perf: add perfrevlogrevision
As part of investigating performance improvements to revlog reading,
I needed a mechanism to measure every part of revlog reading so I knew
where time was spent and how effective optimizations were.
This patch implements a perf command for benchmarking the various
stages of reading a single revlog revision.
When executed against a manifest revision at the end of a 30,000+
long delta chain in mozilla-central, the command demonstrates that
~80% of time is spent in zlib decompression.
#require svn svn-bindings
$ cat >> $HGRCPATH <<EOF
> [extensions]
> convert =
> EOF
$ convert()
> {
> startrev=$1
> repopath=A-r$startrev-hg
> hg convert --config convert.svn.startrev=$startrev \
> --config convert.svn.trunk=branches/branch1 \
> --config convert.svn.branches=" " \
> --config convert.svn.tags= \
> --datesort svn-repo $repopath
> hg -R $repopath log -G \
> --template '{rev} {desc|firstline} files: {files}\n'
> echo
> }
$ svnadmin create svn-repo
$ svnadmin load -q svn-repo < "$TESTDIR/svn/startrev.svndump"
Convert before branching point
$ convert 3
initializing destination A-r3-hg repository
scanning source...
sorting...
converting...
3 removeb
2 changeaa
1 branch, changeaaa
0 addc,changeaaaa
o 3 addc,changeaaaa files: a c
|
o 2 branch, changeaaa files: a
|
o 1 changeaa files: a
|
o 0 removeb files: a
Convert before branching point
$ convert 4
initializing destination A-r4-hg repository
scanning source...
sorting...
converting...
2 changeaa
1 branch, changeaaa
0 addc,changeaaaa
o 2 addc,changeaaaa files: a c
|
o 1 branch, changeaaa files: a
|
o 0 changeaa files: a
Convert at branching point
$ convert 5
initializing destination A-r5-hg repository
scanning source...
sorting...
converting...
1 branch, changeaaa
0 addc,changeaaaa
o 1 addc,changeaaaa files: a c
|
o 0 branch, changeaaa files: a
Convert last revision only
$ convert 6
initializing destination A-r6-hg repository
scanning source...
sorting...
converting...
0 addc,changeaaaa
o 0 addc,changeaaaa files: a c