( )⚙ D7121 copies: get copies information directly from _copies

This is an archive of the discontinued Mercurial Phabricator instance.

Differential D7121

copies: get copies information directly from _copies
ClosedPublic

Authored by marmoute on Oct 16 2019, 12:54 PM.

Details

Reviewers
martinvonz
Group Reviewers
hg-reviewers
Commits
rHG82dabad535d2: copies: get copies information directly from _copies
rHG32943e8c8d68: copies: get copies information directly from _copies
Summary

Performance measurement does not show any significant performance movement. This
is not surprising since p1copies() code is self._copies[0].

Diff Detail

Repository
rHG Mercurial
Lint
Automatic diff as part of commit; lint not applicable.
Unit
Automatic diff as part of commit; unit tests not applicable.

Event Timeline

marmoute created this revision.Oct 16 2019, 12:54 PM
martinvonz accepted this revision.Oct 16 2019, 1:12 PM
martinvonz added a subscriber: martinvonz.

Performance measurement does not show any significant performance movement. This
is not surprising since p1copies() code is self._copies[0].

My testing agrees with that. However, making the same change (well, the reverse of it) on top of D7070 makes very significant difference. I have no idea why.

This revision is now accepted and ready to land.Oct 16 2019, 1:12 PM
martinvonz added a comment.Oct 16 2019, 1:15 PM
In D7121#104543, @martinvonz wrote:

Performance measurement does not show any significant performance movement. This
is not surprising since p1copies() code is self._copies[0].

My testing agrees with that. However, making the same change (well, the reverse of it) on top of D7070 makes very significant difference. I have no idea why.

Actually, I may have spoken too soon. hg perfpathcopies FIREFOX_BETA_59_END FIREFOX_BETA_60_BASE --config perf.run-limits='10.0-10' yields:

before:
! wall 4.595685 comb 4.600000 user 4.530000 sys 0.070000 (best of 10)
after:
! wall 4.485065 comb 4.490000 user 4.440000 sys 0.050000 (best of 10)

So there seems to be a measurable difference. Could just the function call overhead be that whole difference? Anyway, it doesn't seem worth spending time trying to figure out.

marmoute added a commit: rHG32943e8c8d68: copies: get copies information directly from _copies.Oct 16 2019, 1:33 PM
Closed by commit rHG32943e8c8d68: copies: get copies information directly from _copies (authored by marmoute). · Explain Why
This revision was automatically updated to reflect the committed changes.
marmoute updated this revision to Diff 17271.Oct 16 2019, 5:49 PM
marmoute added a comment.Oct 16 2019, 5:57 PM
In D7121#104546, @martinvonz wrote:
In D7121#104543, @martinvonz wrote:

Performance measurement does not show any significant performance movement. This
is not surprising since p1copies() code is self._copies[0].

My testing agrees with that. However, making the same change (well, the reverse of it) on top of D7070 makes very significant difference. I have no idea why.

Actually, I may have spoken too soon. hg perfpathcopies FIREFOX_BETA_59_END FIREFOX_BETA_60_BASE --config perf.run-limits='10.0-10' yields:

before:
! wall 4.595685 comb 4.600000 user 4.530000 sys 0.070000 (best of 10)
after:
! wall 4.485065 comb 4.490000 user 4.440000 sys 0.050000 (best of 10)

So there seems to be a measurable difference. Could just the function call overhead be that whole difference? Anyway, it doesn't seem worth spending time trying to figure out.

Function call can have a large overhead, how many revision are we talking about here ?

martinvonz added a comment.Oct 16 2019, 5:59 PM
In D7121#104671, @marmoute wrote:
In D7121#104546, @martinvonz wrote:
In D7121#104543, @martinvonz wrote:

Performance measurement does not show any significant performance movement. This
is not surprising since p1copies() code is self._copies[0].

My testing agrees with that. However, making the same change (well, the reverse of it) on top of D7070 makes very significant difference. I have no idea why.

Actually, I may have spoken too soon. hg perfpathcopies FIREFOX_BETA_59_END FIREFOX_BETA_60_BASE --config perf.run-limits='10.0-10' yields:

before:
! wall 4.595685 comb 4.600000 user 4.530000 sys 0.070000 (best of 10)
after:
! wall 4.485065 comb 4.490000 user 4.440000 sys 0.050000 (best of 10)

So there seems to be a measurable difference. Could just the function call overhead be that whole difference? Anyway, it doesn't seem worth spending time trying to figure out.

Function call can have a large overhead, how many revision are we talking about here ?

5627

marmoute added a comment.Oct 16 2019, 6:26 PM
In D7121#104672, @martinvonz wrote:
In D7121#104671, @marmoute wrote:
In D7121#104546, @martinvonz wrote:

[…]

So there seems to be a measurable difference. Could just the function call overhead be that whole difference? Anyway, it doesn't seem worth spending time trying to figure out.

Function call can have a large overhead, how many revision are we talking about here ?

5627

That is not much to be honest. How do you measure this?

martinvonz added a comment.Oct 16 2019, 6:31 PM
In D7121#104673, @marmoute wrote:
In D7121#104672, @martinvonz wrote:
In D7121#104671, @marmoute wrote:
In D7121#104546, @martinvonz wrote:

[…]

So there seems to be a measurable difference. Could just the function call overhead be that whole difference? Anyway, it doesn't seem worth spending time trying to figure out.

Function call can have a large overhead, how many revision are we talking about here ?

5627

That is not much to be honest. How do you measure this?

By running exactly the command I shared earlier :P This is a version of mozilla-unified I've created by running hg convert --config experimental.copies.write-to=changeset-only.

marmoute added a comment.Oct 16 2019, 6:39 PM
In D7121#104674, @martinvonz wrote:
In D7121#104673, @marmoute wrote:
In D7121#104672, @martinvonz wrote:
In D7121#104671, @marmoute wrote:
In D7121#104546, @martinvonz wrote:

[…]

So there seems to be a measurable difference. Could just the function call overhead be that whole difference? Anyway, it doesn't seem worth spending time trying to figure out.

Function call can have a large overhead, how many revision are we talking about here ?

5627

That is not much to be honest. How do you measure this?

By running exactly the command I shared earlier :P This is a version of mozilla-unified I've created by running hg convert --config experimental.copies.write-to=changeset-only.

I meant the number of changeset :-)

marmoute added a comment.Oct 16 2019, 6:41 PM
In D7121#104675, @marmoute wrote:
In D7121#104674, @martinvonz wrote:
In D7121#104673, @marmoute wrote:
In D7121#104672, @martinvonz wrote:
In D7121#104671, @marmoute wrote:
In D7121#104546, @martinvonz wrote:

[…]

So there seems to be a measurable difference. Could just the function call overhead be that whole difference? Anyway, it doesn't seem worth spending time trying to figure out.

Function call can have a large overhead, how many revision are we talking about here ?

5627

That is not much to be honest. How do you measure this?

By running exactly the command I shared earlier :P This is a version of mozilla-unified I've created by running hg convert --config experimental.copies.write-to=changeset-only.

I meant the number of changeset :-)

I did a fresh clone and fished this myself. I indeed see the same number

hg log -T '.\n' -r FIREFOX_BETA_60_BASE%FIREFOX_BETA_59_END | wc -l

Revision Contents
Changeset List

PathPackages
Mmercurial/copies.py (5 lines)

Diff 17247

mercurial/copies.py

alwaysmatch = match.always() alwaysmatch = match.always()
while work: while work:
r = heapq.heappop(work) r = heapq.heappop(work)
copies = all_copies.pop(r) copies = all_copies.pop(r)
if r == b.rev(): if r == b.rev():
return copies return copies
for i, c in enumerate(children[r]): for i, c in enumerate(children[r]):
childctx = repo[c] childctx = repo[c]
p1copies, p2copies = childctx._copies
if r == childctx.p1().rev(): if r == childctx.p1().rev():
parent = 1 parent = 1
childcopies = childctx.p1copies() childcopies = p1copies
else: else:
assert r == childctx.p2().rev() assert r == childctx.p2().rev()
parent = 2 parent = 2
childcopies = childctx.p2copies() childcopies = p2copies
if not alwaysmatch: if not alwaysmatch:
childcopies = { childcopies = {
dst: src for dst, src in childcopies.items() if match(dst) dst: src for dst, src in childcopies.items() if match(dst)
} }
# Copy the dict only if later iterations will also need it # Copy the dict only if later iterations will also need it
if i != len(children[r]) - 1: if i != len(children[r]) - 1:
newcopies = copies.copy() newcopies = copies.copy()
else: else: