Re: Git's static analysis

fredag 06 februari 2009 02:19:15 skrev Junio C Hamano:

Pieter de Bie [off-list ref] writes:

quoted

I played around a bit with the 'Clang' static analyser, and tried to run git's
source code through it. It comes up with a few possible errors, so I thought
you might find it interesting. I took a quick glance, and it also seems to
have a few false positives, but it might still be worth to take a look.

The results can be found here:

	http://frim.frim.nl/git-analyse/

Hmm, I took a quick look at a few, and they looked nonsense, but perhaps I
am misreading things.

For example:

    http://frim.frim.nl/git-analyse/report-uxXiUR.html#EndPath

I am assuming that we follow the control flow of the labelled comments, so
I followed along from [1] to [7] and then saw these:

    [8] loop condition is false, execution continues on line 1492
    1483:   for (i = 0; i < array->nr; i++) {
                ...
            }

    [9] taking false branch
    1492:   if (array->nr <= i)
                return NULL;

    [10] dereference of null pointer.
    1495:   c->object.flags |= ...

The thing is, if [8] exits, "i < array->nr" is not true anymore, and there
is no way you can take false branch of  "if (array->nr <= i)" in the
immediately next step [9]. and reach point [10].

The code assumes can c become null in the loop [if (!c) continue]. If that
is the last iteration it comes out of the loop with c == NULL and array->nr >=i,
thus not returning. 

I have to dig through history until may 2008 to find this version of this code  so
the analysis seems a bit obsolete. The loop was rewritten in 4603ec0f960e.

-- robin