the prototype decl of memcpy/memcmp/calloc/free should not in sched.h

the prototype decl of memcpy/memcmp/calloc/free should not in sched.h

[黄建忠]

[-- Attachment #1: Type: text/plain, Size: 227 bytes --]

Hi, there,

sched.h in musl should not contains the prototype declarations of "memcpy/memcmp/calloc/free" func.

memcpy/memcmp already in string.h
calloc/free already in stdlib.h

Any reason they should exist in sched.h?

[-- Attachment #2: Type: text/html, Size: 319 bytes --]

Re: the prototype decl of memcpy/memcmp/calloc/free should not in sched.h

[Rich Felker]

On Fri, Oct 31, 2014 at 11:03:35AM +0800, 黄建忠 wrote:
> Hi, there,
> 
> sched.h in musl should not contains the prototype declarations of "memcpy/memcmp/calloc/free" func.
> 
> memcpy/memcmp already in string.h
> calloc/free already in stdlib.h
> 
> Any reason they should exist in sched.h?

They're present only under _GNU_SOURCE mode, and the reason is that
musl's CPU_* macros for working with cpu sets for affinity use these
functions directly rather than having additional exported symbols in a
protected namespace to do the work. Is this approach causing a problem
for you?

Rich

Re: the prototype decl of memcpy/memcmp/calloc/free shouldnot in sched.h

[黄建忠]

[-- Attachment #1: Type: text/plain, Size: 1485 bytes --]

I encounter a problem that use clang/libc++/libc++abi only to build the whole system(As i remeber gcc should be ok).

In firefox, there is a system_wrapper header include sched.h, but in hunspell codes, calloc and free was redefined, so there is a conflicts here.
for example, in "extensions/spellcheck/hunspell/src/hunspell_alloc_hooks.h"
#define calloc(count, size) HunspellAllocator::CountingCalloc(count, size)

It's not a big issue and can be fixed in firefox.

but still feel strange that these declarations exists in sched.h and also a little bit conflict with common sense.

Please consider it.
 
------------------ Original ------------------
From:  "Rich Felker"<dalias@libc.org>;
Date:  Fri, Oct 31, 2014 11:24 AM
To:  "musl"<musl@lists.openwall.com>; 

Subject:  Re: [musl] the prototype decl of  memcpy/memcmp/calloc/free shouldnot in sched.h

 
On Fri, Oct 31, 2014 at 11:03:35AM +0800, 黄建忠 wrote:
> Hi, there,
> 
> sched.h in musl should not contains the prototype declarations of "memcpy/memcmp/calloc/free" func.
> 
> memcpy/memcmp already in string.h
> calloc/free already in stdlib.h
> 
> Any reason they should exist in sched.h?

They're present only under _GNU_SOURCE mode, and the reason is that
musl's CPU_* macros for working with cpu sets for affinity use these
functions directly rather than having additional exported symbols in a
protected namespace to do the work. Is this approach causing a problem
for you?

Rich

[-- Attachment #2: Type: text/html, Size: 1983 bytes --]

Re: the prototype decl of memcpy/memcmp/calloc/free shouldnot in sched.h

[Szabolcs Nagy]

* ????????? <jianzhong.huang@i-soft.com.cn> [2014-10-31 11:42:35 +0800]:
> 
> In firefox, there is a system_wrapper header include sched.h, but in hunspell codes, calloc and free was redefined, so there is a conflicts here.
> for example, in "extensions/spellcheck/hunspell/src/hunspell_alloc_hooks.h"
> #define calloc(count, size) HunspellAllocator::CountingCalloc(count, size)
> 

a simple fix for that in musl is

	void *(calloc)(size_t, size_t);

or

	#define CPU_ALLOC(n) __cpu_alloc(n)
	static __inline cpu_set_t *__cpu_alloc(size_t _n)
	{
		extern void *(calloc)(size_t,size_t);
		return (calloc)(1,CPU_ALLOC_SIZE(_n));
	}

which almost gets the namespace right: only a non-function like
calloc macro defined before sched.h would be an issue

getting the namespace completely right would need a new extern
identifier in musl that aliases or wraps calloc

but _GNU_SOURCE has no well defined namespace requirement
so it is hard to justify workarounds that spill over into
other parts of the libc

Re: the prototype decl of memcpy/memcmp/calloc/free shouldnot in sched.h

[Rich Felker]

On Sat, Nov 01, 2014 at 05:22:30PM +0100, Szabolcs Nagy wrote:
> * ????????? <jianzhong.huang@i-soft.com.cn> [2014-10-31 11:42:35 +0800]:
> > 
> > In firefox, there is a system_wrapper header include sched.h, but in hunspell codes, calloc and free was redefined, so there is a conflicts here.
> > for example, in "extensions/spellcheck/hunspell/src/hunspell_alloc_hooks.h"
> > #define calloc(count, size) HunspellAllocator::CountingCalloc(count, size)
> > 
> 
> a simple fix for that in musl is
> 
> 	void *(calloc)(size_t, size_t);
> 
> or
> 
> 	#define CPU_ALLOC(n) __cpu_alloc(n)
> 	static __inline cpu_set_t *__cpu_alloc(size_t _n)
> 	{
> 		extern void *(calloc)(size_t,size_t);
> 		return (calloc)(1,CPU_ALLOC_SIZE(_n));
> 	}
> 
> which almost gets the namespace right: only a non-function like
> calloc macro defined before sched.h would be an issue

Yes, I like moving the declarations to non-file-scope better.

> getting the namespace completely right would need a new extern
> identifier in musl that aliases or wraps calloc
> 
> but _GNU_SOURCE has no well defined namespace requirement
> so it is hard to justify workarounds that spill over into
> other parts of the libc

I think the problem may be that GCC activates _GNU_SOURCE by default
for C++. Am I right about this? If so, perhaps we should look for
things that could cause excessive breakage under this behavior and try
to fix them. Or (easier) we can just wait until somebody finds a
problem and address them on a per-case basis.

Rich

Re: the prototype decl of memcpy/memcmp/calloc/free shouldnot in sched.h

[Alexander Monakov]

> I think the problem may be that GCC activates _GNU_SOURCE by default
> for C++. Am I right about this?

Yes, that's true; see for example http://lwn.net/Articles/590504/ and
https://gcc.gnu.org/onlinedocs/libstdc++/faq.html#faq.predefined

Alexander

Re: the prototype decl of memcpy/memcmp/calloc/free shouldnot in sched.h

[Rich Felker]

On Sat, Nov 01, 2014 at 01:02:36PM -0400, Rich Felker wrote:
> On Sat, Nov 01, 2014 at 05:22:30PM +0100, Szabolcs Nagy wrote:
> > * ????????? <jianzhong.huang@i-soft.com.cn> [2014-10-31 11:42:35 +0800]:
> > > 
> > > In firefox, there is a system_wrapper header include sched.h, but in hunspell codes, calloc and free was redefined, so there is a conflicts here.
> > > for example, in "extensions/spellcheck/hunspell/src/hunspell_alloc_hooks.h"
> > > #define calloc(count, size) HunspellAllocator::CountingCalloc(count, size)
> > > 
> > 
> > a simple fix for that in musl is
> > 
> > 	void *(calloc)(size_t, size_t);
> > 
> > or
> > 
> > 	#define CPU_ALLOC(n) __cpu_alloc(n)
> > 	static __inline cpu_set_t *__cpu_alloc(size_t _n)
> > 	{
> > 		extern void *(calloc)(size_t,size_t);
> > 		return (calloc)(1,CPU_ALLOC_SIZE(_n));
> > 	}
> > 
> > which almost gets the namespace right: only a non-function like
> > calloc macro defined before sched.h would be an issue
> 
> Yes, I like moving the declarations to non-file-scope better.

I think we should do this, but while looking at how, I noticed a
couple other issues:

1. Useless declaration of memcpy, which is unused.
2. Use of memset without a declaration.

Just making a note here so I don't overlook these when fixing it.

Rich