microlinkhq / html-get / 28578547509

98%

push

github

fix: avoid silent empty HTML when pandoc cannot convert office files (#265)

* fix: avoid silent empty HTML when pandoc cannot convert office files

xlsx/pptx readers require pandoc >= 3.8.3; older binaries exit 0 with
empty stdout, which html-get treated as a successful conversion and
returned a blank page. Probe supported input formats at startup, treat
empty/unsupported conversions as failures, and fall back to the raw body
instead of returning empty HTML.

* refactor: probe office binaries once via execFileSync

- memoize defaultMutool/defaultPandoc so the binary probe runs once
  instead of on every request (both are wired as default params, which
  are evaluated per call)
- use execFileSync instead of execSync for which/--list-input-formats:
  no shell spawn, no interpolation surface, matches api ffprobe.js
- pandoc runtime call uses tinyspawn array form so a filepath with
  spaces stays a single argv entry
- log pandoc:empty/pandoc:error instead of silently swallowing, so a
  fallback to the raw body is visible in logs
- export defaultPandoc and cover the real --list-input-formats probe
  (the mock-injection tests bypassed it)

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* refactor: reuse memoizeOne from @metascraper/helpers

Drop the hand-rolled memoize helper in favor of memoizeOne, already
available transitively via @metascraper/helpers. The runners take no
args, so it behaves as memoize-once.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix: guard pandoc probe so a broken binary disables conversion

Restores the try/catch the original PR had around pandoc setup, lost in
the execFileSync refactor. If pandoc is on PATH but --list-input-formats
throws, defaultPandoc() must return undefined (conversion disabled), not
throw out of the default-parameter evaluation and break every getHTML
call. Adds a regression test with a fake pandoc that fails the probe.

Reported by Cursor Bugbot.

Co-Authored-By: Claude Opus 4.8 <n... (continued)

199 of 210 branches covered (94.76%)

Branch coverage included in aggregate %.

59 of 65 new or added lines in 1 file covered. (90.77%)

783 of 794 relevant lines covered (98.61%)

30.75 hits per line