[FFmpeg-devel] [PATCH] Fix usage of temp_file flag in hls_flags option
adrian-007 at o2.pl
Mon Dec 17 18:12:10 EET 2018
W dniu 17.12.2018 o 16:56, Aleksey Skripka pisze:
>> On 17 Dec 2018, at 18:45, Adrian <adrian-007 at o2.pl> wrote:
>> IMO temp_file+single_file should be mutually exclusive - why would you want to have temp file when you're using only one file anyway? In this mode you would either pool for file changes or just get updated playlist with new byte range (and that one should be updated after video file). I believe this restriction was already in place anyway, because there already were checks for HLS_SINGLE_FILE not being set, before doing actual rename, so this was just a simplification of those conditions.
> not me to decide, please ;)
> if playlist will always be done via .tmp, we will have a choice.
Could you clarify what choice you're talking about? And the purpose of
this choice - frankly you got me lost with this. Regardless, I feel
we're getting off topic.
>> As for playlist file - I'm not sure I follow, I've changed only places where temp files were already used, so I don't think I broke something in that matter.
> not you, it was introduced by 223d2bde22ce33dcbcb6f17f234b609cb98f1fb6
Not quite, temp files were already a feature, this commit narrowed
conditions where they are used. If you're referring to change in
hls_window about not checking TEMP flag - that is unrelated to current
state (before commit introducing regression, check was ignoring TEMP
flag, after commit it does take it into account - probably should be
discussed with person responsible for original change as to why it was
changed that way, so that's out of scope of this patch. Issue at hand is
not using TEMP files at all and that is fixed by this patch.
>> From repeat+level+trace I believe I saw that playlist file actually was using a temp file, so if that's the thing you are concerned about, it seems like it's all good.
>> Adrian Guzowski
>> W dniu 17.12.2018 o 16:36, Aleksey Skripka pisze:
>>> with your variant of patch in case of '-hls_flags +temp_file+single_file' no .tmp logic will be used.
>>> not sure what is better, but all previous versions was doing so.
>>> and also, i'm still sure, that for backward compatibility with all previous versions, index.m3u8 should always be written via .tmp file.
>>> i noticed this bug such way: i do not use +temp_file, but rely on atomic rename of index.m3u8.
>>>> On 17 Dec 2018, at 17:23, Adrian <adrian-007 at o2.pl> wrote:
>>>> after upgrading FFmpeg from 4.0 to 4.1 I noticed that temp files in HLS muxed stopped working.
>>>> It looks like a regression introduced by 223d2bde22ce33dcbcb6f17f234b609cb98f1fb6. I've prepared a patch and tested it cross-compiling for my project's target platform (ARM, Buildroot) and it seems like everything is in order. I ran regression tests and nothing seems to be broken.
>>>> Please review it and possibly include this patch.
>>>> Adrian Guzowski
>>>> ffmpeg-devel mailing list
>>>> ffmpeg-devel at ffmpeg.org
>> ffmpeg-devel mailing list
>> ffmpeg-devel at ffmpeg.org
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
More information about the ffmpeg-devel