[cfe-dev] support for parsing more GCC object-size checking builtins

[Shantonu Sen]

I'm running into problems parsing code that builds with - 
D_FORTIFY_SOURCE=2 on Mac OS X. Specifically, this causes calls to  
certain functions to be replacing like so:
> [shantonu at shantonu-sens-macbook-2]$ cat test.c
> #include <string.h>
>
> extern char *foo;
> extern const char *bar;
>
> void * test(void) {
>     return stpcpy(foo, bar);
> }
> [shantonu at shantonu-sens-macbook-2]$ gcc -c test.c -D_FORTIFY_SOURCE=2
> [shantonu at shantonu-sens-macbook-2]$ nm test.o | grep " U "
>          U ___stpcpy_chk
>          U _bar
>          U _foo
> [shantonu at shantonu-sens-macbook-2]$ gcc -c test.c -D_FORTIFY_SOURCE=0
> [shantonu at shantonu-sens-macbook-2]$ nm test.o | grep " U "
>          U _bar
>          U _foo
>          U _stpcpy
> [shantonu at shantonu-sens-macbook-2]$ clang -D_FORTIFY_SOURCE=2 -E  
> test.c | tail -4
>
> void * test(void) {
>     return ((__builtin_object_size (foo, 0) != (size_t) -1) ?  
> __builtin___stpcpy_chk (foo, bar, __builtin_object_size (foo, 2 >  
> 1)) : __inline_stpcpy_chk (foo, bar));
> }

clang does not appear to support all of the builtins documented in the  
GCC manual at <http://gcc.gnu.org/onlinedocs/gcc/Object-Size-Checking.html 
 >. Attempting to use stpcpy (or strncat, etc.), I get:

> [shantonu at shantonu-sens-macbook-2]$ clang.orig -D_FORTIFY_SOURCE=2  
> test.c
> test.c:7:12: error: incompatible operand types ('int' and 'char *')
>     return stpcpy(foo, bar);
>            ^~~~~~
> 1 diagnostic generated.
> [shantonu at shantonu-sens-macbook-2]$ clang.orig -D_FORTIFY_SOURCE=2  
> test.c -E | clang.orig - -fsyntax-only
> <stdin>:273:3: warning: incompatible integer to pointer conversion  
> returning 'int', expected 'char *'
>   return __builtin___stpcpy_chk (__dest, __src,  
> __builtin_object_size (__dest, 2 > 1));
>   ^       
> ~ 
> ~ 
> ~ 
> ~ 
> ~ 
> ~ 
> ~ 
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> <stdin>:308:3: warning: incompatible integer to pointer conversion  
> returning 'int', expected 'char *'
>   return __builtin___strncat_chk (__dest, __src, __len,  
> __builtin_object_size (__dest, 2 > 1));
>   ^       
> ~ 
> ~ 
> ~ 
> ~ 
> ~ 
> ~ 
> ~ 
> ~ 
> ~ 
> ~ 
> ~ 
> ~ 
> ~ 
> ~ 
> ~ 
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> <stdin>:317:61: error: incompatible operand types ('int' and 'char *')
>     return ((__builtin_object_size (foo, 0) != (size_t) -1) ?  
> __builtin___stpcpy_chk (foo, bar, __builtin_object_size (foo, 2 >  
> 1)) : __inline_stpcpy_chk (foo, bar));
>                                                             ^  
> ~ 
> ~ 
> ~ 
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> 3 diagnostics generated.

I added the missing builtins to at least the parser, and now I get:
> [shantonu at shantonu-sens-macbook-2]$ clang -D_FORTIFY_SOURCE=2 test.c
> [shantonu at shantonu-sens-macbook-2]$ clang -D_FORTIFY_SOURCE=2 test.c  
> -E | clang - -fsyntax-only
> [shantonu at shantonu-sens-macbook-2]$

This doesn't add codegen support, since there does not appear to be  
support for any of the other built-in string functions. But this is  
enough to allow checker to successfully parse my entire project.

The existing prototype for __builtin_memset seemed incorrect, since  
memset(3) takes an "int" as the pattern. __builtin_strcpy/strncpy also  
had issues because the destination string was marked as "const" when  
it shouldn't have been.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: builtin-object-size.patch
Type: application/octet-stream
Size: 1518 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20080901/16d53575/attachment.obj>
-------------- next part --------------

Shantonu

Sent from my MacBook