Context
The golden corpus (MacDownTests/Fixtures/, 31 fixtures, consumed by MPMarkdownRenderingTests.m, MPSyntaxHighlightingTests.m, MPMathJaxRenderingTests.m) is the safety net for both the cmark-gfm swap (v3000.1.1) and the Swift port (v3001.0.0). Two gaps stand between it and that role. From the ground survey (plans/dependency-overhaul-survey.md §6.2 on branch claude/dependencies-overhaul-planning-hw3lms) and decision register D-1/D-3/D-4.
Part 1 — Baseline status annotations (D-1)
The corpus faithfully snapshots hoedown's bugs: #26 (list numbering), #27 (merged blockquotes), #10 (code indentation) are expected to close as side effects of cmark-gfm. With plain XCTAssertEqualObjects, the parser swap will "fail" tests by fixing bugs, and a blind regenerate would stamp over intentional deviations.
Add a sidecar manifest (Fixtures/manifest.json) tagging every fixture:
matches-spec — output is CommonMark/GFM-correct; must survive the swap byte-identical (or with reviewed formatting-only diffs)known-hoedown-bug — output is wrong today; the swap is expected to change it (link the issue)intentional-macdown-deviation — MacDown deliberately differs from GFM; must survive the swap
A sidecar beats renaming files: the loader (MPMarkdownRenderingTests.m:59) stays untouched. Enumerate the third category now, while the knowledge is fresh — it is the migration's most fragile knowledge.
Part 2 — Parser-contract coverage (D-3)
Add fixtures for every contract surface in survey §2.12 not currently covered:
Non-goal
A rendered-DOM/visual corpus is deliberately not part of this issue — that belongs to the WKWebView milestone (#111), where the web view is the variable under test (D-4: two corpora, one variable each).
References
Context
The golden corpus (
MacDownTests/Fixtures/, 31 fixtures, consumed byMPMarkdownRenderingTests.m,MPSyntaxHighlightingTests.m,MPMathJaxRenderingTests.m) is the safety net for both the cmark-gfm swap (v3000.1.1) and the Swift port (v3001.0.0). Two gaps stand between it and that role. From the ground survey (plans/dependency-overhaul-survey.md§6.2 on branchclaude/dependencies-overhaul-planning-hw3lms) and decision register D-1/D-3/D-4.Part 1 — Baseline status annotations (D-1)
The corpus faithfully snapshots hoedown's bugs: #26 (list numbering), #27 (merged blockquotes), #10 (code indentation) are expected to close as side effects of cmark-gfm. With plain
XCTAssertEqualObjects, the parser swap will "fail" tests by fixing bugs, and a blind regenerate would stamp over intentional deviations.Add a sidecar manifest (
Fixtures/manifest.json) tagging every fixture:matches-spec— output is CommonMark/GFM-correct; must survive the swap byte-identical (or with reviewed formatting-only diffs)known-hoedown-bug— output is wrong today; the swap is expected to change it (link the issue)intentional-macdown-deviation— MacDown deliberately differs from GFM; must survive the swapA sidecar beats renaming files: the loader (
MPMarkdownRenderingTests.m:59) stays untouched. Enumerate the third category now, while the knowledge is fresh — it is the migration's most fragile knowledge.Part 2 — Parser-contract coverage (D-3)
Add fixtures for every contract surface in survey §2.12 not currently covered:
data-informationcode-fence accessory (hoedown_html_patch.c:89-91)[TOC]splice shapes, incl. nested lists (MPRenderer.m:218-242)section(hoedown_html_patch.c:182-262)HOEDOWN_EXT_MATH/MATH_EXPLICIT)MPPreprocessMarkdownworkaround — Trouble with two shortcut links side by side #25, Fix code block formatting when no blank line before block #36, Fix code block rendering with square brackets #37, Rendering of lists is off if it doesn't have a blank line before it #254 (MPRenderer.m:109-193). These regexes exist to patch hoedown bugs; a fixture each is the proof they can be deleted under cmark-gfm.CMARK_OPT_UNSAFE/tagfilter decision (decision register D-9) as deliberate behavior, not accidentdata-checkbox-indexordering)Non-goal
A rendered-DOM/visual corpus is deliberately not part of this issue — that belongs to the WKWebView milestone (#111), where the web view is the variable under test (D-4: two corpora, one variable each).
References
plans/dependency-overhaul-survey.md§2.12, §6.2 (branchclaude/dependencies-overhaul-planning-hw3lms)plans/dependency-overhaul-decisions.mdD-1, D-3, D-4