On 07/10/2012 12:51 PM, Alex Converse wrote:
> On Tue, Jul 10, 2012 at 4:24 AM, Justin Ruggles
> <justin.rugg...@gmail.com> wrote:
>> On 07/09/2012 07:14 PM, Alex Converse wrote:
>>> On Mon, Jul 9, 2012 at 2:24 PM, Justin Ruggles <justin.rugg...@gmail.com>
>>> wrote:
>>>> allows for a larger sample size than MIN_CACHE_BITS
>>>> ---
>>>> libavcodec/alac.c | 9 ++-------
>>>> 1 files changed, 2 insertions(+), 7 deletions(-)
>>>>
>>>> diff --git a/libavcodec/alac.c b/libavcodec/alac.c
>>>> index 5f2c8a0..0e62adb 100644
>>>> --- a/libavcodec/alac.c
>>>> +++ b/libavcodec/alac.c
>>>> @@ -84,7 +84,7 @@ static inline int decode_scalar(GetBitContext *gb, int
>>>> k, int readsamplesize)
>>>>
>>>> if (x > 8) { /* RICE THRESHOLD */
>>>> /* use alternative encoding */
>>>> - x = get_bits(gb, readsamplesize);
>>>> + x = get_bits_long(gb, readsamplesize);
>>>> } else if (k != 1) {
>>>> int extrabits = show_bits(gb, k);
>>>>
>>>> @@ -294,6 +294,7 @@ static int alac_decode_frame(AVCodecContext *avctx,
>>>> void *data,
>>>> hassize = get_bits1(&alac->gb);
>>>>
>>>> alac->extra_bits = get_bits(&alac->gb, 2) << 3;
>>>> + readsamplesize = alac->sample_size - alac->extra_bits + channels - 1;
>>>>
>>>> /* whether the frame is compressed */
>>>> is_compressed = !get_bits1(&alac->gb);
>>>> @@ -321,12 +322,6 @@ static int alac_decode_frame(AVCodecContext *avctx,
>>>> void *data,
>>>> alac->output_samples_buffer[ch] = (int32_t
>>>> *)alac->frame.data[ch];
>>>> }
>>>>
>>>> - readsamplesize = alac->sample_size - alac->extra_bits + channels - 1;
>>>> - if (readsamplesize > MIN_CACHE_BITS) {
>>>> - av_log(avctx, AV_LOG_ERROR, "readsamplesize too big (%d)\n",
>>>> readsamplesize);
>>>> - return -1;
>>>> - }
>>>> -
>>>
>>> What about the case where sample_size = 32, channels = 6, and
>>> extra_bits is 0? It seems possible and results in readsamplesize = 37
>>> which is still too big.
>>
>> Yeah, I suppose we should check for that case. In practice it won't
>> happen though. For one thing, channels cannot be 6 here. It's 1 or 2
>> because it is per-element. The most it could be is 33. The reference
>> alac encoder uses 16 extra bits for 32-bit, so it's really only a
>> theoretical problem. I'll send an updated patch that handles the corner
>> cases.
>>
>
> Just to be clear, I'm not proposing you need to handle cases where
> it's more than 32 bits, only that you error out if the bitstream gives
> you that value.
Yeah, that's what I meant by handle. my latest patch errors out when
it's more than 32 bits.
-Justin
_______________________________________________
libav-devel mailing list
libav-devel@libav.org
https://lists.libav.org/mailman/listinfo/libav-devel
