Re: [ft-devel] freetype, undefined behaviour, and clang

[suzuki toshiya]

Hi,

Thanks, I have to thank you (as always) for finding the issue.
The part was introduced by me,

commit 5d3ff05615dda6d1325ed612381a17a0df04c975
Author: suzuki toshiya <address@hidden>
Date:   Sat Aug 1 00:32:24 2009 +0900

    cache: Check higher bits in flags for non ILP32 systems.

and I remember there are more similar codes to deal with the
higher bits in the part mixing long & int variables. Maybe
bitmask calculation would be better than playing with the
MIN/MAX.

Regards,
mpsuzuki

Sean McBride wrote:
> Hi all,
> 
> I recently read this 3 part blog article by the architect of llvm/clang, 
> which I highly recommend:
> 
> <http://blog.llvm.org/2011/05/what-every-c-programmer-should-know.html>
> <http://blog.llvm.org/2011/05/what-every-c-programmer-should-know_14.html>
> <http://blog.llvm.org/2011/05/what-every-c-programmer-should-know_21.html>
> 
> In short, clang has two compiler flags that catch undefined behaviour. He 
> described them thusly:
> 
> "Clang has an experimental -fcatch-undefined-behavior mode that inserts 
> runtime checks to find violations like shift amounts out of range, some 
> simple array out of range errors, etc. This is limited because it slows down 
> the application's runtime and it can't help you with random pointer 
> dereferences (like Valgrind can), but it can find other important bugs. Clang 
> also fully supports the -ftrapv flag (not to be confused with -fwrapv) which 
> causes signed integer overflow bugs to trap at runtime (GCC also has this 
> flag, but it is completely unreliable/buggy in my experience)."
> 
> For fun, I enabled these flags on the VTK project and ran its unit tests.  
> Some tests now crash, in freetype code.  Example:
> 
> ftcbasic.c
> 360         if ( (FT_ULong)(type->flags - FT_INT_MIN) > FT_UINT_MAX )
> 
> (gdb) p type->flags
> $1 = 4
> 
> On my system, INT_MIN is (-2147483647-1) and so this is in fact an overflow.
> 
> I might one day have time to submit patches, but I though I'd point out this 
> tool incase any of you would like to take a stab at it.
> 
> Cheers,
>