Yuya Nishihara <yuya@tcha.org> [Sun, 23 Apr 2017 13:08:58 +0900] rev 33946
encoding: add fast path of from/toutf8b() for ASCII strings
See the previous patch for why.
The added test seems not making much sense because ASCII strings should
never contain "\xed" and be valid UTF-8.
(with mercurial repo)
$ export HGRCPATH=/dev/null HGPLAIN=
$ hg log --time --config experimental.stabilization=all -Tjson > /dev/null
(original)
time: real 6.830 secs (user 6.740+0.000 sys 0.080+0.000)
time: real 6.690 secs (user 6.650+0.000 sys 0.040+0.000)
time: real 6.700 secs (user 6.640+0.000 sys 0.060+0.000)
(fast jsonescape)
time: real 5.630 secs (user 5.550+0.000 sys 0.070+0.000)
time: real 5.700 secs (user 5.650+0.000 sys 0.050+0.000)
time: real 5.690 secs (user 5.640+0.000 sys 0.050+0.000)
(this patch)
time: real 5.190 secs (user 5.120+0.000 sys 0.070+0.000)
time: real 5.230 secs (user 5.170+0.000 sys 0.050+0.000)
time: real 5.220 secs (user 5.150+0.000 sys 0.070+0.000)
Yuya Nishihara <yuya@tcha.org> [Sun, 23 Apr 2017 13:06:23 +0900] rev 33945
encoding: add fast path of from/tolocal() for ASCII strings
This is micro optimization, but seems not bad since to/fromlocal() is called
lots of times and isasciistr() is cheap and simple.
We boldly assume that any non-ASCII characters have at least one 8-bit byte.
This isn't true for some email character sets (e.g. ISO-2022-JP and UTF-7),
but I believe no such encodings are used as a platform default. Shift_JIS,
a major crap, is okay as it should have a leading byte in 0x80-0xff range.
(with mercurial repo)
$ export HGRCPATH=/dev/null HGPLAIN=
$ hg log --time --config experimental.stabilization=all > /dev/null
(original)
time: real 7.460 secs (user 7.420+0.000 sys 0.030+0.000)
time: real 7.670 secs (user 7.590+0.000 sys 0.080+0.000)
time: real 7.560 secs (user 7.510+0.000 sys 0.040+0.000)
(this patch)
time: real 7.340 secs (user 7.260+0.000 sys 0.060+0.000)
time: real 7.260 secs (user 7.210+0.000 sys 0.030+0.000)
time: real 7.310 secs (user 7.260+0.000 sys 0.060+0.000)
Yuya Nishihara <yuya@tcha.org> [Sun, 23 Apr 2017 12:59:42 +0900] rev 33944
encoding: add function to test if a str consists of ASCII characters
Most strings are ASCII. Let's optimize for it.
Using uint64_t is slightly faster than uint32_t on 64bit system, but there
isn't huge difference.
Yuya Nishihara <yuya@tcha.org> [Sun, 23 Apr 2017 14:47:52 +0900] rev 33943
encoding: add fast path of jsonescape() (issue5533)
This isn't highly optimized as it copies characters one by one, but seems
reasonably simple and not slow.
(with mercurial repo)
$ export HGRCPATH=/dev/null HGPLAIN=
$ hg log --time --config experimental.stabilization=all -Tjson > /dev/null
(original)
time: real 6.830 secs (user 6.740+0.000 sys 0.080+0.000)
time: real 6.690 secs (user 6.650+0.000 sys 0.040+0.000)
time: real 6.700 secs (user 6.640+0.000 sys 0.060+0.000)
(this patch)
time: real 5.630 secs (user 5.550+0.000 sys 0.070+0.000)
time: real 5.700 secs (user 5.650+0.000 sys 0.050+0.000)
time: real 5.690 secs (user 5.640+0.000 sys 0.050+0.000)
Yuya Nishihara <yuya@tcha.org> [Sun, 23 Apr 2017 16:10:51 +0900] rev 33942
encoding: extract stub for fast JSON escape
This moves JSON character maps to pure/charencode.py because they will be
used only when the fast-path fails.
Jun Wu <quark@fb.com> [Thu, 24 Aug 2017 21:43:54 -0700] rev 33941
journal: do not use atomictemp (issue5338)
Writing journal files using `atomictemp` leads to quadratic performance that
could be problematic if automation runs many commands. Other logs like
blackbox does not use atomictemp, and journal logs are not critical for repo
correctness. So let's make them non-atomictemp.
Differential Revision: https://phab.mercurial-scm.org/D517
Augie Fackler <raf@durin42.com> [Tue, 22 Aug 2017 14:26:25 -0400] rev 33940
tests: update test-commandserver to pass our import checker
Augie Fackler <raf@durin42.com> [Tue, 22 Aug 2017 16:37:56 -0400] rev 33939
tests: add blank output lines to test-commandserver
This helps the import checker notice when a Python heredoc has
ended. It's not perfect, but it gets nasty quick and I think the
better fix would be to add a real parser for .t-tests, which will have
to wait.