838259 - Update DXR master to a newer version

Status: RESOLVED FIXED

(Release Engineering :: General, defect)

Description

[Joshua Cranmer [:jcranmer]]

This requires changing several scripts and uploading tarballs, so I want to have a test version run first.

Comment 1

[Joshua Cranmer [:jcranmer]]

Attachment: New copy of build environment

Comment 2

[Joshua Cranmer [:jcranmer]]

Attachment: Patch for testing

Modulo the fact that the manifest entry for dxr-build-env.tar.gz has to be changed, and the fact that I'm using a separate repository clone for the test, this should be the patch to the buildbot scripts that would go into production.

Comment 3

[Rail Aliiev [:rail]]

Attachment: dxr.sh fixes

I had to change the script to make it work:

* removed setup.sh from the manifest (it's not used)
* updated dxr-build-env.tar.gz information in the manifest (implying that te attached tarball is suposed to be used)
* chmoded dxr.sh 755 (reverted 644 done by the patch)
* commands passed to mock --shell should be quoted

BTW, can we add "set -x" to the top of the script to make it more verbose?

ATM, the script fails with the following errors:

https://gist.github.com/rail/4719860

Comment 4

[Joshua Cranmer [:jcranmer]]

Attachment: Updated test patch

Okay, I had paths wrong. This new patch adds a set -x and should have correct paths.

Comment 5

[Rail Aliiev [:rail]]

With this change:
-MOCK_PATH=/usr/local/bin:$PWD/clang/bin:$PATH
+MOCK_PATH=/usr/local/bin:$PWD/dxr-build-env/clang/bin:$PATH

the script fails:
https://gist.github.com/rail/4720392

Comment 6

[Joshua Cranmer [:jcranmer]]

So the base versions of gcc/g++ on the builder are too old.

What if you tried adding CC/CXX for the mock_mozilla to build dxr:
mock_mozilla -r mozilla-centos6-x86_64 --cwd=$PWD --shell --unpriv /bin/env PATH=$MOCK_PATH CC=clang CXX=clang++ "make -C dxr"

Comment 7

[Rail Aliiev [:rail]]

Closer :)
https://gist.github.com/rail/4722751

Comment 8

[Joshua Cranmer [:jcranmer]]

Attachment: New copy of build environment

Alas, this one requires changes to dxr-build-env.tar.gz.

Comment 9

[Joshua Cranmer [:jcranmer]]

Attachment: New copy of build environment

Looks like the dxr.config file was busted. Let's try this one.

Comment 10

[Rail Aliiev [:rail]]

Still failing: https://gist.github.com/rail/4724757

Comment 11

[Rail Aliiev [:rail]]

Attachment: udpated patch

I ended up by setting LD_LIBRARY_PATH instead of copying files to system-wide directories. The script fails with the following error now:
https://gist.github.com/rail/4726910

Comment 12

[Rail Aliiev [:rail]]

Attachment: updated patch

Closer: https://gist.github.com/rail/4728815

Comment 13

[Joshua Cranmer [:jcranmer]]

Attachment: Updated test patch

I've moved the dxr.config out of the manifest tarball, which should make deployment easier in theory.

Comment 14

[Rail Aliiev [:rail]]

Attachment: current patch

... and the output: https://gist.github.com/rail/4736020

Almost there! :)

Comment 15

[Rail Aliiev [:rail]]

Attachment: working one

This one worked fine. All green.

Comment 16

[Joshua Cranmer [:jcranmer]]

Attachment: Update dxr building

This should be all the changes that need to go live.

Comment 17

[Rail Aliiev [:rail]]

Comment on attachment 712553 [details] [diff] [review]
Update dxr building

Review of attachment 712553 [details] [diff] [review]:
-----------------------------------------------------------------

Thanks for the patch.

Nothing critical to fix, just some style issues. It'd be better to keep the code more readable so we can easier to change it in the future. 
I'll be glad to test the final patch in staging again.

::: scripts/dxr/dxr.sh
@@ +48,5 @@
> +mock_mozilla -v -r mozilla-centos6-x86_64 --install autoconf213 python zip mozilla-python27-mercurial git ccache glibc-static libstdc++-static perl-Test-Simple perl-Config-General gtk2-devel libnotify-devel alsa-lib-devel libcurl-devel wireless-tools-devel libX11-devel libXt-devel mesa-libGL-devel gnome-vfs2-devel GConf2-devel wget mpfr xorg-x11-font* imake gcc45_0moz3 yasm pulseaudio-libs-devel || exit 2
> +# Old version
> +mock_mozilla -r mozilla-centos6-x86_64 --install pyxdg python-devel || exit 2
> +# New things to add:
> +mock_mozilla -r mozilla-centos6-x86_64 --install python-jinja2 python-pygments sqlite-devel || exit 2

Can you merge the --install commands above into one and wrap the line with backslash so it's more readable? Keeping -v would be great so it'll be easier to debug errors.

@@ +49,5 @@
> +# Old version
> +mock_mozilla -r mozilla-centos6-x86_64 --install pyxdg python-devel || exit 2
> +# New things to add:
> +mock_mozilla -r mozilla-centos6-x86_64 --install python-jinja2 python-pygments sqlite-devel || exit 2
> +# XXX: still need pysqlite?

Not sure what this means. Is the comment still valid?

@@ +61,5 @@
>  rm -rf dxr-build-env
>  mkdir -p dxr-build-env
>  python /tools/tooltool.py -m $(dirname $0)/dxr.manifest -o --url http://runtime-binaries.pvt.build.mozilla.org/tooltool fetch
> +tar xf clang.tar.bz2 -C dxr-build-env || exit 2
> +tar xf dxr-build-env.tar.gz -C dxr-build-env || exit 2

Do you really need all these "|| exit 2". They look ... ugly. You already have "set -e", so it should just exit non zero if it fails.

@@ +79,5 @@
> +# Unpack trilite
> +mv trilite dxr/trilite
> +
> +# Build DXR binary plugins
> +mock_mozilla -r mozilla-centos6-x86_64 --cwd=$PWD --shell --unpriv /bin/env PATH=$MOCK_PATH CC=clang CXX=clang++ "make -C dxr"

Could you wrap this line as well?

@@ +103,5 @@
>  echo "Starting build"
>  set +e
> +# XXX: compile-build hack
> +echo "ac_add_options --enable-stdcxx-compat" > src/.mozconfig
> +mock_mozilla -r mozilla-centos6-x86_64 --cwd=$PWD --shell --unpriv /bin/env PATH=$MOCK_PATH LD_LIBRARY_PATH=$PWD/dxr/trilite "dxr/dxr-build.py -j6 -f dxr.config -s -t $branch" 2>&1 | grep -v 'Unprocessed kind'

And this one.

Comment 18

[Joshua Cranmer [:jcranmer]]

(In reply to Rail Aliiev [:rail] from comment #17)
> Do you really need all these "|| exit 2". They look ... ugly. You already
> have "set -e", so it should just exit non zero if it fails.

I don't pretend to understand releng scripts, but I was thinking that they may have been used to distinguish between different types of failures (Blue versus red, in tbpl color terminology). But I'm not an expert.

Comment 19

[Rail Aliiev [:rail]]

Not a big deal, can live. :)

Comment 20

[Joshua Cranmer [:jcranmer]]

Attachment: Update dxr building

Comment 21

[Rail Aliiev [:rail]]

Comment on attachment 712707 [details] [diff] [review]
Update dxr building

lgtm. if you want I can run a staging test tomorrow.

Comment 22

[Joshua Cranmer [:jcranmer]]

that'd be fine.

Comment 23

[Rail Aliiev [:rail]]

It failed:

+ mock_mozilla -r mozilla-centos6-x86_64 --cwd=/builds/slave/dxr-mozilla-central/dxr-build-env --shell --unpriv /bin/env PATH=/usr/local/bin:/builds/slave/dxr-mozilla-central/dxr-build-env/clang/bin:/usr/local/bin:/usr/lib64/ccache:/usr/local/bin:/bin:/usr/bin:/usr/local/sbin:/usr/sbin:/sbin:/home/cltbld/bin CC=clang CXX=clang++ 'make -C dxr'
INFO: mock_mozilla.py version 1.0.3 starting...
State Changed: init plugins
INFO: selinux disabled
State Changed: start
State Changed: lock buildroot
State Changed: shell
make: *** No targets specified and no makefile found.  Stop.
make: Entering directory `/builds/slave/dxr-mozilla-central/dxr-build-env/dxr'
make: Leaving directory `/builds/slave/dxr-mozilla-central/dxr-build-env/dxr'
State Changed: unlock buildroot
program finished with exit code 2

Sounds like http://hg.mozilla.org/users/Pidgeot18_gmail.com/dxr-test is somehow different from http://hg.mozilla.org/projects/dxr

Comment 24

[Joshua Cranmer [:jcranmer]]

(In reply to Rail Aliiev [:rail] from comment #23)
> Sounds like http://hg.mozilla.org/users/Pidgeot18_gmail.com/dxr-test is
> somehow different from http://hg.mozilla.org/projects/dxr

Oh, I haven't pushed the changes on dxr master yet.

Comment 25

[Rail Aliiev [:rail]]

Comment on attachment 712707 [details] [diff] [review]
Update dxr building

http://hg.mozilla.org/build/tools/rev/2afb2bd56420

Comment 26

[Joshua Cranmer [:jcranmer]]

Attachment: Fix -s -t issue

Comment 27

[Rail Aliiev [:rail]]

Comment on attachment 713244 [details] [diff] [review]
Fix -s -t issue

http://hg.mozilla.org/build/tools/rev/06bdb4f450df

Comment 28

[Rail Aliiev [:rail]]

The scripts is being killed by buildbot after 7200 secs without any output. Can you somehow change the script so it prints something useful so we can debug? Something like:

blah blah | tee log.txt
# check $? 
grep -v 'Unprocessed kind' log.txt && (echo "found"; exit 2)

Comment 29

[Joshua Cranmer [:jcranmer]]

This has long since been checked in.