[PATCH] Warn on mismatching types to sizeof for memset and friends where length is of the form sizeof(Type) * factor.

[Oliver Chang]

Here are some general results after some quick testing (some of which is stolen from thakis's testing on chromium). Unfortunately some of these warnings were difficult to triage and requires a significant amount of time (which I think makes a point for this warning).

Discounting duplicates in the same file, both Firefox and Chromium give around 8-10 warnings.

Some patterns observed:

- Most seem to just be based on size assumptions between types, e.g. 2 * sizeof(uint16_t) == sizeof(uint32_t).

Firefox had about 2 instances where 2 different classes were expected to have the same size (e.g. when 1 is a wrapper around another).

- In Chromium there are some issues with structs and unions that solely contained members of a single type.

e.g.
struct M {

  int A[4];

};

M a;
memset(&a, 0, sizeof(int) * 4);

thakis@ suggested decaying the struct type to the member type in this case for the comparison.

- Firefox also had about 2 warnings related to multidimensional arrays, e.g.

typedef int Foo[10];
Foo m;
memset(&m, 0, sizeof(Foo));

Perhaps we can just compare the type 'Foo' in this case before trying to break it down further in this case.

- There are some other warnings such as differences complex_t vs float[2], w_char vs short.


http://reviews.llvm.org/D8881

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/