test gap: multiple SEQ in same paragraph + state-machine reset / truncation invariants (#104 follow-up)

[kiki830621]

Problem

From verification of PsychQuant/che-word-mcp#104 (verify report: comment-4346442620):

「Multiple SEQ fields in same paragraph ([..., begin1, instrText1, sep1, cached1, end1, begin2, ...]) untested. State machine appears to reset correctly on second begin, but no explicit verification. Same for mixed-identifier within paragraph.」

— Source: team:logic

parseFiveRunSpan (Sources/OOXMLSwift/Parsing/FieldParser.swift:181-246) is a state machine that resets on each fldCharType="begin". Per code review the reset path is correct, but no test exercises:

1. Two consecutive SEQ Figure fields in same paragraph — [caption, begin1, instrText("SEQ Figure"), separate1, cached1, end1, " ", begin2, instrText("SEQ Figure"), separate2, cached2, end2]. Should return 2 ParsedFields.

2. Mixed-identifier SEQ in same paragraph — [..., begin1, instrText("SEQ Figure"), ..., end1, begin2, instrText("SEQ Table"), ..., end2]. Should return 2 ParsedFields with different identifiers; updateAllFields should increment both counters.

3. Truncated paragraph (begin without end) — [begin, instrText] with no separate / end. State machine should leave current non-nil at loop exit and silently emit nothing. Currently doc-comment doesn't pin this invariant.

Type

test gap (P3)

Acceptance criteria

• Add 3 sub-tests in Issue104FieldParserCanonicalFormTests (or a new FieldParserStateMachineEdgeCasesTests file) covering the three scenarios above.
• Document the truncated-paragraph behavior in parseFiveRunSpan doc-comment.

Why low priority

State machine logic appears correct on inspection. Real-world OOXML producers rarely emit multiple SEQ in same paragraph (typical pattern is one caption per paragraph). But pathological documents or future producers might, and the assumption is unverified.

Related: PsychQuant/che-word-mcp#104

---

Current Status

Phase: implemented
Last updated: 2026-05-02 by idd-implement

Key Decisions

• Implemented same-paragraph canonical SEQ parser/update coverage in PR Cover canonical sequence field edge cases #37.
• Documented truncated canonical spans as discarded when no closing fldCharType="end" appears.

Scope Changes

• (none)

Blocking

• Awaiting IDD verify / PR review.

Commits

• 11860b5 test: cover canonical sequence field edge cases

[kiki830621]

Implementation Plan

Scope

Pin canonical 5-run state-machine edge cases without changing parser behavior.

Plan

• Add same-paragraph tests for two consecutive SEQ Figure fields.
• Add same-paragraph tests for mixed SEQ Figure + SEQ Table fields and verify updateAllFields() increments both counters.
• Add a truncated begin+instrText-without-end test asserting no field is emitted.
• Document the truncated-span behavior in FieldParser.parseFiveRunSpan.

[kiki830621]

Implementation Complete

Checklist (synced from TaskList)

• Add same-paragraph two-SEQ Figure canonical parser coverage -> commit 11860b5
• Add same-paragraph mixed SEQ Figure / SEQ Table parser and updateAllFields() coverage -> commit 11860b5
• Add truncated canonical span coverage and document the discard behavior -> commit 11860b5

Changes

• 11860b5: added same-paragraph state-machine coverage and truncated-span behavior tests.
• 11860b5: documented that a canonical span without closing fldCharType="end" emits no partial ParsedField.
• PR: Cover canonical sequence field edge cases #37

Files Changed

Sources/OOXMLSwift/Parsing/FieldParser.swift       |   3 +-
Tests/OOXMLSwiftTests/Issue104FieldParserCanonicalFormTests.swift | 164 +++++++++++++++++++++

Verification

• swift test --filter Issue104FieldParserCanonicalFormTests -> passed
• swift test -> 834 tests, 1 skipped, 0 failures

Scope Compliance

Only the requested canonical state-machine edge cases and doc-comment invariant were changed.

Next: verification pending