[Tinycc-devel] Re: malloc(0)

[Charlie Gordon]

Hello folks!

Here comes one of my favorites: libc behaviour at the limits!

- ANSI C 1990 does not specify precisely what the semantics of malloc(0)
 it only says that free(NULL) is legitimate, and realloc(ptr,0) should free 
ptr.
- a quick note here on free(NULL): whenever you implement a wrapper on 
malloc/free
 make it adhere to this convention, it is very useful and becomes quite 
natural;
 I only test for NULL ptr before calling free() in cases where the function 
call
 would be too costly, a rare occasion.  As a matter of fact, very few C++ 
addicts
 know that its perfectly fine to 'delete' a NULL pointer.
- ISO/IEC 9899:1999 is more specific on the malloc(0) issue: section 7.20.3 
pg 324
 read: "if the size of the space request is zero, the behaviour is 
implementation-
 defined: either a null pointer is returned, or the behaviour is as the 
size
 were some nonzero value, except that the returned pointer shall not be 
used
 to access an object.
- tcc returns non NULL for malloc(0) because it allocates 1 more byte than
 requested to ensure that malloc'd areas are never contiguous.  This is 
needed
 for bound checking.  ISO reads in section 7.20.3 that each successful 
allocation
 shall yield a pointer to an object disjoint from any other object, yet 
that
 order and contiguity of storage is unspecified.

Conclusion: do not rely on implementation defined behaviour. Both glibc and 
ulibc
are OK on this issue, TCC internal libc can implement one or the other, 
preferably
glibc's, but not necessarily.

There are a lot of small details like these to polish in the C language,
keep contributing to tcc by testing it to the limits!

Chqrlie.

___________________________________________________________
Sent by ePrompter, the premier email notification software.
Free download at http://www.ePrompter.com.

_________________________________________________________________
MSN Photos is the easiest way to share and print your photos: 
http://photos.msn.com/support/worldwide.aspx