The Android task has many issues:
Fix all issues by removing the task. Note that the CI env file is kept, so anyone can still run the Android CI.
Also, use the compute credits to promote tsan, a more useful task.
The following sections might be updated with supplementary metadata relevant to reviewers and maintainers.
See the guideline for information on the review process.
If your review is incorrectly listed, please react with 👎 to this comment and the bot will ignore it on the next update.
ACK fa22538e481fa2c4f0b5d6f91166335e60b67fe9 - nuke it
I have only ever seen this task pass (even if all others fail) or fail intermittently.
It should be noted the primary role of the task is to check that a change doesn’t prevent us from building for android, a platform with 3 billion users and a platform that we care about building for. A manual check will be needed for build system changes to ensure that this compatibility is kept. @dergoegge
can you show an example to back up this claim: “I have only ever seen this task pass (even if all others fail)…” - The android CI won’t pass if we can’t build core, if all other tests fail we can’t build core (macOS 10.15 ci task for example). So this sounds impossible.
NACK
I’m not sure if the PR description counts as “many” issues.
If this task was failing intermittently all the time, why are there no issues posted about these intermittent failures occurring all the time, everywhere? We watch this task often in the QML repo and have never had it fail due to a network timeout.
It’s great to speed up a task, and I wish all tasks could run as quickly as possible; that would be really great. But not at the expense of a check on building for a platform we want to support, with a massive user base, more than any other operating system in existence.
The potential to allow Android devices to participate in the network as full nodes is powerful. It’s beautiful to enable a person in a part of the world where one only has an old Android phone to benefit from having a self-sovereign view of the network and their coins, to participate in the enforcement of the network rules, and to broadcast transactions in an unstoppable P2P manner directly from their device.
And, the changes to support Android are “surprisingly small and sensible”, even in the QML repo as we’ve added the ability for it to run in the background as a process.
| Android Tablet | Android Phone |
|---|---|
This feels, in a small way, motivated by the want to remove the GUI by some. There seems to be an idea among a handful of us that all of our issues will be fixed if we rm -rf ./src/qt. This isn’t so, and maintenance of core will still involve headaches and work. All it does is remove a way for many to run a node, reducing the population and harming the demographic of those running nodes. I care about non-technical individuals being able to run core. I would hope others share the same sentiment.
why are there no issues posted about these intermittent failures
Because no one reported them. But they do exist, if you grep for “Warning: An error occurred while preparing SDK package Android Emulator: Connection reset.” For example, https://cirrus-ci.com/task/5766445213155328?logs=build#L2700
It’s great to speed up a task, and I wish all tasks could run as quickly as possible;
It only checks that it can build, and doesn’t run any tests, so I hope the other tasks remain to run the tests.
This feels, in a small way, motivated by the want to remove the GUI by some.
I think this is a far stretch. I’ve never encouraged to remove the GUI and would strongly object the removal. What this does is remove the Android task from the CI config, nothing more. The CI env is explicitly kept, to be able to continue running the task outside. And in fact it is already running on a nightly basis on my nightly CI repo: https://cirrus-ci.com/task/5872120266227712 , along with other important to check tasks that have no place in this repo.
Needless to say this is also no attempt to remove Android compatibility. I agree with you that this is an important platform and demographic to support.
This is merely removing a task that is failing more often due to unrelated issues (network) than to indicate a useful result (true red CI). If you disagree it would be helpful to demonstrate or link to one instance where this task has caught or would have caught an issue.
I don’t feel too strongly, and I am happy to close this, but I think we should be having a discussion based on real data points and not feelings.
Another failure on a doc change: https://cirrus-ci.com/task/5213373952950272?logs=ci#L1404
0 CXX qt/libbitcoinqt_a-qrc_bitcoin.o
1 CXX qt/libbitcoinqt_a-qrc_bitcoin_locale.o
2 AR libbitcoin_node.a
3 AR libbitcoin_util.a
4 AR libbitcoin_wallet.a
5 AR libbitcoin_zmq.a
6 AR libbitcoin_cli.a
7 AR libbitcoin_common.a
8 AR libbitcoin_consensus.a
9 CXXLD crypto/libbitcoin_crypto_base.la
10 CXXLD crypto/libbitcoin_crypto_arm_shani.la
11 CXXLD libunivalue.la
12 CXXLD leveldb/libleveldb.la
13 CXXLD crc32c/libcrc32c.la
14 CXXLD leveldb/libmemenv.la
15 AR libtest_util.a
16 AR qt/libbitcoinqt.a
17 CXXLD qt/bitcoin-qt
18 CXXLD qt/test/test_bitcoin-qt
19make[2]: Leaving directory '/tmp/cirrus-ci-build/src'
20make[1]: Leaving directory '/tmp/cirrus-ci-build/src'
21Making all in doc/man
22make[1]: Entering directory '/tmp/cirrus-ci-build/doc/man'
23make[1]: Nothing to be done for 'all'.
24make[1]: Leaving directory '/tmp/cirrus-ci-build/doc/man'
25make[1]: Entering directory '/tmp/cirrus-ci-build'
26make[1]: Nothing to be done for 'all-am'.
27make[1]: Leaving directory '/tmp/cirrus-ci-build'
28+ cd src/qt
29+ ANDROID_HOME=/tmp/cirrus-ci-build/depends/SDKs/android
30+ ANDROID_NDK_HOME=/tmp/cirrus-ci-build/depends/SDKs/android/ndk/23.2.8568313
31+ make apk
32make -C .. bitcoin_qt_apk
33make[1]: Entering directory '/tmp/cirrus-ci-build/src'
34tar: --exclude=*/*: Cannot open: No such file or directory
35tar: Error is not recoverable: exiting now
36tar: --exclude=*/*: Cannot open: No such file or directory
37tar: Error is not recoverable: exiting now
38mkdir -p qt/android/libs/arm64-v8a
39cp /tmp/cirrus-ci-build/depends/SDKs/android/ndk/23.2.8568313/toolchains/llvm/prebuilt/linux-x86_64/bin//../sysroot/usr/lib/aarch64-linux-android/libc++_shared.so qt/android/libs/arm64-v8a
40tar xf -C qt/android/src/ src/android/jar/src --strip-components=5
41tar: -C: Cannot open: No such file or directory
42tar: Error is not recoverable: exiting now
43make[1]: *** [Makefile:22313: bitcoin_qt_apk] Error 2
44make[1]: Leaving directory '/tmp/cirrus-ci-build/src'
45make: *** [Makefile:11: apk] Error 2
46Exit status: 2
can you show an example to back up this claim: “I have only ever seen this task pass (even if all others fail)…” @jarolrod I can’t find the PR this happened on and I’ve only seen that one time but I swear I’m not making this up. The macOS task has other build flags, so it must have been some weird combination of that and tests/lint failing but some builds succeeding.