Registering GHC for Coverity SCAN

[Austin Seipp]

Hello all,

On IRC today, Nicolas Trangez brought up the idea of registering GHC
for Coverity's SCAN project. SCAN is essentially a free service run by
Coverity, which runs their Static Analyzer on open source projects
('open source' being defined by OSI) and gives the results back to
developers. Coverity is widely regarded as having the absolute best
tool in the business for C codebases as it stands. The idea is to try
and rattle bugs out of the runtime system if possible.

I think this is a good idea, but it needs some public discussion.
Namely, Coverity requires an official Point of Contact within the
project (maintainer/author/governing body) to register the project in
their name. We are left up to determine who this is and verify it. Bug
reports which are found are not also made public, as they could be
potentially serious security vulnerabilities.*

So, if done, this is a rather piece-by-piece project which requires a
bit of commitment, because it will require the registree to not only
assess the bugs, but sensibly move them into the issue tracker or fix
it themselves.

It's also a question of whether it's worth it. Personally, I think it
has the chance to uncover some real bugs. But the RTS is highly
unusual C code (by some standards) and it will probably require tuning
and patience to get tangible results. Bugs will also need to be
assessed by someone. False positives happen etc. This is the double
edged sword of static analysis.** Anyone who does it should be very
aware of this.

So, what do you all think? I'd particularly like input from Simon M
here since he's the primary author of the RTS, of course. :) And
anyone who does this will undoubtedly have to be in contact with him
to some degree, for all the aforementioned reasons.

If nobody else would step up for this, I would be interested in
contacting them if it seems worth it and giving it a shot.

* I think the chance of this in our case is very very low of course.
The cases we really care about for a safety perspective are
type-safety violations at the Haskell level, obviously. Nonetheless
this requires an individual who is committed to sifting through
reports and assessing them in a detailed way.

** Coverity at one point found out the hard way that more analysis is
not necessarily better, and people don't respond well to the results
sometimes. Or they're too complicated to explain. The following is a
great read: http://cacm.acm.org/magazines/2010/2/69354-a-few-billion-lines-of-code-later/fulltext

-- 
Regards,
Austin - PGP: 4096R/0x91384671