+ehmaldonaldo, fyi, in case we hit some weird aspect to trying to recurse into another deps file in the same repo.

I don't think that's possible right now.
As it is, in order to recurse into a DEPS file, we need to add the parent repo as a dependency. So in this case Chromium would need Chromium as a dependency...

We'd need to add an "include" directive or something similar.

Disclaimer: I know nothing.

NB: the example given in the description doesn't affect //build, but //third_party only, as far as I understand.

But I guess //build and //third_party must be treated analogously anyway (same battle).

+Oleh, Dirk. 

Ok, some options after we discussed this via Hangouts:

1. Make //third_party and //build into their own repositories.
I like this idea the most. This would also alleviate another problem from WebRTC, since some of the changes to the code in here break WebRTC and it's hard to check before submitting.
I think it has been proposed before, but things didn't work out.

2. Tag the dependencies that are needed by //build, via a "tags" attribute for the dependency in the deps dict.
Then we could ask gclient to list the dependencies needed by //build and check that WebRTC is including all of them.

3. Add an include mechanism for gclient.
We could tell gclient to include DEPS files within the same repository. This would make things harder if we move to git submodules, since git submodules don't have an analogous mechanism.

Random remarks.

1. Turning //third_party into its own repository
It might not be future-proof, by lack of granularity. If a particular third party A depends on B, but WebRTC doesn't depends on A, we won't need B. 

2. Tag dependencies needed by //build and //third_party
The gn files already state dependencies when linking is involved.
* only tagging other 'special' dependencies would prevent duplication (DRY).
* we may put these annotations in .gn, to keep the following distinction:
  * gn files answer 'who need what?'
  * DEPS files answer 'which version and where to find it?'

> 1. Make //third_party and //build into their own repositories.

Are you talking about moving those directories out of src.git and into their own repos, or simply creating gsubtree mirrors? We do not want to do the former. The latter already exists for //build, but not for //third_party. 

> 2. Tag the dependencies that are needed by //build, via a "tags" attribute for
> the dependency in the deps dict.

This feels like it would be clunky and not scale very well, i.e. to N other consumers of //build.

> 3. Add an include mechanism for gclient.
> We could tell gclient to include DEPS files within the same repository. This would make things 
> harder if we move to git submodules, since git submodules don't have an analogous mechanism.

Is there a reason we couldn't change the implementation of recursedeps to support other DEPS files in the same repo?

I also don't understand the comment about submodules (or see the connection).

> Is there a reason we couldn't change the implementation of recursedeps to support other DEPS files in the same repo?

I don't see any. Currently, when a directory such as 'src/buildtools' is present in recursedeps (the list defined in DEPS file), it is bound to an URL:

deps['src/buildtools'] == 'https://chromium.googlesource.com/chromium/buildtools.git@babecafe'


I can think of two simple ways to declare an already checked-in directory in a parent DEPS file:

1/ Use a special tag instead of an URL. E.g:
   deps = {
            # ...
            'src/third_party': LOCAL_DEPENDENCY
            # ...
          }
2/ Introduce an auxiliary structure. E.g:
   local_deps = ['src/third_party',  
                 #...
                ]

If you have any preference or objection, raise your voice!

3/ Do nothing particular. I.e.:
* 'src/third_party' not in deps, meaning that's not a external dep we should pull
* 'src/third_party' in recursedeps, meaning we must dig into src/third_party/DEPS

A bit less explicit, but neater in my book (DRY).

If this were just about //src/build/ requiring that certain repos be DEPS'd in to //src/build/foo, and the fact that the gsubtreed mirror of //src/build doesn't include those DEPS, then there could be plenty of reasonable solutions.

But this is about //src/build/ requiring that certain repos be DEPS'd in to //src/third_party/foo. There is not any way for the gsubtreed mirror of //src/build/ to represent that in a sane manner, because that mirror repository cannot guarantee that it is checked out adjacent to something that looks like //src/third_party/.

So even if there were a way to do something like recursedeps[1] onto a DEPS file within the same repo, that would not be a good solution to this problem. The good solution to this problem is that, for //src/build/ to be used by other people, it needs to be self-contained. That's a problem for the developers of //src/build to solve, not gclient.

[1] nota bene: recursedeps would not be the solution here anyway. The semantics of recursedeps are such that it takes a list of things that are already defined as dependencies, and recurses upon them. Another DEPS file in a subdirectory of an already-checked-out repo is not a dependency (doesn't have a name, doesn't have a url, doesn't have to be processed in breadth-first order, etc.). This is not to say there isn't a solution here (as edward suggests, a new 'include' directive might work), but recursedeps and adding things to the deps dict aren't it.

Thanks for your input, here are some comments to challenge my understanding.

> for //src/build/ to be used by other people, it needs to be self-contained
For that, isn't it sufficient for //src/build to clearly expose its dependencies?

Also, can't we just amend the semantics of recursedeps? I would find it nice to see these orthogonal responsibilities:
 * deps dict defining dependencies to pull
 * recursedeps listing directories to recursively sync (making no difference with checked-in ones or sub-repo).

BTW the recursedeps unit tests demonstrate this latter behavior: they happily process sub-directories which just happen to be present (i.e. not pulled from a real repo).

I don't understand this sentence:

> There is not any way for the gsubtreed mirror of //src/build/ to represent
> that in a sane manner, because that mirror repository cannot guarantee that 
> it is checked out adjacent to something that looks like //src/third_party/.

Are you saying that there's no way to programmatically express this? If so, I don't think that's true? 

We already require build to be checked out next to third party, in the source code, so expressing this in DEPS files wouldn't change anything. And we've been sharing //build across multiple projects for a couple years.

One potential way to make this work with a new DEPS file would be to have //build/DEPS contain deps of the form 'src/third_party/foo', and this would work fine, it would "simply" require build to be checked out to src/build and third_party checked out to src/third_party. This is effectively what src-internal/DEPS does, for example, and I think this would work fine for, say, NaCl, and WebRTC, which both require a checkout rooted at 'src'. (I don't think this would work for V8, which I think doesn't require that).

A better alternative, though would probably be to do what ANGLE does, and define a variable that defaults to 'src', e.g., 'root', and have //build/DEPS refer to '${root}/third_party/foo'. Or, define a variable for the location of 'third_party' directly. You can then replicate the values of these variables out to GN files, and via the GN files (which are easily parseable) could expose that to other scripts that might need them. I think that probably would work for v8.

> The good solution to this problem is that, for //src/build/ to be used by other people, 
> it needs to be self-contained. 

That's an option, but it's limited. It would mean that in order for something to be used by src/build it would need to live in src/build, and thus if someone else wanted to *also* use that thing, they'd have to get it from src/build (or, alternatively, have two copies). 

I don't see the idea that if you want to use project X you also have to use project Y as problematic, and I think we already have the hooks we need to support this properly.

> The semantics of recursedeps are such that it takes a list of things that are already defined as
> dependencies, and recurses upon them.

I don't think there's anything in the *interface* of gclient that says that the DEPS file has to live in a different repo? The way I've viewed recursedeps is that it tells gclient to go recurse into 'this other DEPS file', not 'this other repo'. Perhaps I am odd here.

So, this would change the implementation and perhaps in a sense changes the semantics, but I'm not sure that it does in a way that would be upsetting or break anything? It feels just like a generalization to me, and better than adding two mechanisms that do the same thing but vary solely based on whether the file is in the same repo or a different one.

Another option, which doesn't require any code change.

3'/ 
* deps['src/third_party'] == None 
   # It must be in deps for recursion to take place, 
   # but it's neither an external repo nor a package.
* 'src/third_party' in recursedeps

C.f. https://chromium-review.googlesource.com/c/chromium/src/+/1121456 which exhibit a working solution.

Re #16:
An include would be easier to implement, so I'd prefer that.

Re #17:
Yes, that works but I would prefer not to do that. That's more of a bug than a feature ;)

Re #16 (cont.)
Though if you think it's clearer/better to use only recursedeps, we could do that.

let's see what agable@ says about #c16.

> An include would be easier to implement, so I'd prefer that.
Could you elaborate on that please? I don't see why, so I'm feeling there are some pieces I'm missing.

> Yes, that works but I would prefer not to do that. That's more of a bug than a feature ;)
Out of curiosity, what is the purpose of 'None' which is explicitly allowed by the schema?

> Though if you think it's clearer/better to use only recursedeps, we could do that.
+1 for recursedeps only :)
What do you think of option 3/ (#13)?

Michael, FYI. I'm not sure how this would affect the release scripts, since they do something with recursedeps.

> > Yes, that works but I would prefer not to do that. That's more of a bug than a feature ;)
> Out of curiosity, what is the purpose of 'None' which is explicitly allowed by the schema?
I think it is because it was used before, and we want to support those historical DEPS files.

Ok, so I've switched to option 3/ (comment #13), which doesn't use 'None' (unclear semantic / deprecated / whatever).

I might be biased, but like the separation of concerns :)
 * deps to declare external dependencies.
 * recursedeps to list which DEPS files to recurse into.

It requires a slight addition in gclient.py (https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/1124858)

It does the job and has a very low risk to break anything. If you really believe it might impede future changes/cleanups, I'm willing to implement the 'include' version which as far as I can see would do exactly the same thing, but iterating over an 'include' variable instead of 'recursedeps'.

The following revision refers to this bug:
  https://webrtc.googlesource.com/src.git/+/401afd51bb7742a750bd121198b1b26492192299

commit 401afd51bb7742a750bd121198b1b26492192299
Author: Yves Gerey <yvesg@webrtc.org>
Date: Wed Sep 19 09:05:20 2018

Auto roller: improved tracking of android dependencies.

Until now, only revision changes were automatically rolled.
This CL detect new and removed dependencies in third_party/android_deps.

Change-Id: I4f83b7308be577115cc3ed57edd9881496428173
Bug:  chromium:855108 
Reviewed-on: https://webrtc-review.googlesource.com/100021
Commit-Queue: Patrik Höglund <phoglund@webrtc.org>
Reviewed-by: Patrik Höglund <phoglund@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#24773}
[modify] https://crrev.com/401afd51bb7742a750bd121198b1b26492192299/tools_webrtc/autoroller/roll_deps.py
[modify] https://crrev.com/401afd51bb7742a750bd121198b1b26492192299/tools_webrtc/autoroller/unittests/roll_deps_test.py
[modify] https://crrev.com/401afd51bb7742a750bd121198b1b26492192299/tools_webrtc/autoroller/unittests/testdata/roll_deps/DEPS
[modify] https://crrev.com/401afd51bb7742a750bd121198b1b26492192299/tools_webrtc/autoroller/unittests/testdata/roll_deps/DEPS.chromium.new
[add] https://crrev.com/401afd51bb7742a750bd121198b1b26492192299/tools_webrtc/autoroller/unittests/testdata/roll_deps/DEPS.chromium.with_android_deps
[add] https://crrev.com/401afd51bb7742a750bd121198b1b26492192299/tools_webrtc/autoroller/unittests/testdata/roll_deps/DEPS.with_android_deps

The following revision refers to this bug:
  https://webrtc.googlesource.com/src.git/+/1548e9a2959360b327f57d2b2851995b9d1a397f

commit 1548e9a2959360b327f57d2b2851995b9d1a397f
Author: Yves Gerey <yvesg@webrtc.org>
Date: Wed Sep 19 12:25:56 2018

Auto roller: don't complain about expected dependencies.

It's ok for some WebRTC dependencies not to be Chromium dependencies.
Explicitly list them instead of relying of CIDP discrimination.

Bug:  chromium:855108 
Change-Id: I2dafce488b28409cbce7e0c3167d92f48859084f
Reviewed-on: https://webrtc-review.googlesource.com/101000
Commit-Queue: Yves Gerey <yvesg@webrtc.org>
Reviewed-by: Patrik Höglund <phoglund@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#24774}
[modify] https://crrev.com/1548e9a2959360b327f57d2b2851995b9d1a397f/tools_webrtc/autoroller/roll_deps.py
[modify] https://crrev.com/1548e9a2959360b327f57d2b2851995b9d1a397f/tools_webrtc/autoroller/unittests/roll_deps_test.py

The following revision refers to this bug:
  https://webrtc.googlesource.com/src.git/+/26a6b5899f4901ef14f09982e01d1b0014639455

commit 26a6b5899f4901ef14f09982e01d1b0014639455
Author: Yves Gerey <yvesg@webrtc.org>
Date: Wed Sep 19 14:33:40 2018

Auto roller: fix list of WEBRTC_ONLY_DEPS.

Very unfortunate typo.

Bug:  chromium:855108 
Change-Id: I236e8537537e7dad805d58c92f804ef021981574
Reviewed-on: https://webrtc-review.googlesource.com/101021
Reviewed-by: Oleh Prypin <oprypin@webrtc.org>
Commit-Queue: Yves Gerey <yvesg@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#24775}
[modify] https://crrev.com/26a6b5899f4901ef14f09982e01d1b0014639455/tools_webrtc/autoroller/roll_deps.py

The following revision refers to this bug:
  https://webrtc.googlesource.com/src.git/+/31eb01faa580cee28e3fec461b325dd50e18678a

commit 31eb01faa580cee28e3fec461b325dd50e18678a
Author: Yves Gerey <yvesg@webrtc.org>
Date: Wed Sep 19 15:51:20 2018

Auto roller: Fix GenerateCommitMessage signature.

TODO: Add integration test.

Bug:  chromium:855108 
Change-Id: Ic892cd09e6712e9b7304e8b10b5fdc147b38a6bd
Reviewed-on: https://webrtc-review.googlesource.com/101040
Commit-Queue: Yves Gerey <yvesg@webrtc.org>
Reviewed-by: Oleh Prypin <oprypin@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#24776}
[modify] https://crrev.com/31eb01faa580cee28e3fec461b325dd50e18678a/tools_webrtc/autoroller/roll_deps.py
[modify] https://crrev.com/31eb01faa580cee28e3fec461b325dd50e18678a/tools_webrtc/autoroller/unittests/roll_deps_test.py

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/a9abce3b979716c611816a6a07ec43e9c7beb145

commit a9abce3b979716c611816a6a07ec43e9c7beb145
Author: chromium-autoroll <chromium-autoroll@skia-public.iam.gserviceaccount.com>
Date: Thu Sep 20 10:20:25 2018

Roll src/third_party/webrtc d574123c5035..ee002e61857b (32 commits)

https://webrtc.googlesource.com/src.git/+log/d574123c5035..ee002e61857b


git log d574123c5035..ee002e61857b --date=short --no-merges --format='%ad %ae %s'
2018-09-19 chfremer@webrtc.org Fix WebRTC fuzzers tests in Chromium missing field trial implementation
2018-09-19 yvesg@webrtc.org Roll chromium_revision e4b02117a9..cc7b9c6822 (592264:592452)
2018-09-19 yvesg@webrtc.org Auto roller: Fix GenerateCommitMessage signature.
2018-09-19 yvesg@webrtc.org Auto roller: fix list of WEBRTC_ONLY_DEPS.
2018-09-19 yvesg@webrtc.org Auto roller: don't complain about expected dependencies.
2018-09-19 yvesg@webrtc.org Auto roller: improved tracking of android dependencies.
2018-09-19 buildbot@webrtc.org Roll chromium_revision f305d23e18..e4b02117a9 (592151:592264)
2018-09-18 mbonadei@webrtc.org Fix WebRTC fuzzers tests in Chromium.
2018-09-18 buildbot@webrtc.org Roll chromium_revision f37f3783a3..f305d23e18 (591975:592151)
2018-09-18 ssilkin@webrtc.org Restrict use of frame rate controller.
2018-09-18 nisse@webrtc.org Simplify includes in p2ptransportchannel.cc
2018-09-18 sprang@webrtc.org Prevent resolution limited max bitrate from going below min
2018-09-18 mbonadei@webrtc.org Reland "Add RTC_EXPORT macro to export WebRTC symbols."
2018-09-18 mbonadei@webrtc.org Do not compile frame_analyzer_host during Chromium builds.
2018-09-18 mbonadei@webrtc.org Add documentation about field_trial/metrics custom impl.
2018-09-18 jtteh@webrtc.org Revert "Add RTC_EXPORT macro to export WebRTC symbols."
2018-09-18 mbonadei@webrtc.org Reland "Compile frame analyzer for the host machine on perf tests."
2018-09-18 buildbot@webrtc.org Roll chromium_revision 0eb6b522cc..f37f3783a3 (591627:591975)
2018-09-18 danilchap@webrtc.org Refactor RtpPacketizerH264 tests
2018-09-18 danilchap@webrtc.org Cleanup modules_common_types
2018-09-17 mbonadei@webrtc.org Revert "Compile frame analyzer for the host machine on perf tests."
2018-09-17 mbonadei@webrtc.org Add RTC_EXPORT macro to export WebRTC symbols.
2018-09-17 mbonadei@webrtc.org Compile frame analyzer for the host machine on perf tests.
2018-09-17 buildbot@webrtc.org Roll chromium_revision 1d75f6fc68..0eb6b522cc (591373:591627)
2018-09-17 mbonadei@webrtc.org Rename RTC_EXPORT to RTC_OBJC_EXPORT.
2018-09-17 mbonadei@webrtc.org Introduce GN arg rtc_exclude_runtime_enabled_features_default.
2018-09-17 danilchap@webrtc.org Register video rtp header extensions in rtp_rtcp by uri
2018-09-17 sprang@webrtc.org Add ability to throttle VideoBitrateAllocation updates.
2018-09-17 mbonadei@webrtc.org Remove obsolete comments.
2018-09-17 saza@webrtc.org Reland "Remove APM internal usage of EchoCancellation"
2018-09-15 mbonadei@webrtc.org Merge field_trial and field_trial_default in a unique target.
2018-09-15 mbonadei@webrtc.org Merge system_wrappers:metrics and system_wrappers:metrics_default.


Created with:
  gclient setdep -r src/third_party/webrtc@ee002e61857b

The AutoRoll server is located here: https://autoroll.skia.org/r/webrtc-chromium-autoroll

Documentation for the AutoRoller is here:
https://skia.googlesource.com/buildbot/+/master/autoroll/README.md

If the roll is causing failures, please contact the current sheriff, who should
be CC'd on the roll, and stop the roller if necessary.

CQ_INCLUDE_TRYBOTS=luci.chromium.try:linux_chromium_archive_rel_ng;luci.chromium.try:mac_chromium_archive_rel_ng

BUG=chromium:None,chromium:855108,chromium:855108,chromium:855108,chromium:855108,chromium:None,chromium:None,chromium:884164,chromium:None,chromium:None,chromium:None,chromium:None,chromium:None
TBR=webrtc-chromium-sheriffs-robots@google.com

Change-Id: If83dd14ed68cd0ebd408e229d672b44eedb9fb98
Reviewed-on: https://chromium-review.googlesource.com/1235418
Reviewed-by: chromium-autoroll <chromium-autoroll@skia-public.iam.gserviceaccount.com>
Commit-Queue: chromium-autoroll <chromium-autoroll@skia-public.iam.gserviceaccount.com>
Cr-Commit-Position: refs/heads/master@{#592746}
[modify] https://crrev.com/a9abce3b979716c611816a6a07ec43e9c7beb145/DEPS

The following revision refers to this bug:
  https://webrtc.googlesource.com/src.git/+/4b9f3908ed39ebda4994f5fc07db95cfad8c74b4

commit 4b9f3908ed39ebda4994f5fc07db95cfad8c74b4
Author: Yves Gerey <yvesg@webrtc.org>
Date: Mon Sep 24 15:30:39 2018

Auto roller: [unittest cleanup] Proper patching mechanism.

Use mock.patch instead of setattr, deemed hackish and less robust.
As an additional benefit, mock is explictly activated and precisely scoped.

Bug:  chromium:855108 
Change-Id: I3644bb6773a4b95e50aa5b671292e108af1fd2e9
Reviewed-on: https://webrtc-review.googlesource.com/101660
Reviewed-by: Oleh Prypin <oprypin@webrtc.org>
Reviewed-by: Patrik Höglund <phoglund@webrtc.org>
Commit-Queue: Yves Gerey <yvesg@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#24804}
[modify] https://crrev.com/4b9f3908ed39ebda4994f5fc07db95cfad8c74b4/tools_webrtc/autoroller/unittests/roll_deps_test.py

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/76346c9b0afb2febbf2bf3c8d2785ce342e8ad0e

commit 76346c9b0afb2febbf2bf3c8d2785ce342e8ad0e
Author: chromium-autoroll <chromium-autoroll@skia-public.iam.gserviceaccount.com>
Date: Mon Sep 24 23:40:00 2018

Roll src/third_party/webrtc 645512ba5966..a36631cdb571 (26 commits)

https://webrtc.googlesource.com/src.git/+log/645512ba5966..a36631cdb571


git log 645512ba5966..a36631cdb571 --date=short --no-merges --format='%ad %ae %s'
2018-09-24 danilchap@webrtc.org Expect late delayed task is deleted.
2018-09-24 buildbot@webrtc.org Roll chromium_revision c92470b71e..695c43e432 (593506:593611)
2018-09-24 niklas.enbom@webrtc.org Revert "Bug in histogram metric reporting."
2018-09-24 srte@webrtc.org Don't reset initial constraints in congestion controller.
2018-09-24 orphis@webrtc.org Add dummy implementation for SetCodecPReferences.
2018-09-24 terelius@webrtc.org Add helper class to process RtcEventLog events in order.
2018-09-24 yvesg@webrtc.org ResultsContainer: Initialize fields.
2018-09-24 yvesg@webrtc.org Auto roller: [unittest cleanup] Proper patching mechanism.
2018-09-24 kwiberg@webrtc.org TestStereo: Don't rely on the ACM to create encoders
2018-09-24 gustaf@webrtc.org AEC3: Delay estimator adapts even when estimated echo saturates
2018-09-24 kron@webrtc.org Refactor to remove direct memory dependency on kMaxId
2018-09-24 aleloi@webrtc.org Bug in histogram metric reporting.
2018-09-24 buildbot@webrtc.org Roll chromium_revision a7544fa319..c92470b71e (592771:593506)
2018-09-24 sprang@webrtc.org Revert "Reland "Enable simulcast screenshare by default""
2018-09-24 phensman@webrtc.org Include stringutils to allow build on chromium
2018-09-24 devicentepena@webrtc.org AEC3: Bounding the nearend spectrum used as input for the suppressor gain computation
2018-09-24 artit@webrtc.org Disabled TestPacketBuffer.SeqNumWrapOneFrame test due to clang update
2018-09-24 nisse@webrtc.org Delete unused Url class.
2018-09-24 nisse@webrtc.org Delete unused HttpData methods.
2018-09-24 kwiberg@webrtc.org AudioCodingModuleTest.TestStereo: Delete write-only variables
2018-09-24 mflodman@webrtc.org Don't throttle key-frame requests per layer.
2018-09-24 jonasolsson@webrtc.org Make fewer copies when using StringBuilder.
2018-09-24 asapersson@webrtc.org Android: Add maxFramerate to RtpParameters.
2018-09-24 artit@webrtc.org Revert "Added support of getting coverage on mac"
2018-09-24 phensman@webrtc.org Add WriteVideoToFile to video_file_reader.
2018-09-24 henrika@webrtc.org Ensures that ADM unittest uses default audio devices for all platforms.


Created with:
  gclient setdep -r src/third_party/webrtc@a36631cdb571

The AutoRoll server is located here: https://autoroll.skia.org/r/webrtc-chromium-autoroll

Documentation for the AutoRoller is here:
https://skia.googlesource.com/buildbot/+/master/autoroll/README.md

If the roll is causing failures, please contact the current sheriff, who should
be CC'd on the roll, and stop the roller if necessary.

CQ_INCLUDE_TRYBOTS=luci.chromium.try:linux_chromium_archive_rel_ng;luci.chromium.try:mac_chromium_archive_rel_ng

BUG=chromium:None,chromium:None,chromium:755660,chromium:855108,chromium:None,chromium:690537,chromium:887464,chromium:844647
TBR=webrtc-chromium-sheriffs-robots@google.com

Change-Id: I906ee8f2b3f7798a61aef95bce71fbcbe0d81221
Reviewed-on: https://chromium-review.googlesource.com/1241517
Reviewed-by: chromium-autoroll <chromium-autoroll@skia-public.iam.gserviceaccount.com>
Commit-Queue: chromium-autoroll <chromium-autoroll@skia-public.iam.gserviceaccount.com>
Cr-Commit-Position: refs/heads/master@{#593747}
[modify] https://crrev.com/76346c9b0afb2febbf2bf3c8d2785ce342e8ad0e/DEPS

The include mechanism wasn't brought to completion.
The roll_deps.py workaround dealing with ANDROID_DEPS covers most of our needs.