tests/test-changelog-exec.t

parsers: optimize filtered headrevs logic

The old native head revs logic would iterate over every node, starting from 0,
and check if every node was filtered (by testing it against the filteredrevs
python set). On large repos with hundreds of thousands of commits, this could
take 150ms.

This new logic iterates over the nodes in reverse order, and skips the filtered
check if we've seen an unfiltered child of the node. This saves approximately a
bagillion filteredrevs set checks, which shaves the time down from 150ms to
20ms during every branch cache write.

#require execbit

b51a8138292a introduced a regression where we would mention in the
changelog executable files added by the second parent of a merge. Test
that that doesn't happen anymore

  $ hg init repo
  $ cd repo
  $ echo foo > foo
  $ hg ci -qAm 'add foo'

  $ echo bar > bar
  $ chmod +x bar
  $ hg ci -qAm 'add bar'

manifest of p2:

  $ hg manifest
  bar
  foo

  $ hg up -qC 0
  $ echo >> foo
  $ hg ci -m 'change foo'
  created new head

manifest of p1:

  $ hg manifest
  foo

  $ hg merge
  1 files updated, 0 files merged, 0 files removed, 0 files unresolved
  (branch merge, don't forget to commit)
  $ chmod +x foo
  $ hg ci -m 'merge'

this should not mention bar but should mention foo:

  $ hg tip -v
  changeset:   3:c53d17ff3380
  tag:         tip
  parent:      2:ed1b79f46b9a
  parent:      1:d394a8db219b
  user:        test
  date:        Thu Jan 01 00:00:00 1970 +0000
  files:       foo
  description:
  merge
  
  

  $ hg debugindex bar
     rev    offset  length  ..... linkrev nodeid       p1           p2 (re)
       0         0       5  .....       1 b004912a8510 000000000000 000000000000 (re)

  $ cd ..