fix(web): /metrics language field emits non-lookup display names (c/c++, c#)

[dekobon]

Decided (2026-06-05): one canonical lowercase slug per language everywhere (cpp/csharp/tsx/…) for Display/FromStr/CLI-JSON/web/py; drop c/c++/c# and the Python override. See the resolved-decisions comment.

Summary

The REST /metrics response language field emits the raw LANG::name()
display string — "c/c++", "c#" — which is not a valid language= lookup
token, and which the Python bindings deliberately override. The two published
surfaces disagree on the canonical name for the same language.

Evidence

• WebMetricsResponse.language (big-code-analysis-web/src/web/metrics.rs:34),
  populated from guess_language's second tuple element
  (big-code-analysis-web/src/web/server.rs:311), which is LANG::name()
  (src/tools.rs). LANG::name() returns "c/c++" for C++ and "c#" for C#.
• The Python bindings override exactly these tokens
  (big-code-analysis-py/src/language.rs:46, lang_to_name): Cpp → "cpp",
  Csharp → "csharp", Tsx → "tsx", precisely because "c/c++" / "c#"
  are not valid lookup tokens.

Why 2.0-worthy

A client cannot round-trip the language string from a /metrics response back
into any language-name parameter, and web vs py disagree on canonical names. The
web response shape is a frozen API; changing "c/c++" → "cpp" is breaking → 2.0.

Proposed change

Decide a single canonical language-name vocabulary across web + Python (+ the CLI
JSON language field) for 2.0. Route the web language field through the same
lang_to_name-style override the bindings already use. This pairs with #508's
Display/FromStr work — the canonical name should be the FromStr-parseable
token, not the human display name.

Acceptance

• /metrics language is a valid lookup token consistent with the Python
  surface and LANG::FromStr.

Part of the pre-2.0 review (#505).

Resolution

Fixed in fix/issue-540 (commit 57e056d9). One canonical lowercase
slug per language is now used on every surface (LANG::name /
Display / FromStr, CLI JSON, web /metrics, Python). Cpp → cpp,
Csharp → csharp, Tsx → tsx; the pretty c/c++ / c# forms are
dropped. Display is injective and round-trips exactly through
FromStr for every variant (enforced by a new all-variants round-trip
test plus a no-punctuation guard). fake::get_true reports cpp for
Objective-C/C++ instead of the unparseable obj-c/c++. The Python
lang_to_name override is removed as redundant. Breaking serialized
language value, deferred to 2.0; version stays 1.x. See the
resolution comment for the full slug table and decision rationale.

[dekobon]

Resolved decisions (2026-06-05)

One canonical lowercase slug per language, used everywhere. A single
round-trippable token (cpp, csharp, tsx, javascript, mozjs, python,
rust, …) becomes the value of LANG::name() / Display / FromStr and of the
CLI JSON language field, the web /metrics language field, and the Python
surface.

• Drop the human-pretty c/c++ / c# display forms entirely (devs read
  cpp/csharp fine); no separate display_name().
• Remove the Python lang_to_name override (big-code-analysis-py/src/ language.rs) — it only existed to paper over c/c++/c#; now redundant.
• The web bug (raw LANG::name() → c/c++) disappears: guess_language's
  second tuple element is already this canonical name.
• FromStr round-trips the slug exactly (revisits feat(lang): implement Display/FromStr/Hash for the LANG enum #508, whose FromStr/Display
  were built around the old name()); the only non-injective case remains the
  Tsx/Typescript display-alias unless given distinct slugs — confirm at
  implementation.

Cross-reference #507 (JS default/mozjs split) and #508 (Display/FromStr/Hash):
the slugs must match the .jsm→mozjs, .js→javascript split. 2.0-breaking
(changes Display output + the serialized language value).

[dekobon]

Fixed in fix/issue-540 (commit 57e056d9)

Implemented the resolved decision: one canonical lowercase slug per language, used everywhere.

Slug table (every variant → canonical slug)

Variant	Slug	Variant	Slug
Javascript	javascript	Cpp	cpp (was c/c++)
Mozjs	mozjs	Csharp	csharp (was c#)
Java	java	Elixir	elixir
Go	go	Python	python
Kotlin	kotlin	Tsx	tsx (was typescript)
Lua	lua	Typescript	typescript
Rust	rust	Bash	bash
Tcl	tcl	Ccomment	ccomment
Irules	irules	Preproc	preproc
Perl	perl	Php	php
Ruby	ruby	Groovy	groovy

All 22 slugs are distinct, so Display is injective and from_str(to_string(l)) == Ok(l) for every variant.

Decisions

• Tsx → tsx (not typescript). The decision comment required injectivity, but Tsx and Typescript both rendered typescript, which is not injective and fails the mandatory round-trip-all-variants test. Splitting Tsx to tsx (the value the old Python override already produced) makes both distinct and round-trippable, matching the Mozjs/Javascript split from refactor(api): normalize &LANG-by-value and drop Java-style get_ getter prefixes #507.
• objc/objc++ → cpp. fake::get_true returned the unparseable "obj-c/c++" pseudo-name. Objective-C/C++ parse with the C/C++ grammar (LANG::Cpp), so it now reports "cpp" — the same slug LANG::Cpp.name() emits and a valid FromStr token.

Root cause

mk_lang! generated name()/Display/FromStr from the table's field-4 display string ("c/c++", "c#"). The web /metrics language field is guess_language(...).1, which is that display string — so C/C++ and C# files surfaced non-lookup tokens. Fix: field-4 now holds the slug for Cpp/Csharp/Tsx.

Changes

• src/langs.rs: Cpp/Csharp/Tsx display → slug; fake::get_true → LANG::Cpp.name() ("cpp"); tests updated to assert slugs + exact round-trip + no-punctuation guard.
• src/macros/mod.rs: Display/FromStr doc comments rewritten (injective; slug is the single cross-surface identifier).
• big-code-analysis-py/src/language.rs: lang_to_name override removed (redundant); callers use LANG::name(); tests updated.
• big-code-analysis-web/src/web/server_tests.rs: new test_web_metrics_language_is_canonical_slug (C++→cpp, C#→csharp, both parse::<LANG>() round-trip).
• big-code-analysis-cli/src/html_report.rs: palette LUT keyed on slugs; tsx row added (typescript tint).
• big-code-analysis-cli/src/markdown_report.rs, docs/file-detection.md, big-code-analysis-book/src/languages.md, STABILITY.md: doc/test updates.

Tests

Full workspace cargo test --all-features green (2553 lib + web 62 + py 48 + cli). No parent .snap.new; integration submodule clean (it does not emit display language names). fmt/clippy (both feature flavours) clean; cargo xtask man-page drift clean; cargo doc -D warnings clean.

CHANGELOG (not edited per workflow)

### Changed under ## [Unreleased], marked (breaking) and deferred to 2.0:

(breaking) LANG::name/Display/FromStr now use one canonical lowercase slug per language; the pretty c/c++ / c# display forms are dropped and Tsx reports tsx. The serialized language value (CLI JSON, web /metrics, Python) changes accordingly and is now always a valid FromStr lookup token. (Deferred to the next major bump.)