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.
Minimal hgk check
$ echo "[extensions]" >> $HGRCPATH
$ echo "hgk=" >> $HGRCPATH
$ hg init repo
$ cd repo
$ echo a > a
$ hg ci -Am adda
adding a
$ hg debug-cat-file commit 0
tree a0c8bcbbb45c
parent 000000000000
author test 0 0
revision 0
branch default
phase draft
adda
$ echo b > b
$ hg ci -Am addb
adding b
$ hg log -T '{node}\n'
102a90ea7b4a3361e4082ed620918c261189a36a
07f4944404050f47db2e5c5071e0e84e7a27bba9
$ hg debug-diff-tree 07f494440405 102a90ea7b4a
:000000 100664 000000000000 1e88685f5dde N b b
$ hg debug-diff-tree 07f494440405 102a90ea7b4a --patch
diff --git a/b b/b
new file mode 100644
--- /dev/null
+++ b/b
@@ -0,0 +1,1 @@
+b
Ensure that diff-tree output isn't affected by diffopts
$ hg --config diff.noprefix=True debug-diff-tree 07f494440405 102a90ea7b4a
:000000 100664 000000000000 1e88685f5dde N b b
$ hg --config diff.noprefix=True debug-diff-tree --patch 07f494440405 102a90ea7b4a
diff --git a/b b/b
new file mode 100644
--- /dev/null
+++ b/b
@@ -0,0 +1,1 @@
+b
$ cd ..