LSP: offer every lib/ module in import completion (#692)

[InauguralPhysicist]

The import completion list was a hand-maintained array in src/eigenlsp.c holding 34 names while lib/ has 75 modules. 41 were invisible in the editor — every ui_w_* widget module, the entire science set (physics, chemistry, biology, calculus, linalg, numerics, optimize, probability, geometry, earth_science, engineering), plus audio, utf8, sync, pkg and others.

The fix deletes the fossil rather than updating it

#590 (merged last week) already added tools/gen_lsp_stdlib_index.sh, which scrapes lib/ at build time into lsp_stdlib_index.h for symbol completion + hover. The hardcoded array was sitting directly next to its own replacement. The generator now also emits stdlib_modules[], and completion iterates that — so the list is structurally incapable of drifting from lib/ again.

Why the module list is NOT derived from stdlib_docs

This is the one real design decision here. stdlib_docs only holds public defines, and five modules have none:

module	why it has no docs rows
ui_layout	all four defines are _-prefixed (private)
text_builder	declares no top-level define
ui_dnd, ui_focus, ui_registry	same shape

All five are importable regardless — and ui_layout was in the old hardcoded 34. So the obvious one-line implementation (walk stdlib_docs, collect distinct module names) would have shipped 70 modules and silently dropped one that previously worked. stdlib_modules[] is therefore generated from the lib/ file set, which is the thing "what can I import" actually means. Both the generator header comment and the test spell this out so nobody later "simplifies" it back.

Validation

tests/test_lsp.py asserts the completion set against the live lib/*.eigs listing rather than a hardcoded number — adding a module can never silently fall out of completion again.

Verified red-then-green: with the fix reverted, the new assertion fails and prints exactly the 41 missing modules, corroborating the issue's headline number independently.

check	result
tests/test_lsp.py	82 passed, 0 failed (2 new)
same test on the old code	FAIL, naming all 41 missing modules
release suite	3185/3185
make lsp + build.sh lsp	both regenerate the index; no new warnings

(Both build paths were checked deliberately — build.sh drifting from the Makefile has bitten before. build.sh lsp caught a -Wcomment warning I'd introduced by writing a glob inside a block comment; fixed. The remaining lint.c warnings are pre-existing.)

Closes #692

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR fixes EigenLSP’s import -context completion so it offers every lib/*.eigs module, by deleting the previously hand-maintained module-name array and instead using a build-generated stdlib_modules[] list emitted alongside the existing stdlib_docs index.

Changes:

• Extend tools/gen_lsp_stdlib_index.sh to also emit stdlib_modules[] derived from the lib/ file set (not from stdlib_docs rows).
• Update src/eigenlsp.c to iterate stdlib_modules[] for module-name completion after import .
• Add a regression test in tests/test_lsp.py that asserts the completion set matches the live lib/*.eigs file listing; document the fix in CHANGELOG.md.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
tools/gen_lsp_stdlib_index.sh	Generator now emits stdlib_modules[] (module list sourced from lib/ file set).
src/eigenlsp.c	import completion now comes from generated stdlib_modules[] instead of a hardcoded list.
tests/test_lsp.py	Regression test asserts import-module completion matches actual lib/*.eigs modules (including modules with no public defines).
CHANGELOG.md	Adds release-note entry describing the completion-list drift and the generated-list fix.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.