Valgrind Finds Thousands Of Potential Issues With Mesa [Phoronix]

While valgrind finds issues, it's totally unclear if these issues are serious. The explanation of the Mesa developer that responded sounds reasonable (valgrind cannot track past ioctl).

I'm wondering, if this article is more about valgrind than about Mesa: what does it help if only a fraction of the reported issues is an actual issue?

Well, most of the valgrind errors should be repetitions, so I would say there are probably 5000 error or so.

It is a great idea to use Valgrind or other automatic tools(we make our own), but it is really hard when you had already developed a project without it.

Going from 507,326 errors to 507,325 or 507,310 is not very motivating(0.000029% progress).

It's a little unfortunate that the author did not at least try to hunt down a few of them and present a conlclusion about if the warnings are benign or serious.

I don't have the exact version of Mesa that the blog-post (http://www.americanteeth.org/2013/08/14/valgrind-is-not-opti...) seems to refer to. But the first two functions in the i965 driver code that valgrind complained about basically do something like the following code:

    uint32_t *surf = brw_state_batch(...)
    surf[0] = ...stuff with flags for the GPU-code...
    (...)
    surf[5] = ...even more stuff, encoding coordinates and so on...
    (...)

Unfortunately that pointer "*surf" seems to be allocated by some batch-buffer-memory allocation magic, so (after only greping through the source for 5 minutes) I would not be able to conclude if it's a fake warning (maybe refering to some memory that is allocated by some DRI kernel interface mmap-magic) or indeed serious (say, a mistake when calculating the proper boundaries of allocated pages). But with a library used as regularly as Mesa I'd give the authors the benefit of doubt and guess that to detect the subtle errors, one would first have to educate valgrind about that DRI-specific way of allocating memory.

This is completely a non-story. If you're going to write a blog entry about a project's valgrind results at least track down one of the reports that's actually a bug.

The joys of safety when using C.

Validation tools and warnings as errors are not optional when using C and its derivatives (Obj-C/C++).

Not everyone is an elite C developer and even those can make mistakes.