[vicmcorrea]

Problem and fix

Formatting a continued tuple body after a comma can turn valid Bend into a parser error. For example, this runs and prints (17, [3, 5]), but formatting moves the second array write out of the first write's column:

import Base
def main() -> U32 & Array<U32>:
  a = [0 : U32 * 2n]
  (17,a[0]<-3
      a[1]<-5)

Start the existing bounded body scope after commas at parenthesis depth, outside generic arguments. Continuation lines then move with their body anchor, as they already do after (. This extends #1816 using #1803's existing scope expiry.

Blast radius

Only tools/bend-fmt-lsp/src/formatter.ts and its existing test file change. Parenthesis tracking limits the new origins to parenthesized comma sites, including call/declaration parentheses; commas inside generic arguments or unparenthesized array/constructor contexts are excluded. No parser, compiler, runtime, API, dependency or gate changes.

Regression evidence

Two fixtures extend the existing exact-output/idempotence and real-CLI harnesses: a continued tuple item and a tuple followed by an observably consumed call argument. The original programs print (17, [3, 5]) and [28, 5]; formatted versions fail parsing on main in default, four-space and tab modes, and retain those outputs with the fix.

Checks

• npm test: 33/33, zero skips, including CLI regressions and the LSP stdio session.
• npm pack --dry-run --json: build succeeds; 14 expected package files.
• bun gates/repo.ts: 54/54 with CI-pinned ttok 0.3.
• git diff --check: passes.

Validation limits

Run locally on macOS with Node v26.8.2 and Bun 1.3.11. No Linux, other Node versions or editor-client UI validation; the LSP was tested over stdio. Cluster gates were not run. The token-based formatter's existing generic-close/assignment tokenization and unchanged-source fallback limitations remain outside this fix.