render: keep literal text from continuing an embedded expression

[aktanazat]

Summary

Typst keeps parsing an embedded code expression after its closing bracket. Markup that starts with ( becomes an argument list, [ a trailing content argument, and .name a field access. A spreadsheet cell reading "New York" over "(07) Western" produced #linebreak()(07) and the whole document failed with Typst compilation failed: expected function, found content.

Two places let bare text follow a call:

• Inside one escaped string, escape_typst now ends the hard break with a semicolon (#linebreak();), the same terminator the code-mode space strings already use.
• At a run boundary, the existing #[...] safety wrap in write_run_content only looked for a preceding ]. It now also covers ), which closes #linebreak(), #h(..), #o2p-pptx-space(), and the PowerPoint grid words.

Related issue

None filed. Reproduced on a public spreadsheet (VA VALERI fee cost schedule, cell "New York\n(07) Western").

Testing

• cargo test -p office2pdf --lib (2780 passed)
• cargo fmt --all --check, cargo clippy -p office2pdf --all-targets (only the pre-existing effective_last_resort_family dead-code warning)
• New tests: test_escape_typst_terminates_hard_linebreak_before_parenthesis, test_escape_typst_terminates_hard_linebreak_before_field_access, test_render_document_hard_linebreak_before_parenthesis (asserts the terminator never renders), test_fixed_page_text_box_soft_line_break_before_parenthesis_compiles, test_fixed_page_text_box_grid_word_before_parenthesised_run_compiles. All five fail on main and pass here.
• Verified with typst 0.15.1 that A#linebreak().5 percent compiles (a leading .5 is a number), A#linebreak().NET fails with linebreak does not have field "NET", and A#h(1em)(07) B fails with expected function, found content.

Visual impact

• No rendered PDF change
• Rendered PDF change or visual evidence added
• Reason: the semicolon and the #[...] wrap are consumed by the parser; documents that compiled before produce byte-identical Typst layout, and documents that failed before now compile. The render test asserts no ; reaches the text layer.

Checklist

• Commits include a Signed-off-by line
• PR scope contains one root cause
• Remaining visual deviations each reference an open issue (none observed)

[developer0hye]

Thank you for this fix, and for the precise diagnosis. Merged as 151e0fd.

What I verified before merging:

• Reproduced the parser behaviour with the typst 0.15.1 CLI: New York#linebreak()(07) Western and A#h(1em)(07) B fail with "expected function, found content", Runtime#linebreak().NET pricing fails with linebreak does not have field "NET", and both the ; terminator and the #[...] wrap fix them. The same holds inside a #text(...)[...] content block, and the semicolon never reaches the text layer.
• Traced the three other places that still emit a bare #linebreak() (the PowerPoint line split, the overlong-Latin chunk boundary from fix(docx): wrap overlong Latin table tokens #1456, and the PPTX soft-break marker). Every one of them is followed by text that goes through write_run_content, so the new ) guard covers them too.
• Nothing else in the render code pattern-matches the literal #linebreak() string, so the trailing ; cannot confuse a measurement path.

One note for the record, not a request: the guard also fires after a literal ) in text (e.g. "(a)" followed by a run starting with "("), which only adds a harmless #[...]. And the protection is implicit in that it relies on all bare text passing through write_run_content; a future emitter that bypasses it would reintroduce the failure. Fine as is.