backend: coerce all-numpy operands in promote_scalars under a forced backend

[jobovy]

What

promote_scalars passed its inputs through unchanged when none was a backend array to anchor on (ref is None), assuming "the namespace's functions handle scalars." That holds for jax but not torch: torch.cos/sqrt/… reject numpy.float64 / python floats. So under a forced torch backend, every promote_scalars caller — the coords transforms (cyl_to_rect/rect_to_cyl/…) and OblateStaeckelWrapperPotential — crashed on all-numpy inputs:

TypeError: cos(): argument 'input' (position 1) must be Tensor, not numpy.float64
   at galpy/util/coords.py:1164  (cyl_to_rect)

This is the root cause behind the migrated RotateAndTilt / Offset / OblateStaeckel / Kuzmin wrapper torch failures (they route coordinates through these transforms). The fix coerces the operands via coerce_coords in that branch:

 ref = next((v for v in vals if is_backend_array(v)), None)
 if ref is None:
-    # ... pass through, the namespace's functions handle scalars
-    return vals
+    # torch's functions REJECT numpy.float64/python floats, so coerce instead
+    return coerce_coords(xp, *vals)

One spot in the central coercion helper, rather than patching each transform.

Safety (adversarially reviewed, two independent passes, CPU-forced)

• numpy byte-identical — the xp is numpy guard short-circuits first; verified object-identical pass-through and identical SHA-256 output hashes for cyl_to_rect/rect_to_cyl/cyl_to_spher/Rz_to_uv/uv_to_Rz over a 2000-point grid.
• jax value-identical — jax already tolerated raw scalars; the coerced float64 values match numpy to 0.0 and autodiff still flows (jax.grad through a coords transform is finite). jax failure sets are byte-identical (functional no-op for jax).
• No regression — 1663 backend tests pass; test_coords torch 20→9 failures (fixes 11), test_quantity torch 110→99 (fixes 11), all remaining are strict subsets; OblateStaeckelWrapper unaffected.
• No caller-reliance break — the one by-reference caller (integrateFullOrbit.py) pins xp=numpy, hitting the guard; no caller depends on the old pass-through returning a raw scalar.
• Also fixes a latent mixed numpy-array + scalar torch bug (the array was anchored-on but itself left un-coerced).

Tests

New tests/test_backend_coerce.py: numpy object-identity pass-through; forced-jax/torch coercion to backend float64 (the fixed branch); the anchored path unchanged; and a coords-transform integration check (fails at coords.py:1164 without the fix). Lives in the test_backend* coverage shard.

File-disjoint from #990 (Potential.py) and the merged #989. Part of the torch potential burndown; the per-potential _anchor fixes (PowerSpherical/SpiralArms/EllipticalDisk/…) are a separate themed PR.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 0% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 46.87%. Comparing base (509ea17) to head (f6ecdca).
⚠️ Report is 1 commits behind head on feat/backends.

Files with missing lines	Patch %	Lines
galpy/backend/_namespaces.py	0.00%	4 Missing ⚠️
galpy/backend/_coerce.py	0.00%	1 Missing ⚠️

❗ There is a different number of reports uploaded between BASE (509ea17) and HEAD (f6ecdca). Click for more details.

HEAD has 26 uploads less than BASE

Flag	BASE (509ea17)	HEAD (f6ecdca)
	28	2

Additional details and impacted files

@@                Coverage Diff                 @@
##           feat/backends     #991       +/-   ##
==================================================
- Coverage          99.93%   46.87%   -53.06%     
==================================================
  Files                254      254               
  Lines              39836    40668      +832     
  Branches             838      843        +5     
==================================================
- Hits               39810    19064    -20746     
- Misses                26    21604    +21578     

☔ 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.

[github-actions]

All-backend test status (jax / torch)

Commit b2a3704f6159e5cddc839c0dcf7c8c3ba21ac956

Green is achieved via the checked-in xfail-ledger (tests/backend_xfail.txt, applied xfail(strict=False)), so the metric to watch is the shrinking xfail count (burndown), not a raw pass count. A FAIL/ERR is an un-ledgered regression (reds the run). Because the ledger is non-strict, a now-passing ledgered test is a plain pass here (no per-push XPASS); burndown candidates -- in both directions -- are surfaced by the scheduled regen run, which rewrites the ledger from real outcomes. deferred is a separate burndown: tests skipped because they are unrunnable under the backend until the port is vectorized (see tests/backend_slow_skip.txt), e.g. the jax spherical-DF sampling/quadrature tests pending the Track F DF migration.

Overall: jax: 1058 passed · 264 xfail · 725 deferred | torch: 998 passed · 1039 xfail · 1 deferred · 6 FAIL/ERR

Ledger size: 2357 entries (jax=284, torch=2073).

Test shard	jax	torch
actionAngle	✅ 112 pass · 89 xfail	✅ 37 pass · 164 xfail
sphericaldf	✅ 164 pass · 26 xfail · 28 deferred	✅ 8 pass · 210 xfail
conversion + util + misc	✅ 87 pass · 4 xfail · 1 deferred	✅ 44 pass · 48 xfail
potential + scf + multipole	— (no result)	— (no result)
quantity + coords	✅ 287 pass · 49 xfail	✅ 228 pass · 108 xfail
orbit (energy/Jacobi + from_name)	✅ 0 pass · 0 xfail · 115 deferred	✅ 71 pass · 44 xfail
orbit + orbits (main)	✅ 0 pass · 0 xfail · 578 deferred	✅ 341 pass · 234 xfail
evolveddiskdf	✅ 35 pass · 0 xfail	✅ 34 pass · 1 xfail
jeans + dynamfric	✅ 17 pass · 2 xfail · 2 deferred	✅ 7 pass · 13 xfail · 1 deferred
qdf + pv2qdf + streamgapdf_impulse + noninertial	✅ 57 pass · 75 xfail · 1 deferred	✅ 15 pass · 118 xfail
streamgapdf	✅ 28 pass · 2 xfail	✅ 28 pass · 2 xfail
diskdf	✅ 129 pass · 0 xfail	✅ 116 pass · 13 xfail
streamdf + streamspraydf + streamTrack	✅ 142 pass · 17 xfail	❌ 69 pass · 84 xfail · 6 FAIL/ERR

Per-shard counts

Test shard	backend	pass	xfail	deferred	XPASS	fail	error
actionAngle	jax	112	89	0	0	0	0
actionAngle	torch	37	164	0	0	0	0
sphericaldf	jax	164	26	28	0	0	0
sphericaldf	torch	8	210	0	0	0	0
conversion + util + misc	jax	87	4	1	0	0	0
conversion + util + misc	torch	44	48	0	0	0	0
potential + scf + multipole	jax	—	—	—	—	—	—
potential + scf + multipole	torch	—	—	—	—	—	—
quantity + coords	jax	287	49	0	0	0	0
quantity + coords	torch	228	108	0	0	0	0
orbit (energy/Jacobi + from_name)	jax	0	0	115	0	0	0
orbit (energy/Jacobi + from_name)	torch	71	44	0	0	0	0
orbit + orbits (main)	jax	0	0	578	0	0	0
orbit + orbits (main)	torch	341	234	0	0	0	0
evolveddiskdf	jax	35	0	0	0	0	0
evolveddiskdf	torch	34	1	0	0	0	0
jeans + dynamfric	jax	17	2	2	0	0	0
jeans + dynamfric	torch	7	13	1	0	0	0
qdf + pv2qdf + streamgapdf_impulse + noninertial	jax	57	75	1	0	0	0
qdf + pv2qdf + streamgapdf_impulse + noninertial	torch	15	118	0	0	0	0
streamgapdf	jax	28	2	0	0	0	0
streamgapdf	torch	28	2	0	0	0	0
diskdf	jax	129	0	0	0	0	0
diskdf	torch	116	13	0	0	0	0
streamdf + streamspraydf + streamTrack	jax	142	17	0	0	0	0
streamdf + streamspraydf + streamTrack	torch	69	84	0	0	6	0