llvm dev - May 2008 - [LLVMdev] optimization assumes malloc return is non-null

One more question on this.

The substitution of malloc() is predicated on the assumption that the
compiler knows the implementation of malloc(), and the argument for that
seems to rest in part on the specification of the Standard C Library.

But I am not aware of any requirement that compliant C code must be
linked with the Standard C Library. If this requirement does not exist,
then no portion of the C library specification can be used to justify a
compile-time optimization unless we have some sort of compilation model
information that tells us the C library can be assumed.

Is there a requirement somewhere in the C *Language* Specification that
ties all of this together in the required way?


shap

Jonathan S. Shapiro wrote:-
> One more question on this.
> 
> The substitution of malloc() is predicated on the assumption that the
> compiler knows the implementation of malloc(), and the argument for that
> seems to rest in part on the specification of the Standard C Library.
> 
> But I am not aware of any requirement that compliant C code must be
> linked with the Standard C Library. If this requirement does not exist,
> then no portion of the C library specification can be used to justify a
> compile-time optimization unless we have some sort of compilation model
> information that tells us the C library can be assumed.
> 
> Is there a requirement somewhere in the C *Language* Specification that
> ties all of this together in the required way?
Reserved identifiers and header inclusion.

Neil.

Hi,
> The substitution of malloc() is predicated on the assumption that the
> compiler knows the implementation of malloc(), and the argument for that
> seems to rest in part on the specification of the Standard C Library.
> 
> But I am not aware of any requirement that compliant C code must be
> linked with the Standard C Library. If this requirement does not exist,
> then no portion of the C library specification can be used to justify a
> compile-time optimization unless we have some sort of compilation model
> information that tells us the C library can be assumed.
> 
> Is there a requirement somewhere in the C *Language* Specification that
> ties all of this together in the required way?
this is what -ffreestanding is for.  More generally, don't run
SimplifyLibcalls
if you are not going to link with the standard C library.

Ciao,

Duncan.

On Wed, 2008-05-14 at 23:23 +0900, Neil Booth wrote:
> Jonathan S. Shapiro wrote:-
> > Is there a requirement somewhere in the C *Language* Specification
that
> > ties all of this together in the required way?
> 
> Reserved identifiers and header inclusion.
> 
> Neil.
What section of the C standard do I need to refresh myself on here?

On May 14, 2008, at 6:31 AM, Jonathan S. Shapiro wrote:
> The substitution of malloc() is predicated on the assumption that the
> compiler knows the implementation of malloc(), and the argument for  
> that
> seems to rest in part on the specification of the Standard C Library.
This is kinda a compilers 101 type question...

The Standard C Library is part of the C Standard.  Compilers compile  
for that standard and can make use of anything contained in it in any  
way they see fit.
> But I am not aware of any requirement that compliant C code must be
> linked with the Standard C Library.
The requirement is present, See

        -ffreestanding
            Assert that compilation takes place in a freestanding  
environment.  This implies -fno-builtin.  A freestanding environment  
is one in
            which the standard library may not exist, and program  
startup may not necessarily be at "main".  The most obvious example is
an OS
            kernel.  This is equivalent to -fno-hosted.

for example.
> Is there a requirement somewhere in the C *Language* Specification  
> that
> ties all of this together in the required way?
Yes.  The C standard.  Likewise the C++ language standard, likewise  
any language standard.