️ dep-bump: cap unxt below 2.0 - #800

Merged
nstarman merged 1 commit into
GalacticDynamics:mainfrom
nstarman:claude/pin-unxt-below-2
Aug 8, 2026
Merged

⬆️ dep-bump: cap unxt below 2.0#800
nstarman merged 1 commit into
GalacticDynamics:mainfrom
nstarman:claude/pin-unxt-below-2

Conversation

@nstarman

@nstarman nstarman commented Aug 8, 2026

Copy link
Copy Markdown
Contributor

galax declares unxt>=1.11.2 with no upper bound, but has never been
resolved or tested against unxt 2.x — the lock has only ever held a 1.x.

Nothing currently declares that boundary. uv does consider unxt 2.0.0
during resolution (it is compatible on requires-python) and backtracks to
1.11.x for transitive reasons. That is incidental rather than a guarantee, and
it disappears as soon as the surrounding constraints shift — a coordinax bump,
a new dependency, an --upgrade.

This makes the tested boundary explicit. The resolved version is unchanged
— 1.11.2, as before — so the diff is metadata only:

- { name = "unxt", specifier = ">=1.11.2" },
+ { name = "unxt", specifier = ">=1.11.2,<2" },

What it guards against

unxt is making unit-system repr/str round-trippable
(GalacticDynamics/unxt#855):

before after
repr unitsystem(kpc, …) or LengthUnitSystem(length=Unit("km")) unitsystem('kpc', 'Myr', 'solMass', 'rad')
str LTMAUnitSystem(length, time, mass, angle) unitsystem(kpc, Myr, solMass, rad)

That rewrites the LTMAUnitSystem( length=Unit("kpc"), ...) form which 32
doctest lines here pin
— 18 in _interop/galax_interop_gala/potential.py, 12
in _interop/galax_interop_galpy/potential.py, and one each in
potential/_src/builtin/burkert.py and potential/_src/params/field.py.

Those updates belong with the port to unxt 2.x rather than ahead of it, since
2.0 already carries its own breaking changes (no-arg unitsystem() is now
dimensionless, unit-system identity is order-independent, Python 3.11 dropped).

Lift this cap in the PR that does that port.

Verification

uv lock produces no resolved-version change. tests/smoke 8 passed;
tests/unit 1032 passed, 170 skipped, 14 xfailed.

galax declares `unxt>=1.11.2` with no upper bound, but has never been resolved
or tested against unxt 2.x -- the lock has only ever held a 1.x. Today nothing
declares that boundary: `uv` considers unxt 2.0.0 during resolution (it is
compatible on `requires-python`) and backtracks to 1.11.x for transitive
reasons. That is incidental, not a guarantee, and it disappears the moment the
surrounding constraints shift.

Make the tested boundary explicit. The resolved version is unchanged --
1.11.2, as before -- so this is metadata only; `uv.lock` records the new
specifier and nothing else.

There is a concrete change coming that this guards against. unxt is making unit
system `repr`/`str` round-trippable
(GalacticDynamics/unxt#855), which rewrites the
`LTMAUnitSystem( length=Unit("kpc"), ...)` form that 32 doctest lines here pin,
18 of them in `_interop/galax_interop_gala/potential.py` and 12 in
`..._galpy/potential.py`. Those updates belong with the port to unxt 2.x, not
ahead of it.

Lift this cap in the PR that does that port.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@codecov

codecov Bot commented Aug 8, 2026

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.86%. Comparing base (e6ef7d5) to head (d0fea38).

Additional details and impacted files
@@           Coverage Diff           @@
##             main     #800   +/-   ##
=======================================
  Coverage   95.86%   95.86%           
=======================================
  Files         160      160           
  Lines        6046     6046           
=======================================
  Hits         5796     5796           
  Misses        250      250           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

@nstarman nstarman added this to the v1.0.0 milestone Aug 8, 2026
@nstarman
nstarman merged commit 9c1342d into GalacticDynamics:main Aug 8, 2026
16 checks passed
@nstarman
nstarman deleted the claude/pin-unxt-below-2 branch August 8, 2026 15:55
nstarman added a commit that referenced this pull request Aug 14, 2026
Bumps quax's floor from 0.3.2 to 0.4.3. No code changes needed --
quax.quaxify's public signature is unchanged across this range, so
this is a pure version-floor bump.

unxt>=2.0 and equinox>=0.13.8 were also considered as part of this
sweep, but:

- unxt 2.0 is not usable yet: coordinax v0.23.3 (latest, and the
  ceiling we're holding to) was never updated for unxt 2.0's
  Quantity/BareQuantity rename, so basic vector construction with
  range-checking crashes (`eqx.error_if` sees a `Quantity` where it
  expects a raw JAX array) and other codepaths hit
  `AttributeError: type object 'Quantity' has no attribute
  'type_parameter'`. 154 doctests and 115 tests fail with unxt>=2.0
  installed against coordinax 0.23.3. The existing `unxt>=1.11.2,<2`
  cap (#800) stays until coordinax ships unxt-2.0 support.
- equinox was already at >=0.13.8 on main; nothing to bump.

Verified: full test suite (1080 passed) and doctests (4964 passed)
pass with quax 0.4.3 alone.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
nstarman added a commit that referenced this pull request Aug 14, 2026
Bumps quax's floor from 0.3.2 to 0.4.3. No code changes needed --
quax.quaxify's public signature is unchanged across this range, so
this is a pure version-floor bump.

unxt>=2.0 and equinox>=0.13.8 were also considered as part of this
sweep, but:

- unxt 2.0 is not usable yet: coordinax v0.23.3 (latest, and the
  ceiling we're holding to) was never updated for unxt 2.0's
  Quantity/BareQuantity rename, so basic vector construction with
  range-checking crashes (`eqx.error_if` sees a `Quantity` where it
  expects a raw JAX array) and other codepaths hit
  `AttributeError: type object 'Quantity' has no attribute
  'type_parameter'`. 154 doctests and 115 tests fail with unxt>=2.0
  installed against coordinax 0.23.3. The existing `unxt>=1.11.2,<2`
  cap (#800) stays until coordinax ships unxt-2.0 support.
- equinox was already at >=0.13.8 on main; nothing to bump.

Verified: full test suite (1080 passed) and doctests (4964 passed)
pass with quax 0.4.3 alone.

Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
nstarman added a commit that referenced this pull request Aug 16, 2026
`coordinax>=0.23.3` has no upper bound. That is safe only by accident right
now -- 0.23.3 is the latest release, so there is nothing above the floor to
resolve to -- and it stops being safe the moment coordinax 0.24 lands, on an
already-published galax wheel that cannot be changed.

coordinax is 0.x and moves with galax: the floor has been raised on every
release since 0.23.0 (#769, #774, #784, #792). #738 already capped it once, at
`>=0.20.1,<0.21`, for an incompatible release; the cap was dropped in #744 when
the floor moved past it. `unxt` carries `<2` for the same reason, and is
currently earning it.

Resolution is unchanged -- coordinax stays at 0.23.3 -- so this only edits the
recorded specifier in the lock, matching #800.

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant