Ah, sorry for my stupid response before. This should have been fixed in r2265. Thanks for such detail report. Regards, nanang On Sat, Sep 6, 2008 at 8:58 PM, ??? <wsycqyz at 126.com> wrote: > Hi, > I think I found why the assertion failed. > The problem is about that I set attr->info.max_bps to the rounded-up 2489. > In stream.c, create_channel() is called to create dec & enc channel. When it > comes to line 1402 which is in create_channel() of stream.c, > > channel->out_pkt_size = sizeof(pjmedia_rtp_hdr) + > stream->codec_param.info.max_bps * > PJMEDIA_MAX_FRAME_DURATION_MS / > 8 / 1000; > > channel->out_pkt_size will be 74 because max_bps is only 2489. > So when appending SDES at line 1333 of stream.c, > > len = create_rtcp_sdes(stream, (pj_uint8_t*)pkt, > stream->enc->out_pkt_size - len); > > We see out_pkt_size - len = 74 - 52 = 22 < 28. So the assertion failed. > > My solution is either to increase the max_bps to 3000 or to redefine the > PJMEDIA_MAX_FRAME_DURATION_MS to 300. > > Well, streamutil.c uses the pjmedia stream in a different way. When > pjmedia_rtcp_build_rtcp () is call, RTCP RR is always built and no SR, so > the returned variable "len" is always 32(sizeof(pjmedia_rtcp_rr_pkt)), not > 52(sizeof(pjmedia_rtcp_sr_pkt)). Finally out_pkt_size = 74 - 32 = 42 > 28. > The assertion won't fail. > > I attached my melp.c, melp.h, and the melp implement code if you need it. I > didn't do many tests so sorry for any bug. > For now, the implement uses float point. There is fixed point version but > the performance is really bad. > > Thank you again for giving me help! >
