Auto-assigning to diagnose & triage.

Logs from the number formatting tests in 58611 and 58597:

[ RUN      ] NumberFormattingTest.FormatPercent
../../base/i18n/number_formatting_unittest.cc:130: Failure
Expected equality of these values:
  WideToUTF16(cases[i].expected_arabic)
    Which is: ٠٪؜
  FormatPercent(cases[i].number)
    Which is: 0‎%‎
Stack trace:
#00: CurrentOsStackTraceExceptTop at ../../third_party/googletest/src/googletest/src/gtest.cc:810
#01: operator= at ../../third_party/googletest/src/googletest/src/gtest.cc:383
#02: TestBody at ../../base/i18n/number_formatting_unittest.cc:129

[00027.824] pkgsvr: 2018/07/09 18:31:18 pkgfs:unsupported(/packages/base_unittests/0): dir unlink "app.log"
../../base/i18n/number_formatting_unittest.cc:130: Failure
[00027.905] pkgsvr: 2018/07/09 18:31:18 pkgfs:unsupported(/packages/base_unittests/0): dir unlink "app.log"
Expected equality of these values:
[00027.943] pkgsvr: 2018/07/09 18:31:18 pkgfs:unsupported(/packages/base_unittests/0): dir unlink "app.log"
  WideToUTF16(cases[i].expected_arabic)
    Which is: ٤٢٪؜
  FormatPercent(cases[i].number)
    Which is: 42‎%‎
Stack trace:
#00: CurrentOsStackTraceExceptTop at ../../third_party/googletest/src/googletest/src/gtest.cc:810
#01: operator= at ../../third_party/googletest/src/googletest/src/gtest.cc:383
#02: TestBody at ../../base/i18n/number_formatting_unittest.cc:129

[00027.981] pkgsvr: 2018/07/09 18:31:18 pkgfs:unsupported(/packages/base_unittests/0): dir unlink "app.log"
../../base/i18n/number_formatting_unittest.cc:130: Failure
[00028.008] pkgsvr: 2018/07/09 18:31:18 pkgfs:unsupported(/packages/base_unittests/0): dir unlink "app.log"
Expected equality of these values:
[00028.098] pkgsvr: 2018/07/09 18:31:18 pkgfs:unsupported(/packages/base_unittests/0): dir unlink "app.log"
  WideToUTF16(cases[i].expected_arabic)
    Which is: ١٬٠٢٤٪؜
  FormatPercent(cases[i].number)
    Which is: 1,024‎%‎
Stack trace:
#00: CurrentOsStackTraceExceptTop at ../../third_party/googletest/src/googletest/src/gtest.cc:810
#01: operator= at ../../third_party/googletest/src/googletest/src/gtest.cc:383
#02: TestBody at ../../base/i18n/number_formatting_unittest.cc:129

[00028.104] pkgsvr: 2018/07/09 18:31:18 pkgfs:unsupported(/packages/base_unittests/0): dir unlink "app.log"
[  FAILED  ] NumberFormattingTest.FormatPercent (6 ms)
[00028.174] pkgsvr: 2018/07/09 18:31:18 pkgfs:unsupported(/packages/base_unittests/0): dir unlink "app.log"
[635/3112] NumberFormattingTest.FormatPercent (6 ms)

Both NumberFormattingTest.FormatPercent and TimeFormattingTest.TimeFormatTimeOfDayDE were modified as part of an ICU dependency roll, at about the time that those flakes were observed, which is suspicious, possibly indicating a missing dependency e.g. on ICU data files, resulting in stale ones being used for some time immediately after the roll?

Broke out the TaskScheduler report to  issue 862582 , since that looks to be something different.

jshin: Looking at this flake, it seems to me that the failures are would be consistent with us running an out-of-date base_unittests against the newly-rolled ICU data - could you take a look at the NumberFormattingTest.FormatPercent failures from comment #2 and confirm that assessment?

If that fits these failures (which seem to be one-offs; there are no other reports on try-flakes, AFAICT) then we'll need to look for a missing dependency, likely somewhere in our Fuchsia package creation logic.

> consistent with us running an out-of-date base_unittests against the newly-rolled ICU data 

Yes, fuschia's base test is out-of-date. The expected values (for both DateFormat and NumberFormat) in that outdated base_unittests (as used by Fuschia) have to be updated to the trunk.   

Why does Fuschia have a separate update-cycle for base_unittests? 
Don't Fuscha trybots use the latest trunk of base_unittests ? 

> If that fits these failures (which seem to be one-offs; there are no other reports on try-flakes, AFAICT) then we'll need to look for a missing dependency

I'm not a good person to deal with the abov. Whoever that is has to take this bug. 

Re #4: We should have seen our base_unittests rebuilt in-sync with the ICU data-set change, so that things would have matched, but for some reason we didn't, and we're not sure why, but a missing GN dependency seems the obvious issue.

Re-assigning to someone on the Chrome-Fuchsia team to look at as a cleanup task, since this is unlikely to keep happening unless that dependency is changed often.

It seems that Fuchsia has a separate forked tree of ICU and it is still 61.1 instead of 62.1. I wonder why Fuchsia forked ICU from Chromium.   There must be up-sides to that, but there are also downside: 

Fuchsia's ICU update would lag behind Chromium's as found here. ICU bug fixes/security fixes wouldn't be propagated in time to Fuschia. 

Of course, there'd be no issue if Fuchsia's copy of ICU is updated as frequently as Chrome's :-)

Re #6: Are we actually using the system-provided ICU in the Chromium build, though?

To clarify, do we think the cause was an out of date base_unittests, or an updated base_unittests that was running against an old ICU data file?

If the latter, I wonder if this could be that we're loading the wrong .so (perhaps only in tests?) We do build icuuc and icui18n, but they're renamed because Fuchsia provides system ones in /system/lib already, that I assume would be 61 or older still.