Recent problems building chrome_public_test_apk (Android)

[Kevin]

I have run into the 64K method limit while building chrome_public_test_apk.  I added enable_multidex=true to the instrumentation_test_apk("chrome_public_test_apk") rule in chorme/Android/BUILD.gn, which solved the problem for release builds.  However, in a debug build, I am getting a related error:

FAILED: gen/chrome/android/chrome_public_test_apk__apk/classes.dex.zip 

python ../../build/android/gyp/dex.py --depfile gen/chrome/android/chrome_public_test_apk__apk__final_dex.d --android-sdk-tools ../../third_party/android_tools/sdk/build-tools/24.0.2 --dex-path gen/chrome/android/chrome_public_test_apk__apk/classes.dex.zip --incremental --multi-dex --main-dex-list-path gen/chrome/android/chrome_public_test_apk__apk/classes.dex.zip.main_dex_list --inputs=@FileArg\(gen/chrome/android/chrome_public_test_apk__apk.build_config:dist_jar:dependency_jars\) gen/chrome/android/chrome_public_test_apk__apk/chrome_public_test_apk__apk.jar

  ... Snipped ...

UNEXPECTED TOP-LEVEL EXCEPTION:

com.android.dex.DexException: Too many classes in --main-dex-list, main dex capacity exceeded

Some quick googlin suggested that I change the minSdk to 1 or use proguard, but the general rule in Chromium seems to be that proguard is not used for debug builds.  In most of the manifests the minimum SDK is set to 16, so I'm not sure that is the right answer either.

Any advice will be appreciated

  - Kevin

Qualcomm Innovation Center, Inc.

Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.

[John Budorick]

What gn arguments are you building with? Are you building with current trunk, or with a build that incorporates other changes?

--
--
Chromium Developers mailing list: chromi...@chromium.org
View archives, change email options, or unsubscribe:
http://groups.google.com/a/chromium.org/group/chromium-dev
---
You received this message because you are subscribed to the Google Groups "Chromium-dev" group.
To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/chromium-dev/52abc40d-c439-463c-aabc-16552649f463%40chromium.org.

[Kevin]

args.gn:

is_component_build = false

is_debug = true

target_os="android"

cc_wrapper="ccache"

This is a relatively recent dev build (59.0.3056.0) with fairly extensive local changes.

  - Kevin

Qualcomm Innovation Center, Inc.

Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.

[Holger]

Is there any reason we are not using the same multidex config as ChromePublic.apk for the tests?

https://cs.chromium.org/chromium/src/chrome/android/chrome_public_apk_tmpl.gni?type=cs&q=enable_multidex+chrome_public+package:%5Echromium$&l=52

ChromePublic already uses MultiDex on Debug builds - the tests should probably do the same.

(I haven't run a Dex-check recently, but it's possible that your local changes bumped the test apk above 65k, I don't think Chrome was very far from the limit last time I checked)

[Holger]

Hm, did you make any gn changes?

I just did a dexcount on ChromePublicTest.apk from today's master:

Overall method count: 35662

That's very far from the limit.

--> Any dependency you added that might have pulled in something you didn't want to pull in?

[Kevin]

My current understanding of my problem is that multidex is enabled (with my change), but because proguard is not enabled for a debug build, the main dex itself ends up with over 65536 entries.  (Release build is OK).  I'm not sure why this wouldn't be a problem for ChromePublic.apk as well.  But I'm pretty fuzzy on all of this.

  - Kevin

[Kevin]

OK, I guess I'll have to run dexcount on the baseline and see where the delta is coming from.  I'm not familiar with dexcount, and the references I've found online all seem to describe it as a gradle plugin.  How do I use dexcount in the context of a gn / ninja build?

[Kevin]

Never mind, using https://github.com/mihaip/dex-method-counts.  The investigation continues.

[Kevin]

I'm getting 61591 method IDs in ChromePublicTest.apk, in a baseline m58 build.

args.gn:

target_os = "android"

is_debug = false

symbol_level = 1

Top-level output of dex-method-counts:

Processing ChromePublicTest.apk

Read in 61591 method IDs.

<root>: 61591

    <default>: 1

    android: 14566

    com: 6839

    dalvik: 6

    java: 1262

    javax: 48

    jp: 50

    junit: 131

    org: 38682

Overall method count: 61591

Details of org:

    org: 38682

        chromium: 38580

            base: 948

            chrome: 30602 

                browser: 29924

                test: 677   

            components: 1208

            content: 1973   

            device: 747

            media: 510

            midi: 78

            mojo: 387

            net: 338

            payments: 228

            policy: 58

            printing: 79

            service_manager: 22

            services: 8

            shape_detection: 157

            test: 16

            third_party: 293

            ui: 583

            url: 6

            webapk: 101

            webshare: 36

[Kevin]

Any idea why I'm getting such different results?  (attached)

On Wednesday, April 5, 2017 at 11:32:31 AM UTC-7, Holger wrote:

[Holger]

Seems you switched over to a Release build when doing the dexcount this time.
Not sure if there should be large differences between release and debug, but my dex count above was based on a debug build.

[John Budorick]

We've started seeing this in some build configurations in trunk and are now tracking it in crbug/710811 (internal due to some of the configurations involved). I'm looking into our main dex list generation in debug builds.

(Somewhat of an aside -- we've been using //build/android/method_count.py to track method counts for a while, though we do so indirectly nowadays via resource_sizes.py. It doesn't break methods down by package, though.)

--

--
Chromium Developers mailing list: chromi...@chromium.org
View archives, change email options, or unsubscribe:
http://groups.google.com/a/chromium.org/group/chromium-dev
---
You received this message because you are subscribed to the Google Groups "Chromium-dev" group.

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/chromium-dev/47136888-a90e-4a36-ab0c-e182723ff2c3%40chromium.org.

[Kevin]

Thanks for following up.  Please let me know if and when this is tracked by a public bug.

[John Budorick]

I've moved the public part of this bug to https://bugs.chromium.org/p/chromium/issues/detail?id=712852.

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/chromium-dev/9ff246e4-c9cd-4004-a627-6b48c837982c%40chromium.org.