fix(parser): preserve fatal error across top-level await reparse

[ammubhave]

Summary

Fixes #23673. oxfmt panicked with string must start with a quote (oxc_formatter/src/utils/string.rs) on input like await(E);import. The root cause is in the parser: a fatal parse error was silently discarded, so a file that should be rejected was reported as parsed successfully with a malformed AST.

Problem

await(E);import is invalid — the trailing import is incomplete, which is a fatal error. On its own (import, or x;import) the parser reports it correctly. But await(E) is a top-level await: in unambiguous mode the parser first parses await as an identifier, records a checkpoint, and once the import proves the file is a module, reparses that statement with await enabled (reparse_potential_top_level_awaits).

That reparse calls rewind(checkpoint), and rewind restores all saved state — including self.fatal_error. The checkpoint was taken before the import, so rewinding reset fatal_error back to None, erasing it. The parser then returned no diagnostics and a malformed ImportDeclaration whose source is a synthetic StringLiteral { span: 0..0, value: "", raw: None }. Because oxc_formatter::format only runs when parsing reports no diagnostics, this malformed program reached the formatter, which sliced the empty source text and tripped the assertion (and would index-underflow in release).

Fix

Stash self.fatal_error before the reparse loop and restore it afterward. Reparsing only re-derives the AST for already-valid await statements, so a fatal error from a later statement is unrelated to them and must outlive the reparse. A fatal error newly surfaced by reparsing comes earlier in source order and takes priority; otherwise the first-pass error is restored.

This cannot cause false positives: the reparse only triggers after an import/export parsed successfully (parsing never stopped earlier), so any surviving fatal_error necessarily came from a statement after the reparsed awaits — which the reparse cannot fix anyway. It also fixes the class of bug generally, not just the empty-import-source symptom.

Example

await(E);import

Before: parsed "successfully" → oxfmt panics. After: Expected frombut foundEOF``, and valid files like await(E); import x from "y"; still format normally.

Testing

• Added tasks/coverage/misc/fail/oxc-23673.js (negative parser fixture).
• cargo coverage -- parser: 100% AST-parsed, no regressions; the only snapshot change is the new fixture counted as a correctly-failing test (Negative Passed: 149 → 150).
• cargo test -p oxc_parser, clippy, and rustfmt pass.

---

AI usage disclosure: this change was investigated and drafted with the help of Claude Code, then reviewed and tested by the contributor.

🤖 Generated with Claude Code

[codspeed-hq]

Merging this PR will not alter performance

✅ 62 untouched benchmarks
⏩ 9 skipped benchmarks¹

---

_{Comparing ammubhave:fix-23673 (104abe6) with main (36009dd)}

Footnotes

1. 9 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. ↩

[Sysix]

this will fix #23681 too