mirror of
https://github.com/rcourtman/Pulse.git
synced 2026-09-10 02:25:56 +00:00
rcourtman
ff97e3943c
TestSummarySection_PaginatesCardGridForManyMetrics failed once on CI (run 31058903048) with metric card "Disk Read" missing and all of page 2 absent from the extracted text. The card-grid pagination is fine; the bug is in the test helper. Its stream regexp (?s)stream\r?\n(.*?)\r?\nendstream let the optional \r consume the final byte of the compressed payload whenever that byte (the last Adler-32 checksum byte) happened to be 0x0D. The truncated stream fails to inflate and the helper silently skipped the entire page, roughly 1 in 256 streams. fpdf's compressed bytes are a pure function of the report text, which embeds the report period, so specific date windows fail deterministically. The CI window (period Jul 30 00:23 to Aug 6 00:23 UTC) reproduces on the first render while neighbouring windows pass, which is why the test passed locally and on the previous run. Slice each stream by the /Length declared in its object dictionary the way a real PDF reader does. fpdf always writes /Length as a direct integer, and compressed bytes can contain any sequence, so keyword scanning can never be exact. Verified against the pinned CI window plus 400 shifted windows, and the full package.