tree 4de5b0ba3d2a18f5873077d46af66cc2546a1bc0
parent 88d82a1dfff173906c1218917d9a266c1c2e7020
author Nico Weber <thakis@chromium.org> 1760463234 -0700
committer Pdfium LUCI CQ <pdfium-scoped@luci-project-accounts.iam.gserviceaccount.com> 1760463234 -0700

Undo incorrect arithmetic decoder change + follow-up

Reverts https://pdfium-review.googlesource.com/c/pdfium/+/52712
Reverts https://pdfium-review.googlesource.com/c/pdfium/+/16791

The last of those two is incorrect per JBIG2 spec, the first papers
over an issue caused by it.

The last change even cites the spec part that says that an arithmetic
decoder should keep returning 1-bits on out-of-data. This is also
described in detail in Annex E.

This includes two tests, one of which has several trailing 7fff
byte pairs stripped before the final ffac pair. pdf.js, Preview.app,
Acrobat Reader all decode these files fine.

The construction of these tests is described in detail in:
* https://github.com/SerenityOS/serenity/pull/26272/commits/ba49ffac4dd
* https://github.com/SerenityOS/serenity/pull/26242/commits/9882a8462b6

I took these two files, converted them to PDF using Meta/jbig2_to_pdf.py
in that repo, and then to .in format using the script in
https://pdfium-review.googlesource.com/c/pdfium/+/136590

This also fixes decoding 042_25.jb2 from the Power JBIG2 test set.

clusterfuzz-testcase-pdf_codec_jbig2_fuzzer-5210024078213120 from
the third linked bug is a 5636 x 60419 image and needs ~35s to
run on my machine. To not re-introduce the timeouts fixed by the
incorrect fix, instead limit regions to 4096 x 4096 while fuzzing.
This makes `pdf_codec_jbig2_fuzzer` process that test case in
0 ms instead of in 35s on my machine as it now early-outs.

I confirmed that bug_947622_min.pdf from comment 7 on
bug 41450095 renders exactly the same before and after this change.

Bug: 432920906,41450095,40540933
Change-Id: Idc5c13c9440d0b4749f0832659bcdf36c1bc3597
Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/136630
Reviewed-by: Nico Weber <thakis@google.com>
Commit-Queue: Nico Weber <thakis@chromium.org>
Reviewed-by: Lei Zhang <thestig@chromium.org>