fix(data): bind promoted constructor parameters

[ChrisRackauckas-Claude]

Ignore this draft until reviewed by @ChrisRackauckas.

What changed and why

Promoting constructors added by #76 use a union between the resolved self type and the singleton-bottom type. When no ordinary field binds every ADT parameter, the all-bottom dispatch path leaves one or more method type parameters unbound. This is observable both through Test.detect_unbound_args and, for partially inferred multi-parameter ADTs, as an UndefVarError during construction.

This change emits the broad union constructor only when ordinary fields bind every parameter. Otherwise a single self-reference keeps its exact constructor, while multiple self-references use disjoint anchored promotion methods: earlier arguments must be singleton-bottom, the anchor binds the ADT parameters, and later arguments accept either the resolved or singleton-bottom type.

Failing before

On unmodified main at dff7f45:

TMPDIR="$PWD/.tmp" julia +1.10 --project=. -e 'include("test/data/emit/singleton_promote.jl")'

generated constructors bind type parameters: Test Failed
Evaluated: isempty(Method[... Lst.Cons(...) where T ...])
Test Summary:                               | Fail  Total
generated constructors bind type parameters |    1      1

A standalone two-parameter Moshi reproducer on the same base produced:

unbound_count=1
construction_error=UndefVarError: UndefVarError(:S)

Passing after

TMPDIR="$PWD/.tmp" julia +1.10 --project=. -e 'using Pkg; Pkg.test()'
data            | 324 passed
match           | 140 passed
derive          |  30 passed
perf regression |   5 passed
Testing Moshi tests passed

TMPDIR="$PWD/.tmp" julia +1.12 --project=. -e 'include("test/data/emit/singleton_promote.jl")'
focused generated-constructor tests | 17 passed

TMPDIR="$PWD/.tmp" julia +1.13 --project=. -e 'include("test/data/emit/singleton_promote.jl")'
focused generated-constructor tests | 17 passed

The focused tests include zero detected unbound parameters and zero detected ambiguities.

Downstream verification

SymbolicUtils was tested with its Moshi cap removed and this branch developed locally:

TMPDIR="$PWD/.tmp" GROUP=QA julia +1.10 --project=. -e 'using Pkg; Pkg.test()'
SciMLTesting QA | 18 passed, 18 total
AdjView JET     | 37 passed, 37 total
Testing SymbolicUtils tests passed

Formatting was applied with JuliaFormatter 2.13.0 to both changed files. git diff --check upstream/main passed. git diff upstream/main | typos --config .tmp/typos.toml - passed with a temporary exact-identifier allowance for the pre-existing internal name is_inferrable.

Review considerations

The anchored signatures preserve promotion when a concrete self-reference exists without leaving an all-bottom union branch unbound. A partially inferred multi-parameter all-bottom call now fails dispatch with MethodError instead of entering an invalid method and raising UndefVarError; choosing values for absent type parameters is intentionally left outside this focused fix.

Documentation was not built because this changes no public API or documentation. Nightly Julia and non-Linux platforms were not run locally.

Links

• Original singleton-promotion issue: Singleton with parameters #34
• Original singleton-promotion implementation: fix(data): promote parametric singletons in variant constructors #76
• Downstream SymbolicUtils change: Fix strict QA method ambiguities JuliaSymbolics/SymbolicUtils.jl#1034

🤖 Generated with Codex (version unknown; model: unknown; session: local session ID 01a04608-d980-7e61-88d9-f6486b0c1916)

[vercel]

@ChrisRackauckas is attempting to deploy a commit to the roger-luo's projects Team on Vercel.

A member of the Team first needs to authorize it.

[github-actions]

Benchmark Results (Julia v1)

Time benchmarks

	main	b6d7f41...	main / b6d7f41...
adt_transform/transform n=100	6.77 ± 0.17 μs	7.09 ± 0.15 μs	0.955 ± 0.031
adt_transform/transform n=1000	0.0668 ± 0.0011 ms	0.0689 ± 0.00083 ms	0.968 ± 0.02
linked_list/sum n=100	0.471 ± 0.01 μs	0.471 ± 0.001 μs	1 ± 0.021
linked_list/sum n=1000	5.12 ± 0.031 μs	5.18 ± 0.061 μs	0.988 ± 0.013
time_to_load	0.0774 ± 0.0015 s	0.079 ± 0.0017 s	0.979 ± 0.028

Memory benchmarks

	main	b6d7f41...	main / b6d7f41...
adt_transform/transform n=100	0.204 k allocs: 6.34 kB	0.204 k allocs: 6.34 kB	1
adt_transform/transform n=1000	2.02 k allocs: 0.0619 MB	2.02 k allocs: 0.0619 MB	1
linked_list/sum n=100	0 allocs: 0 B	0 allocs: 0 B
linked_list/sum n=1000	0 allocs: 0 B	0 allocs: 0 B
time_to_load	0.149 k allocs: 11.1 kB	0.149 k allocs: 11.1 kB	1

[ChrisRackauckas-Claude]

Benchmark red-team result: the first 117-line implementation showed a repeatable ~7% package load-time regression, so I narrowed the generator change to 48 added lines and removed the extra partial-inference machinery. The final benchmark at #92 (comment) reports time_to_load 77.4 ± 1.5 ms on main versus 79.0 ± 1.7 ms on this branch (main / branch = 0.979 ± 0.028) with identical allocations. The benchmark workflow passed.

The normal CI workflow is awaiting maintainer approval for this forked PR at https://github.com/Roger-luo/Moshi.jl/actions/runs/33164273641; it has not executed or failed.

[ChrisRackauckas]

@Roger-luo what do you think?

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
moshi-jl	Ignored	Preview	Sep 21, 2026 4:39pm UTC

[Roger-luo]

Thanks. Sorry, I was on vacation and was not checking GitHub.

[codecov]

Codecov Report

❌ Patch coverage is 96.15385% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 91.99%. Comparing base (dff7f45) to head (b6d7f41).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/data/emit/cons.jl	96.15%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #92      +/-   ##
==========================================
+ Coverage   91.91%   91.99%   +0.07%     
==========================================
  Files          43       43              
  Lines        1732     1748      +16     
==========================================
+ Hits         1592     1608      +16     
  Misses        140      140              

☔ 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.