On Sat, Jun 18, 2011 at 2:03 PM, Alex Converse <[email protected]> wrote:
> 2011/6/18 Måns Rullgård <[email protected]>:
>> Alex Converse <[email protected]> writes:
>>
>>> 2011/6/18 Måns Rullgård <[email protected]>:
>>>> Alex Converse <[email protected]> writes:
>>>>
>>>>> 2011/6/18 Måns Rullgård <[email protected]>:
>>>>>> Alex Converse <[email protected]> writes:
>>>>>>
>>>>>>> 2011/6/18 Måns Rullgård <[email protected]>:
>>>>>>>> Yusuke Nakamura <[email protected]> writes:
>>>>>>>>
>>>>>>>>> 2011/6/18 Måns Rullgård <[email protected]>
>>>>>>>>>
>>>>>>>>>> Yusuke Nakamura <[email protected]> writes:
>>>>>>>>>>
>>>>>>>>>> > This patch fixes my 10L of following.
>>>>>>>>>> >
>>>>>>>>>> http://git.libav.org/?p=libav.git;a=commitdiff;h=5f0bb0baefd506d684adfa1ad4259c65973b455e
>>>>>>>>>> >
>>>>>>>>>> > From f3c030ebedabc9a17e377c6f91dc417e6578712b Mon Sep 17 00:00:00 
>>>>>>>>>> > 2001
>>>>>>>>>> > From: Yusuke Nakamura <[email protected]>
>>>>>>>>>> > Date: Sun, 5 Jun 2011 01:28:43 +0900
>>>>>>>>>> > Subject: [PATCH] mov: Fix empty edit detection.
>>>>>>>>>> >
>>>>>>>>>> > ---
>>>>>>>>>> >  libavformat/mov.c |    2 +-
>>>>>>>>>> >  1 files changed, 1 insertions(+), 1 deletions(-)
>>>>>>>>>> >
>>>>>>>>>> > diff --git a/libavformat/mov.c b/libavformat/mov.c
>>>>>>>>>> > index e6ada4e..2d1d726 100644
>>>>>>>>>> > --- a/libavformat/mov.c
>>>>>>>>>> > +++ b/libavformat/mov.c
>>>>>>>>>> > @@ -2230,7 +2230,7 @@ static int mov_read_elst(MOVContext *c, 
>>>>>>>>>> > AVIOContext
>>>>>>>>>> *pb, MOVAtom atom)
>>>>>>>>>> >              time     = avio_rb64(pb);
>>>>>>>>>> >          } else {
>>>>>>>>>> >              duration = avio_rb32(pb); /* segment duration */
>>>>>>>>>> > -            time     = avio_rb32(pb); /* media time */
>>>>>>>>>> > +            time     = (int32_t)avio_rb32(pb); /* media time */
>>>>>>>>>>
>>>>>>>>>> This cast is invalid if the value is >INT_MAX.
>>>>>>>>>
>>>>>>>>> What's wrong? If I get 0x80000000 from avio_rb32(pb), then I want to 
>>>>>>>>> set
>>>>>>>>> 0xffffffff80000000 to int64_t time.
>>>>>>>>
>>>>>>>> The cast is not guaranteed to do that.
>>>>>>>
>>>>>>> On GCC it is...
>>>>>>
>>>>>> Are you sure about that?  What about all those cases where it optimises
>>>>>> assuming signed integer arithmetic will never overflow since the results
>>>>>> are undefined/unspecified if it does?
>>>>>>
>>>>>
>>>>> There isn't any signed arithmetic going on here. Just two well defined
>>>>> type conversions.
>>>>>
>>>>> I cited how the spec and gcc explain how both conversions should be
>>>>> handled. Show me where I'm wrong here? It feels wrong isn't good
>>>>> enough.
>>>>
>>>> It is unspecified by the C standard, and there are many other compilers
>>>> than gcc.  Can you prove that none of them do something silly here?
>>>>
>>>
>>> It's not unspecified. It's implementation defined.
>>
>> The distinction is only relevant if dealing with exactly one
>> implementation, which we are not.  Quoting gcc simply isn't good
>> enough.  Besides, even gcc might change in the future.
>>
>
> Because signed integer representation (two's complement; ones's
> complement; sign magnitude) is implementation defined this also needs
> to to be implementation defined.
>
> See also:
>
> 6.5.4
>> Some operators (the unary operator ~, and the binary operators <<, >>, &, ^, 
>> and |,
>> collectively described as bitwise operators) are required to have operands 
>> that have
>> integer type. These operators yield values that depend on the internal 
>> representations of
>> integers, and have implementation-defined and undefined aspects for signed 
>> types.
>
> Yet we allow all of these
>

Feel free to propose a better solution.
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to