1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
|
CONTEXT
=======
The FFmpeg project merges all the changes from the Libav project
(https://libav.org) since the origin of the fork (around 2011).
With the exceptions of some commits due to technical/political disagreements or
issues, the changes are merged on a more or less regular schedule (daily for
years thanks to Michael, but more sparse nowadays).
WHY
===
The majority of the active developers believe the project needs to keep this
policy for various reasons.
The most important one is that we don't want our users to have to choose
between two distributors of libraries of the exact same name in order to have a
different set of features and bugfixes. By taking the responsibility of
unifying the two codebases, we allow users to benefit from the changes from the
two teams.
Today, FFmpeg has a much larger user database (we are distributed by every
major distribution), so we consider this mission a priority.
A different approach to the merge could have been to pick the changes we are
interested in and drop most of the cosmetics and other less important changes.
Unfortunately, this makes the following picks much harder, especially since the
Libav project is involved in various deep API changes. As a result, we decide
to virtually take everything done there.
Any Libav developer is of course welcome anytime to contribute directly to the
FFmpeg tree. Of course, we fully understand and are forced to accept that very
few Libav developers are interested in doing so, but we still want to recognize
their work. This leads us to create merge commits for every single one from
Libav. The original commit appears totally unchanged with full authorship in
our history (and the conflict are solved in the merge one). That way, not a
single thing from Libav will be lost in the future in case some reunification
happens, or that project disappears one way or another.
DOWNSIDES
=========
Of course, there are many downsides to this approach.
- It causes a non negligible merge commits pollution. We make sure there are
not several level of merges entangled (we do a 1:1 merge/commit), but it's
still a non-linear history.
- Many duplicated work. For instance, we added libavresample in our tree to
keep compatibility with Libav when our libswresample was already covering the
exact same purpose. The same thing happened for various elements such as the
ProRes support (but differences in features, bugs, licenses, ...). There are
many work to do to unify them, and any help is very much welcome.
- So much manpower from both FFmpeg and Libav is lost because of this mess. We
know it, and we don't know how to fix it. It takes incredible time to do
these merges, so we have even less time to work on things we personally care
about. The bad vibes also do not help with keeping our developers motivated.
- There is a growing technical risk factor with the merges due to the codebase
differing more and more.
MERGE GUIDELINES
================
The following gives developer guidelines on how to proceed when merging Libav commits.
Before starting, you can reduce the risk of errors on merge conflicts by using
a different merge conflict style:
$ git config --global merge.conflictstyle diff3
tools/libav-merge-next-commit is a script to help merging the next commit in
the queue. It assumes a remote named libav. It has two modes: merge, and noop.
The noop mode creates a merge with no change to the HEAD. You can pass a hash
as extra argument to reference a justification (it is common that we already
have the change done in FFmpeg).
Also see tools/murge, you can copy and paste a 3 way conflict into its stdin
and it will display colored diffs. Any arguments to murge (like ones to suppress
whitespace differences) are passed into colordiff.
TODO/FIXME/UNMERGED
===================
Stuff that didn't reach the codebase:
-------------------------------------
- HEVC DSP and x86 MC SIMD improvements from Libav (see https://ffmpeg.org/pipermail/ffmpeg-devel/2015-December/184777.html)
- 1f821750f hevcdsp: split the qpel functions by width instead of by the subpixel fraction
- 818bfe7f0 hevcdsp: split the epel functions by width
- 688417399 hevcdsp: split the pred functions by width
- a853388d2 hevc: change the stride of the MC buffer to be in bytes instead of elements
- 0cef06df0 checkasm: add HEVC MC tests
- e7078e842 hevcdsp: add x86 SIMD for MC
- 7993ec19a hevc: Add hevc_get_pixel_4/8/12/16/24/32/48/64
- use av_cpu_max_align() instead of hardcoding alignment requirements (see https://ffmpeg.org/pipermail/ffmpeg-devel/2017-September/215834.html)
- f44ec22e0 lavc: use av_cpu_max_align() instead of hardcoding alignment requirements
- 4de220d2e frame: allow align=0 (meaning automatic) for av_frame_get_buffer()
- Support recovery from an already present HLS playlist (see 16cb06bb30)
- Remove all output devices (see 8e7e042d41, 8d3db95f20, 6ce13070bd, d46cd24986 and https://ffmpeg.org/pipermail/ffmpeg-devel/2017-September/216904.html)
- avcodec/libaomenc: export the Sequence Header OBU as extradata (See a024c3ce9a)
Collateral damage that needs work locally:
------------------------------------------
- Merge proresenc_anatoliy.c and proresenc_kostya.c
- Fix MIPS AC3 downmix
Extra changes needed to be aligned with Libav:
----------------------------------------------
- Switching our examples to the new encode/decode API (see 67d28f4a0f)
- HEVC IDCT bit depth 12-bit support (Libav added 8 and 10 but doesn't have 12)
|
