Signed-off-by: Mattia Rizzolo <mattia@debian.org>

Signed-off-by: Mattia Rizzolo <mattia@debian.org>

Gbp-Dch: Ignore
Signed-off-by: Mattia Rizzolo <mattia@debian.org>

--max-report-size:
  Old: in --html-dir this limited only the parent page
  New: in --html-dir this applies across all pages

--max-diff-block-lines:
  Old: in --html-dir 4 * this number applied across all pages (for a given diff block)
  New: in --html-dir this applies across all pages (for a given diff block)

--max-page-size:
  New flag
  Applies to the sole --html page, or the top-level --html-dir page

--max-report-child-size
  Renamed to
--max-page-size-child:
  No behavioural changes

--max-diff-block-lines-parent
  Renamed to
--max-page-diff-block-lines:
  Old: Only applied to the top-level --html-dir page
  New: Applies to the sole --html page, or the top-level --html-dir page

The reasoning behind these changes is that it's unlikely someone would want to
generate a 500MB single html page, but they might in theory generate a 500MB html
directory split up into several 200KB pages, plus a single 200KB html page as
a summary. The new semantics for these flags allows both to be generated in one
run using the same set of flags.

- Don't toggle buttons for diffs that haven't been loaded yet
- Make the diffcontrol headers clickable

In d5b71fab I extracted the side-by-side diff logic away from the HTML presenter
into its own class. In doing so I moved MAX_LINE_SIZE from executing before the
linediff algorithm, to executing after it. The reason was that I saw the old
truncation logic was incorrect, it could cause two different lines who had a
matching prefix longer than MAX_LINE_SIZE to look identical. However the Wagner
Fischer algorithm used to calculate the linediff blows up when dealing with
inputs much longer than 1K.

This commit restores a MAX_WF_SIZE to inside the linediff algorithm to prevent
RAM blowup. It also adds some optimisations to effectively relax the limit in
certain common safe cases, as well as implements a truncation algorithm that
preserves the distinct identities of very-long lines that are different.

3.6 is not yet default so some existing modules e.g. rpm are missing their
compiled components.

Gbp-Dch: Ignore
Signed-off-by: Mattia Rizzolo <mattia@debian.org>

Signed-off-by: Mattia Rizzolo <mattia@debian.org>

* experimental: (154 commits)
  More ydiff/linediff from diffoscope.{difference => diff} to group unified_diff related things together
  difference: has_children -> has_visible_children, and take into account comments
  Add various traverse_* methods to Difference
  Move side-by-side and linediff algorithms to difference.py
  Refactor html-dir presenter to be a class instance, avoiding global state
  Document the 'h' variable in our raw_feeder
  Split diffoscope.difference into feeders.
  Tidy diffoscope/difference.py
  html-dir: show/hide diff comments, which can be very large
  PEP8-ify some files.
  Explicitly specify keepends=True over comment.
  Reorder and rename FIFOFeeder arguments to prevent tools incorrectly parsing as a SyntaxError.
  Remove extraneous whitespace around cmdclass invokation
  Reword comment.
  Add comparator for Fontconfig cache files.
  Make tests less brittle and remove pointless cases
  tools: extend external Arch Linux tools
  Use unicode chars for the +/- controls instead of hacky punctuation
  Don't show +/- controls for differences with no children
  Fix create_limited_print_func
  ...