I think that this is a full list of broken locales (out of all the ones returned by l10n_util::GetAvailableLocales()):

de-CH
en-150
en-GB
en-IN
en-NZ
en-ZA
es-419
fr-CA

Per try jobs with my test, this looks like it's broken on all platforms (and for the same locales, I think).

Thank you for the report and sorry for getting to this late. 

I'll look into the ICU data. 

> en_GB only defines "gregorian"

That should be fine (even if generic is necessary for this API) because the parent of en_GB is en_001 which has 'generic'. 

It's possible that trim_data.sh messes up something as you suspected. 

Measure unit data come from data/unit. Even before trimming, en_001 (the parent of en_GB) does not have entries for {units,unitsNarrow}.duration.{day,hour,month,second} while it does have unitsShort.duration.*.  
en.txt does have {units, unitsNarrow}.duration.* as well as unitsShort.duration.*.   I guess the fallback path is en_GB => en_001 => en. 

Because unit*.duration.* are kept (not trimmed away), trimming does not make any difference (at least, should not; and I couldn't find any difference). 

Nonetheless, with the full ICU data, Dan's test passes while with Chrome's trimmed data it fails. 

I may be hitting a latent bug in ICU's fallback algorithm (en_GB => en_001 => en)  with Chrome's trimmed ICU data file. 

I think I found a likely cause. With data trimming, en_001 (the parent of en_GB) ends up having an empty units and unitsNarrow. Because Chrome only needs duration and en_001 does not have any duration entry for units/unitsNarrow to start with, they become empty after trimming.  And, this empty units and unitsNarrow blocks ICU from falling back to en.txt. 

    units{
    }
    unitsNarrow{
    }

The same is true of all other locales listed in comment 1. 

That was it. I rebuilt the ICU data and Dan's test passes for all locales. 

> I tried to build them locally using the instructions in src/third_party/icu/README.chromium but was unsuccessful.

Sorry about that. source/data/trnslit/trnslocal.mk should have root_subset.txt instead of css3transform.txt.  That file was left not updated. 

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/deps/icu.git/+/d5c238dcc2e801210749f3bb421b9227fe1c0948

commit d5c238dcc2e801210749f3bb421b9227fe1c0948
Author: Jungshik Shin <jshin@chromium.org>
Date: Tue Apr 11 21:58:59 2017

Update trim_data to deal with locale fallback failure for units

Delete empty units,units{Narrow,Short} blocks after trimming units data.
Empty units* blocks in en_GB and a few other locales after trimming
causes ICU to fail to fall back to get the duration data for those
locales.

In addition, fix source/data/translit/root_subset.txt. Rule*Ids block has
to be present even though it's empty. When dropping Hans-Hant transform
rules, root_subset.txt was changed to be completely empty, which broke
"components_unittests --g_test_filter=AutofillProfileComparato*" .

With these changes, regenerate ICU data files. The size is slightly smaller.

android/icudtl.dat  6573872 => 6573792
common/icudt*dat    10130560 => 10130480

BUG= 707515 , 677043 , 684609 
TEST=components_unittests --gtest_filter=AutofillProfileComparato*
TEST=ui_base_unittests --gtest_filter=L10nUtilTest.TimeDurationForm*
R=derat@chromium.org

Review-Url: https://codereview.chromium.org/2812943003 .

[modify] https://crrev.com/d5c238dcc2e801210749f3bb421b9227fe1c0948/android/icudtl.dat
[modify] https://crrev.com/d5c238dcc2e801210749f3bb421b9227fe1c0948/common/icudtb.dat
[modify] https://crrev.com/d5c238dcc2e801210749f3bb421b9227fe1c0948/common/icudtl.dat
[modify] https://crrev.com/d5c238dcc2e801210749f3bb421b9227fe1c0948/scripts/trim_data.sh
[modify] https://crrev.com/d5c238dcc2e801210749f3bb421b9227fe1c0948/source/data/translit/root_subset.txt

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/a8550bc1391772685da6e86adc72a587bd964289

commit a8550bc1391772685da6e86adc72a587bd964289
Author: derat <derat@chromium.org>
Date: Tue Apr 11 22:11:31 2017

Add test for TimeDurationFormat() error.

base::TimeDurationFormat() fails due to a
U_RESOURCE_TYPE_MISMATCH error from ICU when called with
certain locals like en_GB due to missing resource data.
Add a test (disabled for now) to verify that it works
for all supported locales.

BUG= 677043 , 707515 

Review-Url: https://codereview.chromium.org/2793613004
Cr-Commit-Position: refs/heads/master@{#463796}

[modify] https://crrev.com/a8550bc1391772685da6e86adc72a587bd964289/ui/base/l10n/l10n_util_unittest.cc

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/cf65b91bc7496b3dd2c958c890984e73c26ad22c

commit cf65b91bc7496b3dd2c958c890984e73c26ad22c
Author: jshin <jshin@chromium.org>
Date: Wed Apr 12 00:34:12 2017

Roll third_party/icu from 450be73 to b34251f

http://chromium.googlesource.com/chromium/deps/icu.git/+log/450be73..b34251f

Changes include (along with minor data build script updates):

  1. f0449ad Update IANA timezone db to 2017a from 2016i
  2. c781b5f Drops unused Hans-Hant ICU transliteration data.
     ( cuts down the ICU data file size by ~25kB.)
  3. d5c238d Update trim_data to deal with locale fallback failure for units
     and fix root_subset.txt in ICU's transliteration configuration.
  4. b34251f Update timezone db to 2017b

BUG= 684609 ,473288, 707515 , 677043 
TEST=components_unittests --gtest_filter=AutofillProfileComparato*
TEST=ui_base_unittests --gtest_filter=L10nUtilTest.TimeDurationForm*
TEST=For timezone change test, see CL #1 and #4 TEST lines.
TBR=riesa@chromium.org,derat@chromium.org

Review-Url: https://codereview.chromium.org/2755963002
Cr-Commit-Position: refs/heads/master@{#463856}

[modify] https://crrev.com/cf65b91bc7496b3dd2c958c890984e73c26ad22c/DEPS
[modify] https://crrev.com/cf65b91bc7496b3dd2c958c890984e73c26ad22c/ui/base/l10n/l10n_util_unittest.cc

Will ask for merge to M58 after baking in canary for a day or two. Just a DEPS roll would suffice in M58 branch. 

Verified in a canary build (en-GB locale: in the ash-tray, I confirmed that '1% - 2:28 until full' instead of '1% -  until full' is shown;  see  bug 677043 ). I also confirmed it in es-419 locale. 

Requesting for merge to M58.  It's rather late in the cycle, but it has regressed and affects en-GB and es-419 users. (again, see  bug 677043 ). 

This bug requires manual review: We are only 6 days from stable.
Please contact the milestone owner if you have questions.
Owners: amineer@(Android), cmasso@(iOS), bhthompson@(ChromeOS), govind@(Desktop)

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot

There is a huge risk in taking a change post branch as we're cutting M58 stable RC today. Can you please confirm if this is critical? Based on comment 12, seems like it's already verified in Canary. Is there enough unit test coverage for this?

And also seems like this bug has been there since M56? My recommendation is to wait until M59, but if you think it's critical and it's a safe merge, please let us know. 

Sorry that I forgot to give how safe/complicated a cherry-pick is. 

It'll be just a ICU data rebuild with no code change. Dan added a unittest to verify the fix (ICU).  

So,  I think it's pretty safe. It's unfortunate that it's been broken since M56, but it's  bad for en-GB and es-419 users.(no remaining hours for battery. how long to wait until full). 

Since change is affecting Chrome OS more, adding bhthompson@ for merge review. 

As we still have a beta before stable for CrOS, if engineering is confident this is a clean cut safe fix, we can merge this if it is only for CrOS. 

The following revision refers to this bug:
  https://chrome-internal.googlesource.com/chrome/tools/buildspec/+/cd0a1e71a466bf114861203e987fb7ed07e5efd6

commit cd0a1e71a466bf114861203e987fb7ed07e5efd6
Author: Jungshik Shin <jungshik@google.com>
Date: Thu Apr 20 19:47:02 2017