----- Taras Glek <[email protected]> wrote:
> Ok. Thanks for the examples. Can you assign the sixgill bug to yourself?

Hi, I added an nsDebug.h patch to bug 541220.  How do I assign the bug to 
myself?

> I think the way we will land this is by adding another build option to 
> Mozilla(similar to --with-static-checking flag).

This might not be necessary.  The process I've been using doesn't require any 
modifications to the mozilla build, but works by wrapping that build in a 
script which does two things:
1. Forks a manager process which collects and collates the plugin's results 
into some output databases.
2. Modifies PATH to intercept calls to gcc/g++/etc. and insert the appropriate 
xgill plugin arguments.  The plugin communicates with the manager via a socket.

> Once we land this we will need to land annotations to fix the false 
> positives. These annotations would have to broken up into patches with 
> module-granularity. Generally that is done by filing individual bugs for 
> each component. Should set the bugs as depedent on 541220. Hopefully, 
> module-maintaines will help out.

Are these component-level bugs mainly for dependency tracking, i.e. would it be 
possible to commit a patch that adds just a few annotations?  It would be great 
to have a low barrier to adding annotations to the source, since they don't 
change the behavior of the code at all.  One thing that could help here is 
having a web interface for annotations so that they can be added without 
modifying the source.  I had a prototype for this working last month and one of 
my TODO items is to get this working for real, so that any annotation you can 
add directly to the source can also be added over the web.

> What would it take to be airtight? Is that even realistic?

Yes, this is realistic, and is a main goal of this project.  I want to be doing 
verification, but the analysis needs to be practical first and foremost (these 
have historically been mutually exclusive).  There's some on this in the 
slides, and will be a lot more once the documentation is done.  Basically 
though, there are two holes regarding integer overflow and clobbered null 
terminators that should be fixed by writing separate analyses (using the same 
framework), and several minor holes that need to be fixed directly (mainly for 
interprocedural aliasing and call modsets).

Brian
_______________________________________________
dev-static-analysis mailing list
[email protected]
https://lists.mozilla.org/listinfo/dev-static-analysis

Reply via email to