[FFmpeg-cvslog] r14484 - in trunk/libavcodec: audioconvert.c audioconvert.h

[Ivan Kalvachev]

On 8/3/08, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Sun, Aug 03, 2008 at 01:55:50PM +0100, M?ns Rullg?rd wrote:
>> The Wanderer <inverseparadox at comcast.net> writes:
>> 
>> > Diego Biurrun wrote:
>> >
>> >> On Sat, Aug 02, 2008 at 02:55:30PM +0200, Diego Biurrun wrote:
>> >> 
>> >>> On Sat, Aug 02, 2008 at 12:12:11AM +0200, Michael Niedermayer
>> >>> wrote:
>> >>> 
>> >>>> Would it be possible that you could spell out the rule you think
>> >>>> is a rule so we all know beyond doubt if there are exceptions and
>> >>>> if so which exactly.
>> >>> 
>> >>> All files should #include everything they need directly.
>> >> 
>> >> For system headers like inttypes.h that are documented to always
>> >> #include other headers like stdint.h, I consider just inttypes.h
>> >> enough.
>> >
>> > What about for non-system headers which are, or could be, "documented"
>> > (that is, explicitly guaranteed) to always include particular other
>> > headers - or, more to the point, provide particular symbols?
>> 
>> Same thing.  Let me try to state exactly what I want the rule to be:
>> 
>>   For each type, symbol, or macro used in a file (source or header),
>>   the file shall contain an #include line guaranteed to provide,
>>   directly or indirectly, a declaration or definition of the type,
>>   symbol, or macro.
>> 
>> Is this unambiguous enough?
>
> Yes, thank you!
>
> I do not want to make a premature statement but i think this suggestion
> would be acceptable.
> Though it is certainly not what diego suggested nor how i understood
> your previous mails.

I do not like it.

1. It doesn't allow declaration or definition to be in the same file (source or header) where it is used.

2. The rule doesn't imply or limit inclusions to the necessary. You can fill headers with useless stuff, without breaking the rule.

3. The "indirect" inclusion is definitely huge step back.

4. There is no need for all headers to be self sustained. I know this is the key to that system, but definition that says "... a file (source or header included by the source)..." would work just as well.

Actually as long as we talk about definitions "... file (source)..." is sufficient. 
This is because the trick is in the word "guaranteed".  As long as it is not explicitly defined the whole definition is pure academic exercise .