more robust circleci testing by MartinNowak · Pull Request #6022 · dlang/dmd

MartinNowak · 2016-08-06T17:01:06Z

No description provided.

wilzbach · 2016-08-06T17:06:03Z

circle.yml

 test:
  override:
-    - case $CIRCLE_NODE_INDEX in 0) MODEL=32 ./travis.sh ;; 1) MODEL=64 ./travis.sh ;; esac:
+    - case $CIRCLE_NODE_INDEX in 0) MODEL=32 ./circleci.sh ;; 1) MODEL=64 ./circleci.sh ;; esac:


@MartinNowak this is a hard-coded hack to support parallelism.
However of course, if only one container is enabled for a repo only the 32-bit will be run.

It's a horrible YAML abomination b/c apparently they aren't capable to allow hashes as argument (and we don't want to use multiline bash scripts as hash key, even in YAML).

- move all the logic into a separate shell script (CircleCI's YAML is really unfriendly for slightly more complicated stuff) - use installer script to fetch bootstrap compiler (same retry logic as on Travis-CI) - cache installed bootstrap compiler

- more consisten with other dmd make options

This reverts commit cffa4c3.

MartinNowak · 2016-08-06T18:36:18Z

Mainly did this b/c I wanted to restructure the Travis-CI tests to start less workers, but failed due to all the extra cases for CircleCI.
Better to have that in a separate file and deal w/ only a single provider at a time.

codecov-io · 2016-08-06T18:51:06Z

Current coverage is 87.42% (diff: 100%)

Merging #6022 into master will increase coverage by 0.39%

@@             master      #6022   diff @@
==========================================
  Files            90         90          
  Lines         55686      55686          
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
+ Hits          48459      48681   +222   
+ Misses         7227       7005   -222   
  Partials          0          0

Powered by Codecov. Last update ec75122...48e3640

wilzbach · 2016-08-06T18:54:53Z

circleci.sh

+case $CIRCLE_NODE_INDEX in
+    0) MODEL=64 ;;
+    1) MODEL=32 ;;
+esac


Could you please enable two containers, s.t. 32-bit gets tested too?

Should be here https://circleci.com/gh/dlang/dmd/edit#parallel-builds

I enabled a second one.

We're not testing the backend for coverage, shouldn't be much difference, but well.

I asked @wilzbach to stick with one, because the Terms Of Service only offer 4 instances for open source projects. It was unclear what CircleCI's definition of open source was, so while we await clarification I'd like to stay with one. (If their definition is "public repository", then we're golden.)

I asked @wilzbach to stick with one, because the Terms Of Service only offer 4 instances for open source projects.

Did you miss this comment?

I just found the following in the settings at the org page

Additionally, projects that are public on GitHub will build with 3 extra containers -- our gift to free and open source software.

and then this in their FAQ

What if I am building open-source?

We offer a total of four free linux containers ($2400 annual value) for open-source projects. Simply keeping your project public will enable this for you!

#6006 (comment)

I wouldn't think there is any credible risk associated with that, but if it makes you uncomfortable, feel free to just toggle it back in the admin UI (see @wilzbach's link).

If we can get away with only one configuration, this would be preferred either way, though, since parallel PR builds eat into the same allowance.

@wilzbach yes, I had missed that. Glad this is cleared up. Thanks!

@klickverbot Looks like we're good on that, then. But there's that other issue you mentioned - maybe throttling it by using only 1 instance will be more effective for us than running dry halfway through the month.

@WalterBright: It's not so much a question of "running dry" than just the latency of parallel builds on many updated PRs.

than just the latency of parallel builds on many updated PRs.

@klickverbot Let's just see whether it makes a difference and/or clogs our CircleCi queue. It's just an experiment, so we can always switch back to 1 container ;-)

wilzbach · 2016-08-06T19:05:57Z

Mainly did this b/c I wanted to restructure the Travis-CI tests to start less workers, but failed due to all the extra cases for CircleCI.
Better to have that in a separate file and deal w/ only a single provider at a time.

Thanks a lot for this massive cleanup 👍
It looks a lot more professional than my hack.

MartinNowak · 2016-08-06T19:09:09Z

circleci.sh

+        local base_branch=$(curl -fsSL https://api.github.com/repos/dlang/dmd/pulls/$CIRCLE_PR_NUMBER | jq -r '.base.ref')
+    else
+        local base_branch=$CIRCLE_BRANCH
+    fi


This is the snippet to get the target branch. Hopefully we won't hit any rate limit of github, b/c of other people on the test machine.

I hope we don't , otherwise we can try 1) direct authenticated access or 2) rolling our own, small API cache

Or extend CircleCI, but I don't it'll be a problem.

Let's see How to get the pull request upstream branch? - Build Environment - CircleCI Community Discussion.

Btw didn't you want to merge into the current base_branch?
Something like simple like this should do it?

if [ -n ${CIRCLE_PR_NUMBER:-} ]; then git remote add upstream git@github.com:dlang/dmd.git git fetch upstream git merge "upstream/${base_branch}" else git merge ${base_branch} fi

Yes merging would be good.

MartinNowak · 2016-08-06T19:34:21Z

Auto-merge toggled on

JackStouffer · 2016-08-08T18:17:31Z

@MartinNowak This broke building dmd on OS X

$ make -f posix.mak AUTO_BOOTSTRAP=1
/Library/Developer/CommandLineTools/usr/bin/make -C src -f posix.mak
no cpu specified, assuming X86
mkdir -p /tmp/.host_dmd-2.068.2
curl -fsSL http://downloads.dlang.org/releases/2.x/2.068.2 /dmd.2.068.2 .osx.tar.xz | tar -C /tmp/.host_dmd-2.068.2  -Jxf - || rm -rf /tmp/.host_dmd-2.068.2
curl: (3) <url> malformed
curl: (6) Could not resolve host: .osx.tar.xz
tar: Unrecognized archive format
tar: Error exit delayed from previous errors.
CC=c++ /tmp/.host_dmd-2.068.2 /dmd2/osx/bin/dmd -conf=/tmp/ /dmd2/osx/bin/dmd.conf idgen.d
/bin/sh: /tmp/.host_dmd-2.068.2: No such file or directory
make[1]: *** [idgen] Error 127
make: *** [all] Error 2

JackStouffer · 2016-08-08T18:17:44Z

src/posix.mak

 else
  # Auto-bootstrapping, will download dmd automatically
-  HOST_DMD_VER=2.068.2
+  HOST_DMD_VER=2.068.2 # keep in sync with other occurences of that variable, e.g. in circleci.sh


@MartinNowak This line is the culprit

That's bash craziness 😢
@JackStouffer do quotes work for you?

Deleting the space after the 2 and moving the comment to its own line works

Deleting the space after the 2 and moving the comment to its own line works

How about quickly submitting this as PR? ;-)

wilzbach reviewed Aug 6, 2016
View reviewed changes

MartinNowak force-pushed the cleanup_circleci branch 7 times, most recently from 86fc186 to 3809f7c Compare August 6, 2016 18:02

MartinNowak added 3 commits August 6, 2016 20:28

more robust circleci testing

f961c40

- move all the logic into a separate shell script (CircleCI's YAML is really unfriendly for slightly more complicated stuff) - use installer script to fetch bootstrap compiler (same retry logic as on Travis-CI) - cache installed bootstrap compiler

rename make option to ENABLE_COVERAGE

95d4bc8

- more consisten with other dmd make options

Revert "Fix CI checkout branch for non Travis CIs"

73f5a90

This reverts commit cffa4c3.

MartinNowak force-pushed the cleanup_circleci branch from 3809f7c to bdddede Compare August 6, 2016 18:34

MartinNowak changed the title ~~WIP - more robust circleci testing~~ more robust circleci testing Aug 6, 2016

MartinNowak force-pushed the cleanup_circleci branch from bdddede to c842a1c Compare August 6, 2016 18:38

remove circleci stuff from travis.sh

48e3640

wilzbach reviewed Aug 6, 2016
View reviewed changes

MartinNowak force-pushed the cleanup_circleci branch from c842a1c to 48e3640 Compare August 6, 2016 19:01

MartinNowak mentioned this pull request Aug 6, 2016

reduce Travis-CI workload #6023

Merged

MartinNowak reviewed Aug 6, 2016
View reviewed changes

MartinNowak merged commit cf0a174 into dlang:master Aug 6, 2016

MartinNowak deleted the cleanup_circleci branch August 6, 2016 21:54

This was referenced Aug 7, 2016

Make Coverage checking more stable dlang/druntime#1625

Merged

Switch to CircleCi dlang/phobos#4719

Merged

JackStouffer reviewed Aug 8, 2016
View reviewed changes

This was referenced Aug 12, 2016

Fix whitespace issue in posix.mak #6051

Closed

At least GNU make 4.1 on Linux Mint introduces a space in a variable … #6052

Merged

Uh oh!

Conversation

MartinNowak commented Aug 6, 2016

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

MartinNowak commented Aug 6, 2016

Uh oh!

codecov-io commented Aug 6, 2016 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Current coverage is 87.42% (diff: 100%)

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

wilzbach commented Aug 6, 2016

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

MartinNowak commented Aug 6, 2016

Uh oh!

JackStouffer commented Aug 8, 2016

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

codecov-io commented Aug 6, 2016 •

edited

Loading