GitLab

Also rewrite matching to be more assertive on particular sub-parts of
the string. Hopefully this would allow resolving particular cases before
the perf/memory regressions creep in.

The new parsing finds extra ~2000 symbols compared to the previous
version.

For testing I compared the extracted symbol infos with the output of
'arm-linux-androideabi-nm --print-size --size-sort --reverse-sort' on the
instrumented binary. The differences are:
* 'linker_script_end_of_text' is ignored by the symbol extractor (the
  'linker_script_start_of_text' is output as a special case because we
  rely on it later)
* the symbol extractor seems to rightfully ignore local labels and
  pointers deep into functions, like: 'PRE_LOOP_STAGES_7_OR_8' or
  'jsimd_idct_islow_neon_consts'
* v8 builtins like 'Builtins_RecordWrite' have the size 0 and
  unfortunately are not distinguishable from labels above

Bug: 893981
Change-Id: Ie2ac9a76508832cb90fdedda1ceb46f6c2b24dba
Reviewed-on: https://chromium-review.googlesource.com/c/1326486Reviewed-by: Matthew Cary <mattcary@chromium.org>
Commit-Queue: Egor Pasko <pasko@chromium.org>
Cr-Commit-Position: refs/heads/master@{#606475}