[cfe-commits] unix.Malloc static checker improvement: memory.LeakPtrValChanged

[Anna Zaks]

On Feb 8, 2013, at 6:13 AM, Branden Archer <b.m.archer4 at gmail.com> wrote:

> Anna,
> 
> Thanks for committing these!
> 
> Who should I contact to have the LeakPtrValChanged section of the following page updated/removed, as the checker is now in?
> http://clang-analyzer.llvm.org/potential_checkers.html
> 
Just send in a patch. The html page is in the clang repository.

Anna.
> - Branden
> 
> On Thu, Feb 7, 2013 at 6:09 PM, Anna Zaks <ganna at apple.com> wrote:
> Branden,
> 
> Thanks for working on this and polishing it up! I've committed the changes in r174677 and r 174678.
> 
> I had to change the tests from the first patch slightly since we've made analyzer more pessimistic as a quick fix to a regression in r174468. This resulted in the expected leaks not being reported.
> 
> Cheers,
> Anna.
> On Feb 3, 2013, at 8:10 PM, Branden Archer <b.m.archer4 at gmail.com> wrote:
> 
>> Here is the commit message for the first patch:
>> [analyzer] Pass pointer escape type to checkPointerEscape checker
>>     
>> The checkPointerEscape checker previously did not specify how a
>> pointer escaped. This change includes an enum which describes the
>> different ways a pointer may escape. This enum is passed to the
>> checkPointerEscape checker when a pointer escapes. If the escape
>> is due to a function call, the call is passed. This changes
>> previous behavior where the call is passed as NULL if the escape
>> was due to indirectly invalidating the region the pointer referenced.
>> 
>> Here is the commit message for the second patch:
>> [analyzer] malloc checker reports bugs when freeing memory with offset pointer
>>     
>> The malloc checker will now catch the case when a previously malloc'ed
>> region is freed, but the pointer passed to free does not point to the
>> start of the allocated memory. For example:
>>     
>> int *p1 = malloc(sizeof(int));
>> p1++;
>> free(p1); // warn
>>     
>> From the "memory.LeakPtrValChanged enhancement to unix.Malloc" entry
>> in the list of potential checkers.
>> 
>> 
>> What I meant is that the tests should not pass before the patch is applied. Then, after you apply the patch, the tests should start passing.
>> 
>> Thanks for the reminder. The other tests for the second patch checked out, but what I was trying for the first patch was not. I decided to take a more direct approach to testing the first patch. Instead of trying to figure out a system header function that would take a struct containing a pointer (or for the simple stream checker, a FILE pointer), I decided to 'create my own' by adding a fake system header function in the header included in the test files. In doing so, I can easily see that previous to the change there was a false negative, but after the change a bug was found.
>> 
>> - Branden
>> 
>> On Fri, Feb 1, 2013 at 1:20 PM, Anna Zaks <ganna at apple.com> wrote:
>> 
>> On Jan 31, 2013, at 7:20 PM, Branden Archer <b.m.archer4 at gmail.com> wrote:
>> 
>>> Oh, one other thing. I do not know who does the committing when the patch is approved. Should I prove a patch that has my commit message in it and username? Would either of you be pushing the change, or would I be pushing it?
>>> 
>> 
>> One of us will push this and mention that the patch is by you in the commit message. Please do send the commit message!
>> 
>>> Also, thanks for your comments and discussion on this patch, and for keeping with it for so long! I appreciate your feedback and the opportunity to learn some about the static analyzer of clang.
>>> 
>>> - Branden
>>> 
>>> On Thu, Jan 31, 2013 at 10:10 PM, Branden Archer <b.m.archer4 at gmail.com> wrote:
>>> Have you double checked that the tests did not generate the warnings before the patch?
>>> 
>>> When you mentioned this, I had a moment of doubt. I checked for compiler warnings and the 'make test' before I emailed my patches, and all came back clean. Just to be sure I updated my source and compiled again, all still clean. Did you see something in particular, or just wanted to make sure?
>>> 
>> 
>> What I meant is that the tests should not pass before the patch is applied. Then, after you apply the patch, the tests should start passing.
>> 
>>> 
>>> +void testPassConstPointerIndirectly() {
>>> +  struct HasPtr hp;
>>> +  hp.p = fopen("myfile.txt", "w");
>>> +  fputc(0, (FILE *)&hp);
>>> +  return; // expected-warning {{Opened file is never closed; potential resource leak}}
>>> +}
>>> 
>>> Heh. Did you really want this test case? It's not actually valid (&hp is a FILE**, not a FILE*):
>>> 
>>> I knew the test was not proper code, but it was the only way I could think of to pass a structure to a known library function that was known to not close a file. I replaced it in the attached patches with passing the HasPtr structure to a function that accepts it as a const parameter. I am not sure this still tests the same thing (as I do not fully appreciate how the analyzer knows that a function will not close the stream. I am hoping that if the parameter is passed as a const that it will assume this).
>>> 
>>> - Branden
>>> 
>>> 
>>> On Thu, Jan 31, 2013 at 8:26 PM, Jordan Rose <jordan_rose at apple.com> wrote:
>>> 
>>> On Jan 31, 2013, at 5:21 , Branden Archer <b.m.archer4 at gmail.com> wrote:
>>> 
>>>> Basically, you need to pass a pointer which we are tracking to a function call indirectly (ex: as a field in a struct..). You should pass it to a function which is known not to free memory or close stream. Finally, you leak that resource/pointer.
>>>> 
>>>> Previously, we would have a false negative - no leak would be reported. Now, we should be catching the leak.
>>>> 
>>>> Ah, got you. See the first attached patch for these added cases.
>>>> 
>>>> - Branden
>>> 
>>> 
>>> +void testPassConstPointerIndirectly() {
>>> +  struct HasPtr hp;
>>> +  hp.p = fopen("myfile.txt", "w");
>>> +  fputc(0, (FILE *)&hp);
>>> +  return; // expected-warning {{Opened file is never closed; potential resource leak}}
>>> +}
>>> 
>>> Heh. Did you really want this test case? It's not actually valid (&hp is a FILE**, not a FILE*):
>>> 
>>> 
>>> A few remaining comments for the MallocChecker patch:
>>> 
>>> +  if (ExplodedNode *N = C.generateSink()) {
>>> 
>>> Please use an early return here.
>>> 
>>> 
>>> +    int offsetBytes = Offset.getOffset()/C.getASTContext().getCharWidth();
>>> 
>>> Very nitpicky, but can you put spaces around the /?
>>> 
>>> 
>>> +       << ((abs(offsetBytes) > 1) ? "bytes" : "byte")
>>> 
>>> Perfect!
>>> 
>>> Jordan
>>> 
>>> 
>> 
>> 
>> <01-modify-checkPointerEscape.patch><02-update-malloc-checker.patch>
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130208/b4bd0dfc/attachment.html>