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
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
//! What travels between elements.
//!
//! [`MediaBuffer`] is an enum rather than one opaque buffer type, because
//! `ffmpeg-next` already hands back strongly-typed packets and frames and
//! collapsing them would only mean unwrapping again downstream. Its own
//! documentation covers why each payload is shared rather than copied, and
//! why a video frame arrives through a pool reference.
use Arc;
use ;
use crateUnboundObjectPoolRef;
/// The unit of data that flows between elements.
///
/// Compressed and uncompressed data are kept as distinct variants (rather
/// than a single opaque `Buffer` type like GStreamer) because ffmpeg-next
/// already gives us strongly-typed `Packet`/`Frame` types — collapsing them
/// into one type would just mean unwrapping again downstream.
///
/// Payloads are `Arc`-wrapped so `MediaBuffer` is cheaply `Clone` —
/// duplicating a buffer (e.g. [`crate::elements::Tee`] fanning packets out
/// to a decode branch and a remux branch) is a refcount bump, never a copy
/// of the encoded/decoded data.
///
/// `Video` specifically wraps an [`UnboundObjectPoolRef`], not a plain
/// `ffmpeg::frame::Video` — that's what lets whichever element produced it
/// (see [`crate::pool::UnboundObjectPool`], owned as that element's own
/// struct field) get the underlying buffer back automatically once every
/// `Arc` clone downstream has been dropped, instead of it just being freed.
/// Which buffer a video frame's pixels live in.
///
/// Not the frame: a producer with nothing new to show re-emits a fresh
/// `AVFrame` referencing the same picture every tick — a screen capture of a
/// still desktop does exactly that — so comparing frames, or the `Arc`s
/// around them, answers "changed" every time while the pixels have not
/// moved. The plane pointers do not.
///
/// The first two planes are enough for every layout this crate carries:
/// packed formats use one, and the semi-planar and planar ones this crate
/// composites in differ in the first two whenever they differ at all.
///
/// Only sound as an identity while the frame it came from is still
/// referenced. A picture whose buffer has been released can be handed out
/// again at the same address, so every caller here holds that reference for
/// as long as it holds the identity.
pub
/// Lets go of the picture a pooled wrapper was pointing at, as it returns to
/// its pool.
///
/// The `release` an [`UnboundObjectPool`](crate::pool::UnboundObjectPool) of
/// *wrappers* wants: an empty frame is given a picture with `av_frame_ref` on
/// every checkout, and without this it would keep that reference until its
/// next one. Nothing reads those pixels in the meantime, but everything that
/// asks whether a picture is still in use — [`picture_is_referenced`] — would
/// go on answering yes, so a producer would keep a buffer, or a screen-sized
/// texture, alive for an idle wrapper.
///
/// Only for a pool whose items own no picture of their own. A pool of real
/// frames composited or decoded into would be emptied by this.
pub
/// Whether anything besides this frame itself still points at its picture.
///
/// The companion to [`picture_id`], for an element that offers an unchanged
/// picture again by pointing an empty wrapper at it with `av_frame_ref`.
/// Such a wrapper shares the picture's *buffer*, not the pool slot the frame
/// came from, so an [`UnboundObjectPoolRef`] that has gone back to its pool
/// says nothing about whether a wrapper downstream is still showing those
/// pixels — the buffer's own reference count is the only record of it, and
/// a producer that recycles its frames has to keep one out of the pool until
/// this reads false for it.
///
/// Only the first buffer is examined, for the same reason [`picture_id`]
/// reads only the first plane pointers: a wrapper references either all of a
/// frame's buffers or none of them.
pub