[Soc-coordination] Weekly Status Report 3 - scan-build the archive

[Léo Cavaillé]

Hello,


Three main topics to report this week (with according time ratio)!

(5% of the time)
1/ Follow up of the cloog problem blocking the llvm-toolchain-snapshot
builds.

We had an email conversation with Matthias Klose (debian cloog
maintainer), and another with Tobias Grosser (polly developer).
Sylvestre has opened a bug (#714830) though I am not convinced that is
the whole gist of the problem.
The isl library can be installed from the archive stand alone, but cloog
uses an inner version of isl which is packaged in the source cloog
package producing then libcloog-isl-dev.
Tobias claims that the new isl version is ABI-compatible.
I will not take too much time to push this for now as it is a bit
driving me away from my "scan-build" subject.

(/me hoping to see the sunshine again in the LLVM Jenkins)

(70%)
2/ debuild.me

I have made some PR [1]
We had some serious discussions with Paul, who helped me to setup my
instance of debuild.me.
After hacking with it locally, I managed to install it at a bigger scale
to have a more meaningful test platform.
Some friends gave me some slack on servers (French holidays, almost
noone on the campus), I have 4 builders/checkers ready for amd64. When I
will go I may setup a few native i386 servers to see multiarch support
in debuild.me (or hack with qemu...).
I am confident we can have a working version, maybe partially packaged,
in 2/3 weeks. Then, we can switch from testing to preproduction and use
the servers Sylvestre told me we would have for the scan-build project,
beginning to thing at the whole archive scale.

(25%)
3/ scan-build and sbuild/schroot

debuild.me mainly uses schroot/sbuild to build or run checkers. Simple
static analysis checkers work with schroot very well. The complexity of
scan-build is that it needs compiling to work, basically we do :
	scan-build dpkg-buildpackage

Paul had the idea to also use sbuild to do that (I was using cowbuilder
before), and to setup the build environment prior to the build we start
a schroot session where I can apt-get install clang for instance.
Desperately, this cannot work in sbuild.
I sent a mail to Roger Leigh (sbuild/schroot maintainer) who kindly
confirmed what I found in the code, and gave me a go for "the idea" of
my patch.
I am working on it : adding some Perl classes, trying not to interfere
with the existing code which is working like a charm.
My first test of the patch today has been a total success with Sbuild
test suite (100%) but a total failure at execution. Let's wait the next
iteration of test ! I will add some test cases as well.


For next week, here is what I plan to do :
* Finish the sbuild patch and submit it to Roger
* Continue to drown Paul with PRs on debuild.me
* Commit some improvement on the scan-build log parsing I have been
doing this week (forgot this tiny part above), add tests
* Test the clanganalyzer plugin I wrote in debuild.me with the fresh
sbuild patch


Cheers,

-- 
Léo


[1] https://github.com/LeoCavaille