change log for rtems (2010-06-15)

[Chris Johns]

On 16/06/10 5:26 PM, Sebastian Huber wrote:
> On 06/15/2010 06:11 PM, rtems-vc at rtems.org wrote:
> [...]
>> diff -u rtems/cpukit/ChangeLog:1.2346.2.16 rtems/cpukit/ChangeLog:1.2346.2.17
>> --- rtems/cpukit/ChangeLog:1.2346.2.16	Tue Jun 15 08:35:48 2010
>> +++ rtems/cpukit/ChangeLog	Tue Jun 15 10:17:42 2010
>> @@ -1,5 +1,6 @@
>>   2010-06-15	Ralf Corsépius<ralf.corsepius at rtems.org>
>>
>> +	* libfs/src/rfs/rtems-rfs-file-system.h: Remove bogus typecast.
>>   	* libfs/src/rfs/rtems-rfs-block.c, libfs/src/rfs/rtems-rfs-buffer.c,
>>   	libnetworking/nfs/bootp_subr.c: Misc. 64bit-compatibility fixes.
>>   	* posix/include/rtems/posix/pthread.h: Remove stray comment.
>>
>> diff -u rtems/cpukit/libfs/src/rfs/rtems-rfs-file-system.h:1.3 rtems/cpukit/libfs/src/rfs/rtems-rfs-file-system.h:1.4
>> --- rtems/cpukit/libfs/src/rfs/rtems-rfs-file-system.h:1.3	Thu Feb 25 23:54:59 2010
>> +++ rtems/cpukit/libfs/src/rfs/rtems-rfs-file-system.h	Tue Jun 15 10:17:35 2010
>> @@ -308,7 +308,7 @@
>>    *
>>    * @param _fs Pointer to the file system.
>>    */
>> -#define rtems_rfs_fs_size(_fs) (((uint64_t) rtems_rfs_fs_blocks (_fs)) * \
>> +#define rtems_rfs_fs_size(_fs) (rtems_rfs_fs_blocks (_fs) * \
>>                                   rtems_rfs_fs_block_size (_fs))
>>
>>   /**
>>
> [...]
>
> The blocks and block_size fields are of type size_t, thus I don't think that
> this type cast is bogus.
>

Hmm indeed the change would seem broken. The original and I think 
correct code created an intermediate value in compiler that is 64bits in 
size so the maths would not truncate. The cast is needed and valid. The 
RFS is a 64bit file system. It was carefully tested on large disks and 
large files. If uint64_t is not supported by a target or broken then I 
would assert the target is broken as my understanding is uint64_t is 
standard.

If a warning appears when assigning the result then there is a problem 
in that area.

If the implementation of how the macro is done is not correct then lets 
fix that. If the 64bit target needs something different or creates 
warnings lets seem them and fix them.

I will now have to go back and review all the changes in detail to the 
RFS. Unless these type of changes, ie not printf ones are reverted, I 
would have to say the RFS is broken and so this effects the 4.10 
release. I see the comment on the 4.10 branch but view cvs is broken on 
the 4.10 branch so not review the "bogus" cast claim.

> Since size_t may have only 16-bits on some architectures we should consider to
> use uint32_t for these fields.

If this is the case then I may review this for 4.11.

Thanks for pointing this out.

Chris