Thanks for your attention, Adrian!
just one moment,
in hls_write_packet():
instead:
+ if (use_temp_file) {
+ hls_rename_temp_file(s, oc);
should be:
+ if (use_temp_file && !(hls->flags & HLS_SINGLE_FILE))
+ hls_rename_temp_file(s, oc);
or we a killing 'single' filename in few iterations.
Steven,
i recall my version of patch. Adrian's version fixes more issues and hope v3
will be the best :)
Aleksey Skripka
> On 18 Dec 2018, at 01:30, Adrian <[email protected]> wrote:
>
> Thanks for the explanation, now your intent is clear. I agree that keeping
> playlist file always as temp as long as it's a file protocol makes sense and
> won't do harm, though I'm still curious why original change to accept
> HLS_TEMP_FILE flag was sent to include and was included in official release,
> I believe there must be a reason.
>
> Given that, I've prepared another patch with fix for original problem +
> proper file picked up for computing 'use_temp_file' variable value. I don't
> know what is the exact flow here, so I'll send it as another patch, previous
> can be discarded and hopefully this one won't raise any more objections,
> though I'd still appreciate a review.
>
> Thanks for your input Aleksey!
>
> Regards
> Adrian Guzowski
>
> W dniu 17.12.2018 o 22:35, Aleksey Skripka pisze:
>> Evening!
>>
>> First of all, about playlist writeout:
>> before 223d2bde22ce33dcbcb6f17f234b609cb98f1fb6 - playlist was always(!)
>> creating via .tmp file.
>> after 223d2bde22ce33dcbcb6f17f234b609cb98f1fb6 - playlists .tmp logic become
>> dependent on +temp_file flag.
>>
>> I suggest to return to original logic.
>> Non-atomic playlist writeout - is a disaster to just any hls player. This
>> way we will force everyone (except offline VOD) to start using +temp_file.
>> Always!
>> Also, if somebody relying on this - they will get broken streams just after
>> upgrade (It seems to me, that it can be a lot).
>>
>> So, here maintainer should decide how to be.
>>
>>
>> and about single_file:
>> if it will be decided to stay with current way (playlist's .tmp logic
>> controlled by +temp_file flag), you are right - temp_file+single_file case
>> become unreal.
>> if it will be decided to return original behaviour (playlist always via
>> .tmp), someone can choose how to writeout media file (despite i understand
>> temp+single is very-very rare case).
>>
>> Thanks!
> <0001-Fix-usage-of-temp_file-flag-in-hls_flags-option-2nd-attempt.patch>_______________________________________________
> ffmpeg-devel mailing list
> [email protected]
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
_______________________________________________
ffmpeg-devel mailing list
[email protected]
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
