fix(language): a macro template nothing calls is held to every rule a call site is

[FBumann]

Prompt: Review the packages code as a whole. Do you see room for simplififcation? Or reuction of concepts? … Lets tackle that. With stacked PRs.

Note

The following content was generated by AI.

_check_template_names was a second resolver that checked only names. It is deleted. The resolver now checks a template, with the template's formals left bare. So a template now gets every refusal that a call site gets. Stacked on #627.

Method, gate output, alternatives

What this changes

• _Resolver has a formals field, and resolve_expression(..., formals=) passes it in. A formal stays bare in a value, edge=, a dimension kwarg, dual(), and by= with its column roles.
• validation._check_template_names (69 lines) is deleted.
• The listing of unknown names still puts the formals first, and dual(), over= and by= say "or a formal of this macro".

Found in review, fixed in the second commit

• A formal along= beside by=/within=, such as shift(x, along=d, offset=1, by=tg, within=g), was handed to the partition as a dimension and refused as 'tg' has no key column over 'd', even with a valid call. A formal along= now builds no partition, and the call site does.
• sum(x, by=nope, over=a, into=b) with a and b formals loaded, because the formal columns sent the call back before the relation was read. The relation is read first.
• The over= and by= refusals had lost "or a formal of this macro".

The five tests for these fail on the first commit and pass on the second.

What the language now refuses

A template that nothing calls now gets the same refusals as a call site. Before, these passed load:

• A str or bool parameter used as a value (x * tag).
• A relation used as a value, and a dimension used as a value.
• A column that the relation does not have (sum(x, by=lk, over=nope, into=h)).
• over=1, and edge='nope'.
• A list of relations in by=. It now gets the "one call reads one table" refusal, not "does not name a relation" for the first name.

Test coverage moved: a-typo-in-a-relation-list is now a-list-of-relations and asserts the resolver's message.

Tests

• test_a_template_is_held_to_the_rules_a_call_site_is: both cases fail on the base branch and pass here.
• test_a_formal_stands_where_a_call_site_will_bind_it: one case for each guard. I deleted each guard in turn and ran the suite:
  • The guard in _arith: 20 tests fail.
  • _edge: an-edge fails.
  • _dual: a-constraint fails.
  • The relation guard: both relation cases fail.

Gates

• pixi run lint: green.
• pixi run test: 1560 passed on the first commit; 1541 with tests/test_docs.py deselected on the second, where the docs dependencies were absent.
• docs-build and compile-tex did not run: the session proxy blocks their downloads.

🤖 Generated with Claude Code

https://claude.ai/code/session_017XwKgY5wZXv1bKgkfCW2q1

---

Generated by Claude Code

[read-the-docs-community]

Documentation build overview

📚 math-spec | 🛠️ Build #34716545 | 📁 Comparing 89335fc against latest (f9cb305)

  🔍 Preview build  

9 files changed · ± 9 modified

± Modified

• CHANGELOG/index.html
• reference/notation/index.html
• reference/reading/index.html
• reference/language/expressions/index.html
• reference/math_spec/dimensions/index.html
• reference/math_spec/program/index.html
• reference/math_spec/resolution/index.html
• reference/math_spec/sos/index.html
• reference/math_spec/validation/index.html