Keep is/extends inline for multi-argument template heritage

[oha-4]

Fixes the is-breaking part of #11009 (the blank-line/over-indent part is addressed separately in #11010).

Problem

When a model/scalar heritage clause (is X / extends X) references a template with multiple arguments, the formatter wrapped the keyword and the template reference in a single group(indent(...)). Prettier measures that group against the full flat width of the template argument list, so the group broke and pushed the keyword onto its own indented line — even when the declaration line itself was short:

model Picked
  is PickProperties<
    Sample,
    ...
  >;

Fix

printHeritageClause now keeps the keyword on the declaration line when the base is a template reference with multiple (breakable) arguments, and lets the template argument list control the line break:

model Picked is PickProperties<
  Sample,
  ...
>;

Single-argument (hugged) and non-template bases keep the previous behavior of breaking the keyword onto its own line when the declaration is too long.

Tests

• Added regression tests for the multi-argument case on model is, model extends, and scalar extends in formatter.test.ts.
• Reformatted one committed http-client-java test spec (arm-customization.tsp) that contained the old output (whitespace only, no semantic change).
• Formatter test suite, eslint, tsc, and prettier --check "**/*.tsp" all pass locally.

Note: the crossLanguageVersion hash in the corresponding *_metadata.json is whitespace-sensitive; it will be updated from the RegenCheck output once CI runs.

🤖 Generated with Claude Code

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/@typespec/compiler@11011

commit: 95e7383

[github-actions]

All changed packages have been documented.

• ✅ @typespec/compiler
• ✅ @typespec/http-client-java

Show changes

@typespec/compiler - fix ✏️

Keep the is/extends keyword on the declaration line when the base is a template reference with multiple arguments. The template argument list now controls the line breaking instead of the keyword being pushed onto its own indented line.

@typespec/http-client-java - internal ✏️

Reformat a test spec affected by the compiler formatter change keeping is/extends inline for multi-argument template references (whitespace only, no semantic change).

[timotheeguerin]

I'll ask the java team what is going on here if the CI still fails, but you shouldn't be changing those files in that pr

[oha-4]

Thanks for catching this, and for offering to check with the java team — much appreciated! I've reverted the *_metadata.json change so it's out of this PR. The arm-customization.tsp reformat stays since format:check enforces it.

For reference, earlier formatter PRs (#10556, #10558) reformatted these same Java test specs with no metadata change — crossLanguageVersion was added later (#10674), so reformatting had no hash impact back then. This PR is the first formatter change to hit that coupling.

Heads up: I expect RegenCheck to fail, because tcgc derives crossLanguageVersion from the raw .tsp source text (computeCrossLanguageVersion in package.ts), so the whitespace-only reformat changes the hash. Could you check with the java team on the right way to handle this?