GitLab

The current heuristics always creates a new paragraph whenever the
font-size changes, which is undesirable.

Simply removing the size check causes many of the existing tests to fail
because the size-change is often the only reason that we have multiple
paragraphs at all.  The problem is our PDF file has too few lines on
each page to compute reasonable line & paragraph size thresholds.

So this change required changing the heuristics.  The new heuristic is
as follows:

1. We keep track of the top & bottom of the current line, as weighted
averages of the (recent) text boxes on the line.
2. When we encounter a new text box, if it significantly overlaps the
top-to-bottom range, it's considered part of the same line.
3. If we are starting a new line, we also check the paragraph threshold
to see if we should also start a new paragraph.  If the paragraph
threshold couldn't be computed (because there weren't enough lines on
the page), we compare against the line size.

We also introduce the `PDFExtensionAccessibilityTextExtractionTest`
test suite.  These tests are like the tree-dump tests, but they dump
raw text content, split into lines and paragraphs.  (Compared to
tree-dump tests, this approach allows us to test the kNextOnLine and
kPreviousOnLine attributes are correct.)

Bug: 985604
Change-Id: Idfce6edfef42580e7fac4d8a7753c82495c15bd1
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1743032
Commit-Queue: Ian Prest <iapres@microsoft.com>
Reviewed-by: Lei Zhang <thestig@chromium.org>
Reviewed-by: Kevin Babbitt <kbabbitt@microsoft.com>
Cr-Commit-Position: refs/heads/master@{#690153}