Add benchmark to 'path' package, of the most common methods.

Add benchmark to 'path' package, of the most common methods.

BUG=
R=nweiz@google.com, rnystrom@google.com

Committed: https://code.google.com/p/dart/source/detail?r=39321

[Anders Johnsen, 2014-08-05 14:23:20]

Note that I did not call it _test, as we should not run this when running normal
tests (currently takes ~30 seconds).

[Bob Nystrom, 2014-08-05 15:41:58]

A couple of suggestions, but LGTM!

https://codereview.chromium.org/443643002/diff/1/pkg/path/test/benchmark.dart
File pkg/path/test/benchmark.dart (right):

https://codereview.chromium.org/443643002/diff/1/pkg/path/test/benchmark.dart...
pkg/path/test/benchmark.dart:27: group(style.name, () {
I'd consider not using unittest here since it doesn't add much value. Instead,
just add the style name to the string you interpolate in benchmark() and be done
with it?

https://codereview.chromium.org/443643002/diff/1/pkg/path/test/benchmark.dart...
pkg/path/test/benchmark.dart:75: 'C:\\User\\me\\',
Nit: use a raw string here and below?

[nweiz, 2014-08-05 19:40:33]

https://codereview.chromium.org/443643002/diff/1/pkg/path/test/benchmark.dart
File pkg/path/test/benchmark.dart (right):

https://codereview.chromium.org/443643002/diff/1/pkg/path/test/benchmark.dart...
pkg/path/test/benchmark.dart:76:
'\\server\\share\\my\\folders\\some\\file.data',
It doesn't seem like we should be running Windows paths against posix and URL
styles. It also seems like we should be testing URL-specific paths as well.

[nweiz, 2014-08-05 19:41:10]

Also: this should be in benchmark/, not test/.

[Anders Johnsen, 2014-08-06 07:42:27]

PTAL

FYI, https://codereview.chromium.org/439223002/ does the following to isAbsolute
and rootPrefix:

Without:
$ dart pkg/path/benchmark/benchmark.dart rootPrefix isAbsolute
posix-rootPrefix: 0.4179151882916881 iter/us (0:00:00.239838)
posix-isAbsolute: 0.45271811959001845 iter/us (0:00:00.220895)
url-rootPrefix: 0.4040991821032554 iter/us (0:00:00.247472)
url-isAbsolute: 0.4040289769582274 iter/us (0:00:00.247532)
windows-rootPrefix: 0.3380799761991697 iter/us (0:00:00.295796)
windows-isAbsolute: 0.34164323563475607 iter/us (0:00:00.292709)

With:
$ dart pkg/path/benchmark/benchmark.dart rootPrefix isAbsolute
posix-rootPrefix: 4.2865103519225 iter/us (0:00:00.023856)
posix-isAbsolute: 9.106638739641198 iter/us (0:00:00.010988)
url-rootPrefix: 3.0705929314950717 iter/us (0:00:00.032574)
url-isAbsolute: 3.391094984570518 iter/us (0:00:00.029495)
windows-rootPrefix: 5.104384666428462 iter/us (0:00:00.019597)
windows-isAbsolute: 6.214654154496302 iter/us (0:00:00.016097)

which is pretty nice.

https://codereview.chromium.org/443643002/diff/1/pkg/path/test/benchmark.dart
File pkg/path/test/benchmark.dart (right):

https://codereview.chromium.org/443643002/diff/1/pkg/path/test/benchmark.dart...
pkg/path/test/benchmark.dart:27: group(style.name, () {
On 2014/08/05 15:41:58, Bob Nystrom wrote:
> I'd consider not using unittest here since it doesn't add much value. Instead,
> just add the style name to the string you interpolate in benchmark() and be
done
> with it?

Done.

https://codereview.chromium.org/443643002/diff/1/pkg/path/test/benchmark.dart...
pkg/path/test/benchmark.dart:75: 'C:\\User\\me\\',
On 2014/08/05 15:41:58, Bob Nystrom wrote:
> Nit: use a raw string here and below?

Done.

I think we should continue to add to this list over time, as having a nice large
set of paths will make this benchmark a lot more valuable.

If you have other paths that you want to have in, initially, please let me know.

https://codereview.chromium.org/443643002/diff/1/pkg/path/test/benchmark.dart...
pkg/path/test/benchmark.dart:76:
'\\server\\share\\my\\folders\\some\\file.data',
On 2014/08/05 19:40:33, nweiz wrote:
> It doesn't seem like we should be running Windows paths against posix and URL
> styles. It also seems like we should be testing URL-specific paths as well.

Adding some url paths as well.

I'd actually prefer to use the same paths for all methods. That makes it
possible to see the extra overhead between platforms. E.g., if windows is 3x
slower than posix, it can indicate that it may be less optimized than posix.
Interestingly enough, it may also indicate that the differences (root, path
separators) may cause an extra overhead.

Overall, I think the value of using the same paths outweighs the concerns.

[nweiz, 2014-08-06 21:08:28]

https://codereview.chromium.org/443643002/diff/1/pkg/path/test/benchmark.dart
File pkg/path/test/benchmark.dart (right):

https://codereview.chromium.org/443643002/diff/1/pkg/path/test/benchmark.dart...
pkg/path/test/benchmark.dart:76:
'\\server\\share\\my\\folders\\some\\file.data',
On 2014/08/06 07:42:27, Anders Johnsen wrote:
> On 2014/08/05 19:40:33, nweiz wrote:
> > It doesn't seem like we should be running Windows paths against posix and
URL
> > styles. It also seems like we should be testing URL-specific paths as well.
> 
> Adding some url paths as well.
> 
> I'd actually prefer to use the same paths for all methods. That makes it
> possible to see the extra overhead between platforms. E.g., if windows is 3x
> slower than posix, it can indicate that it may be less optimized than posix.
> Interestingly enough, it may also indicate that the differences (root, path
> separators) may cause an extra overhead.
> 
> Overall, I think the value of using the same paths outweighs the concerns.

I don't think you should take out the platform-specific paths, just that you
should only run them on contexts that support them. I don't think running
path.posix.dirname('C:\\User\\me\\') tells us anything particularly useful.

[Anders Johnsen, 2014-08-07 08:56:10]

https://codereview.chromium.org/443643002/diff/1/pkg/path/test/benchmark.dart
File pkg/path/test/benchmark.dart (right):

https://codereview.chromium.org/443643002/diff/1/pkg/path/test/benchmark.dart...
pkg/path/test/benchmark.dart:76:
'\\server\\share\\my\\folders\\some\\file.data',
On 2014/08/06 21:08:28, nweiz wrote:
> On 2014/08/06 07:42:27, Anders Johnsen wrote:
> > On 2014/08/05 19:40:33, nweiz wrote:
> > > It doesn't seem like we should be running Windows paths against posix and
> URL
> > > styles. It also seems like we should be testing URL-specific paths as
well.
> > 
> > Adding some url paths as well.
> > 
> > I'd actually prefer to use the same paths for all methods. That makes it
> > possible to see the extra overhead between platforms. E.g., if windows is 3x
> > slower than posix, it can indicate that it may be less optimized than posix.
> > Interestingly enough, it may also indicate that the differences (root, path
> > separators) may cause an extra overhead.
> > 
> > Overall, I think the value of using the same paths outweighs the concerns.
> 
> I don't think you should take out the platform-specific paths, just that you
> should only run them on contexts that support them. I don't think running
> path.posix.dirname('C:\\User\\me\\') tells us anything particularly useful.

Done.

[Anders Johnsen, 2014-08-07 08:56:22]

PTAL

[nweiz, 2014-08-12 21:31:40]

lgtm

Thanks!

[Anders Johnsen, 2014-08-18 06:32:56]

Committed patchset #5 manually as 39321 (presubmit successful).