Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8976649076753981440

=== Auto-CCing suspected CL author rockot@chromium.org ===

Hi rockot@chromium.org, the bisect results pointed to your CL, please take a look at the
results.


=== BISECT JOB RESULTS ===
Perf regression found with culprit

Suspected Commit
  Author : Ken Rockot
  Commit : 6f2ba28da2225847e4769fb89aad0522d53a7f41
  Date   : Wed Jun 07 02:30:44 2017
  Subject: Introduce Mojo.bindInterface IDL and related test API

Bisect Details
  Configuration: mac_10_11_perf_bisect
  Benchmark    : v8.runtimestats.browsing_desktop
  Metric       : v8-gc-phantom-handle-callback_sum/browse_search/browse_search_google
  Change       : 26.73% | 1.38666666667 -> 1.75733333333

Revision             Result                   N
chromium@477448      1.38667 +- 0.105429      6      good
chromium@477505      1.43367 +- 0.155214      6      good
chromium@477513      1.56183 +- 0.16561       6      good
chromium@477517      1.46983 +- 0.120527      6      good
chromium@477519      1.554 +- 0.348683        9      good
chromium@477520      1.81717 +- 0.178406      6      bad       <--
chromium@477534      1.83233 +- 0.175343      6      bad
chromium@477562      1.75733 +- 0.145977      6      bad

To Run This Test
  src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests --story-filter=browse.search.google v8.runtimestats.browsing_desktop

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8976649076753981440

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=6017970314149888


| O O | Visit http://www.chromium.org/developers/speed-infra/perf-bug-faq
|  X  | for more information addressing perf regression bugs. For feedback,
| / \ | file a bug with component Speed>Bisection.  Thank you!

I have no idea what this benchmark even means, but I don't think the bisect is accurate.

For completeness: Graphs: https://chromeperf.appspot.com/group_report?bug_id=733952

The benchmark measures time that is spent by V8 in in processing weak handles with finalizers attached.

The bisect clearly point to this CL but lets try another one. In any case, this is just one benchmark on one machine, so we can close this as WontFix.

Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8976360144430589888

=== BISECT JOB RESULTS ===
Perf regression found with culprit

Suspected Commit
  Author : Ken Rockot
  Commit : 6f2ba28da2225847e4769fb89aad0522d53a7f41
  Date   : Wed Jun 07 02:30:44 2017
  Subject: Introduce Mojo.bindInterface IDL and related test API

Bisect Details
  Configuration: mac_10_11_perf_bisect
  Benchmark    : v8.runtimestats.browsing_desktop
  Metric       : v8-gc-phantom-handle-callback_sum/browse_search/browse_search_google
  Change       : 21.65% | 1.46516666667 -> 1.78233333333

Revision             Result                    N
chromium@477400      1.46517 +- 0.22007        6      good
chromium@477500      1.4695 +- 0.143908        6      good
chromium@477513      1.56383 +- 0.0965652      6      good
chromium@477519      1.5445 +- 0.219507        6      good
chromium@477520      1.82317 +- 0.0875033      6      bad       <--
chromium@477521      1.771 +- 0.143527         9      bad
chromium@477522      1.81233 +- 0.205366       6      bad
chromium@477525      1.784 +- 0.115421         6      bad
chromium@477550      1.75433 +- 0.153406       6      bad
chromium@477600      1.78233 +- 0.173992       6      bad

To Run This Test
  src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests --story-filter=browse.search.google v8.runtimestats.browsing_desktop

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8976360144430589888

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5846548908343296


| O O | Visit http://www.chromium.org/developers/speed-infra/perf-bug-faq
|  X  | for more information addressing perf regression bugs. For feedback,
| / \ | file a bug with component Speed>Bisection.  Thank you!

There's definitely something happening here. Can you check whether it is intentional that your code also changed production behavior?

It sure looks like something's happening from the bisect, but it still doesn't make any sense.

The change adds new JS API surface only to test environments or runtime environments with a specific Blink feature enabled behind commandline flags.

It's pretty trivial to verify by looking at the CL that it doesn't actually do anything interesting to production environments, modulo some kind of egregious bug in generated IDL bindings.

dcheng@ can you please take a look at the CL as someone with more Blink expertise and see if I'm overlooking something?

Yeah... I'm not sure I see anything particularly related. The changes in non-test code are trivial, and certainly not something I'd expect to have an effect on v8 finalization.

And the interfaces shouldn't be exposed...

+jbroman@ for another pair of eyes

Again, feel free to ignore since the regression is very isolated and minor.

Was just wondering since the change is actually binding related and the metric is about executing finalizers.

Yeah, this is . . . surprising. I wouldn't expect a noticeable change from this. Not sure what's up here.