fix(formatter): break call args instead of = for hugged template assignments

[jsmecham]

Summary

Fixes #21908

When a long-LHS variable declaration assigns the result of a single-template-arg call (e.g. graphql(/* GraphQL */ `...`)), oxfmt broke at = and kept the call hugged. Prettier instead keeps the assignment on one line and breaks the call's arguments out around the template.

Input

const xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx = graphql(
  /* GraphQL */ `
    query Q {
      x
    }
  `
);

Before (incorrect)

const xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx =
  graphql(/* GraphQL */ `
    query Q {
      x
    }
  `);

After (matches Prettier)

const xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx = graphql(
  /* GraphQL */ `
    query Q {
      x
    }
  `
);

Short-LHS cases stay hugged (matches Prettier):

const a = graphql(`
  query { x }
`);

Root cause

In print/call_like_expression/arguments.rs, the three template-hug predicates (is_multiline_template_only_args, is_graphql_call_with_single_template_arg, is_huggable_html_embed_single_arg) emitted (template) directly with no group. The template's hard newlines made the parent assignment's Fluid group exceed the print width, so the assignment broke at = — even though breaking the call's args would have produced a shorter, Prettier-matching layout.

In Prettier, shouldExpandLastArg returns true for embedded single-template args and uses conditionalGroup([hugged, hugged-with-break, allArgsBrokenOut]), so the assignment never breaks at =; the call args break instead when the hugged form overflows.

Fix

Split the existing hug branch into two paths:

1. Always-hug (require, test calls, react hook deps) — unchanged, emits (args) directly. These paths never deal with multiline content, so unconditional hug is still correct.

2. Template-hug (graphql / multiline template / html embed) — when the call is the immediate RHS of a VariableDeclarator or AssignmentExpression and the template was already multi-line in the source, wrap the args in best_fitting!:

   • Variant 1 (preferred): (template) — hugged
   • Variant 2 (fallback): (soft_block_indent(template)) — broken-out

   best_fitting acts as a "fits" boundary, so the assignment's Fluid group sees the args as fitting and doesn't break at =. When the hugged variant doesn't fit on the first line, the printer falls through to the broken-out variant.

   Single-line templates that get rewritten to multi-line by the embedded formatter still fall back to the unconditional hug, which lets will_break propagate to break the assignment at = — matching Prettier's breakParent behavior for embed labels.

   Non-assignment contexts (member chains, plain expression statements) keep the unconditional hug so the surrounding chain group decides how to break — preserving the existing expect(content).toMatch(`...`) chain-break behavior.

Tests

A new fixture crates/oxc_formatter/tests/fixtures/js/calls/template-arg-hug.js exercises five layout scenarios at both printWidth: 80 and printWidth: 100:

• Short LHS hugs, no = break
• Short LHS with leading comment still hugs
• Long LHS at width 80 → broken-out args; at width 100 → hugs (validates the width-aware decision)
• Member chain breaks at the dot, args stay hugged
• Generic (non-graphql) multi-line template arg behaves the same way

Prettier conformance: 746/753 JS, 591/601 TS — no regressions vs. main.

AI Disclosure

This PR was co-authored with Claude Code (AI assistant), as noted in the commit. The fix was reviewed, tested against the formatter test suite and full Prettier conformance suite, and verified to produce no regressions.

[codspeed-hq]

Merging this PR will not alter performance

✅ 44 untouched benchmarks
⏩ 7 skipped benchmarks¹

---

_{Comparing jsmecham:worktree-fixissue-21908 (1aadd72) with main (4380812)}

Footnotes

1. 7 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩