humanspeak / svelte-markdown / 25453385123

93%

main: 81%

push

github

perf(cleanup): collapse shrinkHtmlTokens to a single htmlparser2 pass (#288)

Folds the legacy three-pass shape (containsMultipleTags regex sweep +
parseHtmlBlock flat tokenization + processHtmlTokens stack-based
nesting) into one inline-nesting htmlparser2 traversal. The fast
two-indexOf hasMultipleTags check replaces the global regex; multi-tag
html tokens emit already-nested tokens directly, so the post-pass walk
no longer re-traverses every nested descendant.

Public API unchanged. parseHtmlBlock, processHtmlTokens, and
containsMultipleTags remain exported with their original behavior so
their dedicated test files keep passing.

Numbers on parse-html-heavy (145.5 KB / 1001 tokens, median of 5 runs):

  cleanupMs WARM:     15.8 ms -> 8.0 ms  (1.98x)
  cleanupMs COLD:     15.8 ms -> 8.9 ms  (1.78x)
  parseColdMs WARM:   25.9 ms -> 16.8 ms (-35%)
  parseColdMs COLD:   29.5 ms -> 18.4 ms (-38%)

Variance also tightens substantially: baseline WARM cleanupMs ranged
14.2-26.2 ms; after, 7.6-8.6 ms.

A new unit test pins the single-pass shape so future refactors cannot
silently regress to the three-pass form.

Closes #284

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

506 of 578 branches covered (87.54%)

Branch coverage included in aggregate %.

75 of 79 new or added lines in 1 file covered. (94.94%)

1044 of 1081 relevant lines covered (96.58%)

5042.38 hits per line