dirent.h and __NEED_size_t

dirent.h and __NEED_size_t

[Daniel Cegiełka]

[-- Attachment #1: Type: text/plain, Size: 15 bytes --]

regads,
daniel

[-- Attachment #2: dirent_h__NEED_size_t.diff --]
[-- Type: application/octet-stream, Size: 316 bytes --]

diff -urN musl.orig//include/dirent.h musl/include/dirent.h
--- musl.orig//include/dirent.h	2012-08-24 14:16:18.000000000 +0200
+++ musl/include/dirent.h	2012-08-24 17:11:03.156857822 +0200
@@ -55,6 +55,7 @@
 #endif
 
 #ifdef _BSD_SOURCE
+#define __NEED_size_t
 int getdents(int, struct dirent *, size_t);
 #endif
 

Re: dirent.h and __NEED_size_t

[Daniel Cegiełka]

Sorry - this patch doesn't work properly. However, it breaks
compilation if we don't have a visible size_t...


2012/8/24 Daniel Cegiełka <daniel.cegielka@gmail.com>:
> regads,
> daniel

Re: dirent.h and __NEED_size_t

[Daniel Cegiełka]

[-- Attachment #1: Type: text/plain, Size: 269 bytes --]

Ok! :)

2012/8/24 Daniel Cegiełka <daniel.cegielka@gmail.com>:
> Sorry - this patch doesn't work properly. However, it breaks
> compilation if we don't have a visible size_t...
>
>
> 2012/8/24 Daniel Cegiełka <daniel.cegielka@gmail.com>:
>> regads,
>> daniel

[-- Attachment #2: dirent_h__NEED_size_t.diff --]
[-- Type: application/octet-stream, Size: 344 bytes --]

diff -urN musl.orig//include/dirent.h musl/include/dirent.h
--- musl.orig//include/dirent.h	2012-08-24 14:16:18.000000000 +0200
+++ musl/include/dirent.h	2012-08-24 17:30:57.701972837 +0200
@@ -55,6 +55,8 @@
 #endif
 
 #ifdef _BSD_SOURCE
+#define __NEED_size_t
+#include <bits/alltypes.h>
 int getdents(int, struct dirent *, size_t);
 #endif
 

Re: Re: dirent.h and __NEED_size_t

[Gregor Richards]

[-- Attachment #1: Type: text/plain, Size: 532 bytes --]

It's probably best to avoid multiply including bits/alltypes.h within 
one file (although it's not a big deal since that file will be included 
dozens of times regardless).

With valediction,
  - Gregor Richards

On 08/24/2012 11:32 AM, Daniel Cegiełka wrote:
> Ok! :)
>
> 2012/8/24 Daniel Cegiełka <daniel.cegielka@gmail.com>:
>> Sorry - this patch doesn't work properly. However, it breaks
>> compilation if we don't have a visible size_t...
>>
>>
>> 2012/8/24 Daniel Cegiełka <daniel.cegielka@gmail.com>:
>>> regads,
>>> daniel


[-- Attachment #2: dirent_size_t.diff --]
[-- Type: text/x-diff, Size: 310 bytes --]

diff -r fab1146cec43 include/dirent.h
--- a/include/dirent.h	Thu Aug 23 14:38:34 2012 -0400
+++ b/include/dirent.h	Fri Aug 24 11:39:38 2012 -0400
@@ -7,6 +7,9 @@
 
 #define __NEED_ino_t
 #define __NEED_off_t
+#ifdef _BSD_SOURCE
+#define __NEED_size_t /* for getdents */
+#endif
 
 #include <bits/alltypes.h>
 

Header testing [Re: [musl] Re: dirent.h and __NEED_size_t]

[Rich Felker]

On Fri, Aug 24, 2012 at 11:40:34AM -0400, Gregor Richards wrote:
> diff -r fab1146cec43 include/dirent.h
> --- a/include/dirent.h	Thu Aug 23 14:38:34 2012 -0400
> +++ b/include/dirent.h	Fri Aug 24 11:39:38 2012 -0400
> @@ -7,6 +7,9 @@
>  
>  #define __NEED_ino_t
>  #define __NEED_off_t
> +#ifdef _BSD_SOURCE
> +#define __NEED_size_t /* for getdents */
> +#endif

It would be nice to have a header test suite for this kind of issue.
Basically, it would attempt compiling an otherwise-empty source file
for each system header, with each combination of feature test macros,
to check for errors in any of the cases. We could add -std=c89 to the
mix too if we want to check for c89-compatibility errors at the same
time...

Rich

Re: Header testing [Re: [musl] Re: dirent.h and __NEED_size_t]

[John Spencer]

On 08/24/2012 08:23 PM, Rich Felker wrote:
>
> It would be nice to have a header test suite for this kind of issue.
> Basically, it would attempt compiling an otherwise-empty source file
> for each system header, with each combination of feature test macros,
> to check for errors in any of the cases. We could add -std=c89 to the
> mix too if we want to check for c89-compatibility errors at the same
> time...
>
> Rich
>
results: http://sprunge.us/JURU
script:  http://sprunge.us/HIFE

apart from "inline" there seem to be only 2 problems (at least from a 
quick glance):
bits/alltypes.h:

#if defined(__NEED_struct_timespec) && !defined(__DEFINED_struct_timespec)
struct timespec { time_t tv_sec; long tv_nsec; }; // time_t can be undefined
#define __DEFINED_struct_timespec
#endif

arpa/nameser.h should include support for uint16_t


.. and: <nsz> the stdio_ext.h bug is that stdio.h is not included

Re: Header testing [Re: [musl] Re: dirent.h and __NEED_size_t]

[Rich Felker]

On Fri, Aug 24, 2012 at 10:04:22PM +0200, John Spencer wrote:
> On 08/24/2012 08:23 PM, Rich Felker wrote:
> >
> >It would be nice to have a header test suite for this kind of issue.
> >Basically, it would attempt compiling an otherwise-empty source file
> >for each system header, with each combination of feature test macros,
> >to check for errors in any of the cases. We could add -std=c89 to the
> >mix too if we want to check for c89-compatibility errors at the same
> >time...
> >
> >Rich
> >
> results: http://sprunge.us/JURU
> script:  http://sprunge.us/HIFE
> 
> apart from "inline" there seem to be only 2 problems (at least from
> a quick glance):
> bits/alltypes.h:
> 
> #if defined(__NEED_struct_timespec) && !defined(__DEFINED_struct_timespec)
> struct timespec { time_t tv_sec; long tv_nsec; }; // time_t can be undefined
> #define __DEFINED_struct_timespec
> #endif

It's up to the header requesting timespec to also request time_t. We
could avoid this by just duplicating the underlying type for time_t in
the struct definition; I'm not sure if this would be desirable, since
it might cause errors of omitting time_t from headers that need to
define it to get missed.

> arpa/nameser.h should include support for uint16_t

Fixed.

> ... and: <nsz> the stdio_ext.h bug is that stdio.h is not included

Fixed.

Rich

Re: Header testing [Re: [musl] Re: dirent.h and __NEED_size_t]

[John Spencer]

On 08/24/2012 10:16 PM, Rich Felker wrote:
> On Fri, Aug 24, 2012 at 10:04:22PM +0200, John Spencer wrote:
>>
>> bits/alltypes.h:
>>
>> #if defined(__NEED_struct_timespec)&&  !defined(__DEFINED_struct_timespec)
>> struct timespec { time_t tv_sec; long tv_nsec; }; // time_t can be undefined
>> #define __DEFINED_struct_timespec
>> #endif
> It's up to the header requesting timespec to also request time_t. We
> could avoid this by just duplicating the underlying type for time_t in
> the struct definition; I'm not sure if this would be desirable, since
> it might cause errors of omitting time_t from headers that need to
> define it to get missed.
>
in that case, sys/sem.h (which caused the compilation error) needs to be 
fixed to set __NEED_time_t

Re: Header testing [Re: [musl] Re: dirent.h and __NEED_size_t]

[Rich Felker]

On Fri, Aug 24, 2012 at 10:48:18PM +0200, John Spencer wrote:
> On 08/24/2012 10:16 PM, Rich Felker wrote:
> >On Fri, Aug 24, 2012 at 10:04:22PM +0200, John Spencer wrote:
> >>
> >>bits/alltypes.h:
> >>
> >>#if defined(__NEED_struct_timespec)&&  !defined(__DEFINED_struct_timespec)
> >>struct timespec { time_t tv_sec; long tv_nsec; }; // time_t can be undefined
> >>#define __DEFINED_struct_timespec
> >>#endif
> >It's up to the header requesting timespec to also request time_t. We
> >could avoid this by just duplicating the underlying type for time_t in
> >the struct definition; I'm not sure if this would be desirable, since
> >it might cause errors of omitting time_t from headers that need to
> >define it to get missed.
> >
> in that case, sys/sem.h (which caused the compilation error) needs
> to be fixed to set __NEED_time_t

Fixed, along with a couple other bugs I found reading the
specification for that header..

Rich