FFmpeg源码：av_probe_input_format3函数和AVInputFormat结构体分析（FFmpeg源码5.0.3版本）

=================================================================
FFmpeg源码：av_probe_input_format3函数分析：
FFmpeg源码：av_probe_input_format3函数和AVInputFormat结构体分析（FFmpeg源码5.0.3版本）
7.0.1版本的FFmpeg源码中av_probe_input_format3函数和AVInputFormat结构体的改变
=================================================================



一、av_probe_input_format3函数的声明

av_probe_input_format3函数声明在FFmpeg源码（本文演示用的FFmpeg源码版本为5.0.3）的头文件libavformat/avformat.h中：

1. /**
2. * Guess the file format.
3. *
4. * @param is_opened Whether the file is already opened; determines whether
5. *                  demuxers with or without AVFMT_NOFILE are probed.
6. * @param score_ret The score of the best detection.
7. */
8. const AVInputFormat *av_probe_input_format3(const AVProbeData *pd,
9.                                             int is_opened, int *score_ret);

复制代码

该函数在avformat_open_input函数内部被调用。其作用是：推测文件的格式。


形参pd：输入型参数，为AVProbeData范例的指针。
AVProbeData结构体声明在libavformat/avformat.h中：

1. /**
2. * This structure contains the data a format has to probe a file.
3. */
4. typedef struct AVProbeData {
5.     const char *filename;
6.     unsigned char *buf; /**< Buffer must have AVPROBE_PADDING_SIZE of extra allocated bytes filled with zero. */
7.     int buf_size;       /**< Size of buf except extra allocated bytes */
8.     const char *mime_type; /**< mime_type, when known. */
9. } AVProbeData;

复制代码

pd->filename为：须要被推测格式的文件的路径。
pd->buf：假如路径为pd->filename的文件还没被打开，pd->buf的值为NULL。假如路径为pd->filename的文件已经被打开，pd->buf指向“存放从该文件读取出来的二进制数据”的缓冲区。
pd->buf_size：缓冲区pd->buf的巨细，单位为字节。
pd->mime_type：一样平常为NULL，可忽略。

形参is_opened：输入型参数。值为0表现路径为pd->filename的文件还没被打开，值为1表现路径为pd->filename的文件已经被打开了。


形参score_ret：输出型参数，指向一个int型变量。实行av_probe_input_format3函数后，score_ret指向的int型变量的值会变为：最匹配“路径为pd->filename的文件”的格式的分值。
分值的范围为0到100分，值越大表现越匹配。分值对应的宏界说在头文件libavformat/avformat.h中被界说：

1. #define AVPROBE_SCORE_EXTENSION  50 ///< score for file extension
2. #define AVPROBE_SCORE_MIME       75 ///< score for file mime type
3. #define AVPROBE_SCORE_MAX       100 ///< maximum score

复制代码

返回值：为AVInputFormat范例的指针。假如形参is_opened的值为0（文件还没被打开），返回值为NULL；假如is_opened的值为1（文件已经被打开了），则返回最匹配“路径为pd->filename的文件”的格式的AVInputFormat结构体指针。
AVInputFormat结构体声明在libavformat/avformat.h中：

1. /**
2. * @addtogroup lavf_decoding
3. * @{
4. */
5. typedef struct AVInputFormat {
6.     /**
7.      * A comma separated list of short names for the format. New names
8.      * may be appended with a minor bump.
9.      */
10.     const char *name;
12.     /**
13.      * Descriptive name for the format, meant to be more human-readable
14.      * than name. You should use the NULL_IF_CONFIG_SMALL() macro
15.      * to define it.
16.      */
17.     const char *long_name;
19.     /**
20.      * Can use flags: AVFMT_NOFILE, AVFMT_NEEDNUMBER, AVFMT_SHOW_IDS,
21.      * AVFMT_NOTIMESTAMPS, AVFMT_GENERIC_INDEX, AVFMT_TS_DISCONT, AVFMT_NOBINSEARCH,
22.      * AVFMT_NOGENSEARCH, AVFMT_NO_BYTE_SEEK, AVFMT_SEEK_TO_PTS.
23.      */
24.     int flags;
26.     /**
27.      * If extensions are defined, then no probe is done. You should
28.      * usually not use extension format guessing because it is not
29.      * reliable enough
30.      */
31.     const char *extensions;
33.     const struct AVCodecTag * const *codec_tag;
35.     const AVClass *priv_class; ///< AVClass for the private context
37.     /**
38.      * Comma-separated list of mime types.
39.      * It is used check for matching mime types while probing.
40.      * @see av_probe_input_format2
41.      */
42.     const char *mime_type;
44.     /*****************************************************************
45.      * No fields below this line are part of the public API. They
46.      * may not be used outside of libavformat and can be changed and
47.      * removed at will.
48.      * New public fields should be added right above.
49.      *****************************************************************
50.      */
51.     /**
52.      * Raw demuxers store their codec ID here.
53.      */
54.     int raw_codec_id;
56.     /**
57.      * Size of private data so that it can be allocated in the wrapper.
58.      */
59.     int priv_data_size;
61.     /**
62.      * Internal flags. See FF_FMT_FLAG_* in internal.h.
63.      */
64.     int flags_internal;
66.     /**
67.      * Tell if a given file has a chance of being parsed as this format.
68.      * The buffer provided is guaranteed to be AVPROBE_PADDING_SIZE bytes
69.      * big so you do not have to check for that unless you need more.
70.      */
71.     int (*read_probe)(const AVProbeData *);
73.     /**
74.      * Read the format header and initialize the AVFormatContext
75.      * structure. Return 0 if OK. 'avformat_new_stream' should be
76.      * called to create new streams.
77.      */
78.     int (*read_header)(struct AVFormatContext *);
80.     /**
81.      * Read one packet and put it in 'pkt'. pts and flags are also
82.      * set. 'avformat_new_stream' can be called only if the flag
83.      * AVFMTCTX_NOHEADER is used and only in the calling thread (not in a
84.      * background thread).
85.      * @return 0 on success, < 0 on error.
86.      *         Upon returning an error, pkt must be unreferenced by the caller.
87.      */
88.     int (*read_packet)(struct AVFormatContext *, AVPacket *pkt);
90.     /**
91.      * Close the stream. The AVFormatContext and AVStreams are not
92.      * freed by this function
93.      */
94.     int (*read_close)(struct AVFormatContext *);
96.     /**
97.      * Seek to a given timestamp relative to the frames in
98.      * stream component stream_index.
99.      * @param stream_index Must not be -1.
100.      * @param flags Selects which direction should be preferred if no exact
101.      *              match is available.
102.      * @return >= 0 on success (but not necessarily the new offset)
103.      */
104.     int (*read_seek)(struct AVFormatContext *,
105.                      int stream_index, int64_t timestamp, int flags);
107.     /**
108.      * Get the next timestamp in stream[stream_index].time_base units.
109.      * @return the timestamp or AV_NOPTS_VALUE if an error occurred
110.      */
111.     int64_t (*read_timestamp)(struct AVFormatContext *s, int stream_index,
112.                               int64_t *pos, int64_t pos_limit);
114.     /**
115.      * Start/resume playing - only meaningful if using a network-based format
116.      * (RTSP).
117.      */
118.     int (*read_play)(struct AVFormatContext *);
120.     /**
121.      * Pause playing - only meaningful if using a network-based format
122.      * (RTSP).
123.      */
124.     int (*read_pause)(struct AVFormatContext *);
126.     /**
127.      * Seek to timestamp ts.
128.      * Seeking will be done so that the point from which all active streams
129.      * can be presented successfully will be closest to ts and within min/max_ts.
130.      * Active streams are all streams that have AVStream.discard < AVDISCARD_ALL.
131.      */
132.     int (*read_seek2)(struct AVFormatContext *s, int stream_index, int64_t min_ts, int64_t ts, int64_t max_ts, int flags);
134.     /**
135.      * Returns device list with it properties.
136.      * @see avdevice_list_devices() for more details.
137.      */
138.     int (*get_device_list)(struct AVFormatContext *s, struct AVDeviceInfoList *device_list);
140. } AVInputFormat;

复制代码

AVInputFormat是FFmpeg中的解复用器结构体，每种作为输入的封装格式(比方BMP、FLV、MP4、TS、H.264裸流文件等)都对应一种AVInputFormat 结构。

好比BMP这种图片的封装格式，它对应的AVInputFormat结构为：

1. const AVInputFormat ff_image_bmp_pipe_demuxer = {
2. .name           = "bmp_pipe",
3. .long_name      = "piped bmp sequence",
4. .priv_data_size = 1208,
5. .read_probe     = bmp_probe,
6. .read_header    = ff_img_read_header,
7. .read_packet    = ff_img_read_packet,
8. .priv_class     = &imagepipe_class,
9. .flags          = 256,
10. .raw_codec_id   = AV_CODEC_ID_BMP,
11. };

复制代码

H.264裸流文件对应的AVInputFormat结构为：

1. const AVInputFormat ff_h264_demuxer= {
2. .name           = "h264",
3. .long_name      = "raw H.264 video",
4. .priv_data_size = 1208,
5. .read_probe     = h264_probe,
6. .read_header    = ff_raw_video_read_header,
7. .read_packet    = ff_raw_read_partial_packet,
8. .priv_class     = &ff_rawvideo_demuxer_class,
9. .flags          = 256,
10. .raw_codec_id   = AV_CODEC_ID_H264,
11. };

复制代码

二、av_probe_input_format3函数的界说

av_probe_input_format3函数界说在源文件libavformat/format.c中：

1. const AVInputFormat *av_probe_input_format3(const AVProbeData *pd,
2.                                             int is_opened, int *score_ret)
3. {
4.     AVProbeData lpd = *pd;
5.     const AVInputFormat *fmt1 = NULL;
6.     const AVInputFormat *fmt = NULL;
7.     int score, score_max = 0;
8.     void *i = 0;
9.     const static uint8_t zerobuffer[AVPROBE_PADDING_SIZE];
10.     enum nodat {
11.         NO_ID3,
12.         ID3_ALMOST_GREATER_PROBE,
13.         ID3_GREATER_PROBE,
14.         ID3_GREATER_MAX_PROBE,
15.     } nodat = NO_ID3;
17.     if (!lpd.buf)
18.         lpd.buf = (unsigned char *) zerobuffer;
20.     if (lpd.buf_size > 10 && ff_id3v2_match(lpd.buf, ID3v2_DEFAULT_MAGIC)) {
21.         int id3len = ff_id3v2_tag_len(lpd.buf);
22.         if (lpd.buf_size > id3len + 16) {
23.             if (lpd.buf_size < 2LL*id3len + 16)
24.                 nodat = ID3_ALMOST_GREATER_PROBE;
25.             lpd.buf      += id3len;
26.             lpd.buf_size -= id3len;
27.         } else if (id3len >= PROBE_BUF_MAX) {
28.             nodat = ID3_GREATER_MAX_PROBE;
29.         } else
30.             nodat = ID3_GREATER_PROBE;
31.     }
33.     while ((fmt1 = av_demuxer_iterate(&i))) {
34.         if (fmt1->flags & AVFMT_EXPERIMENTAL)
35.             continue;
36.         if (!is_opened == !(fmt1->flags & AVFMT_NOFILE) && strcmp(fmt1->name, "image2"))
37.             continue;
38.         score = 0;
39.         if (fmt1->read_probe) {
40.             score = fmt1->read_probe(&lpd);
41.             if (score)
42.                 av_log(NULL, AV_LOG_TRACE, "Probing %s score:%d size:%d\n", fmt1->name, score, lpd.buf_size);
43.             if (fmt1->extensions && av_match_ext(lpd.filename, fmt1->extensions)) {
44.                 switch (nodat) {
45.                 case NO_ID3:
46.                     score = FFMAX(score, 1);
47.                     break;
48.                 case ID3_GREATER_PROBE:
49.                 case ID3_ALMOST_GREATER_PROBE:
50.                     score = FFMAX(score, AVPROBE_SCORE_EXTENSION / 2 - 1);
51.                     break;
52.                 case ID3_GREATER_MAX_PROBE:
53.                     score = FFMAX(score, AVPROBE_SCORE_EXTENSION);
54.                     break;
55.                 }
56.             }
57.         } else if (fmt1->extensions) {
58.             if (av_match_ext(lpd.filename, fmt1->extensions))
59.                 score = AVPROBE_SCORE_EXTENSION;
60.         }
61.         if (av_match_name(lpd.mime_type, fmt1->mime_type)) {
62.             if (AVPROBE_SCORE_MIME > score) {
63.                 av_log(NULL, AV_LOG_DEBUG, "Probing %s score:%d increased to %d due to MIME type\n", fmt1->name, score, AVPROBE_SCORE_MIME);
64.                 score = AVPROBE_SCORE_MIME;
65.             }
66.         }
67.         if (score > score_max) {
68.             score_max = score;
69.             fmt       = fmt1;
70.         } else if (score == score_max)
71.             fmt = NULL;
72.     }
73.     if (nodat == ID3_GREATER_PROBE)
74.         score_max = FFMIN(AVPROBE_SCORE_EXTENSION / 2 - 1, score_max);
75.     *score_ret = score_max;
77.     return fmt;
78. }

复制代码

三、av_probe_input_format3函数的内部实现分析

当形参is_opened的值为1，也就是须要被推测格式的文件已经被打开时。函数av_probe_input_format3中，起首会通过语句：

1. AVProbeData lpd = *pd;

复制代码

让lpd.buf指向“存放从该文件读取出来的二进制数据”的缓冲区，从而得到须要被推测格式的文件的二进制数据。


然后会循环实行：

1. while ((fmt1 = av_demuxer_iterate(&i))) {
2. //...
3. }

复制代码

av_demuxer_iterate函数界说在libavformat/allformats.c中，作用是得到形参opaque（也就是i）对应的已被注册的解封装器，可以看到它内部获取了全局数组demuxer_list中的元素：

1. const AVInputFormat *av_demuxer_iterate(void **opaque)
2. {
3.     static const uintptr_t size = sizeof(demuxer_list)/sizeof(demuxer_list[0]) - 1;
4.     uintptr_t i = (uintptr_t)*opaque;
5.     const AVInputFormat *f = NULL;
6.     uintptr_t tmp;
8.     if (i < size) {
9.         f = demuxer_list[i];
10.     } else if (tmp = atomic_load_explicit(&indev_list_intptr, memory_order_relaxed)) {
11.         const AVInputFormat *const *indev_list = (const AVInputFormat *const *)tmp;
12.         f = indev_list[i - size];
13.     }
15.     if (f)
16.         *opaque = (void*)(i + 1);
17.     return f;
18. }

复制代码

全局数组demuxer_list界说在libavformat/demuxer_list.c中，数组中的每个元素都对应一种AVInputFormat结构，也就是说数组中的每个元素都对应一种输入的封装格式(BMP、FLV、MP4、TS、H.264裸流文件等)：

1. static const AVInputFormat * const demuxer_list[] = {
2.     &ff_aa_demuxer,
3.     &ff_aac_demuxer,
4.     &ff_aax_demuxer,
5.     &ff_ac3_demuxer,
6.     &ff_ace_demuxer,
7.     &ff_acm_demuxer,
8.     &ff_act_demuxer,
9.     &ff_adf_demuxer,
10.     &ff_adp_demuxer,
11.     &ff_ads_demuxer,
12.     &ff_adx_demuxer,
13.     &ff_aea_demuxer,
14.     &ff_afc_demuxer,
15.     &ff_aiff_demuxer,
16.     &ff_aix_demuxer,
17.     &ff_alp_demuxer,
18.     &ff_amr_demuxer,
19.     &ff_amrnb_demuxer,
20.     &ff_amrwb_demuxer,
21.     &ff_anm_demuxer,
22.     &ff_apc_demuxer,
23.     &ff_ape_demuxer,
24.     &ff_apm_demuxer,
25.     &ff_apng_demuxer,
26.     &ff_aptx_demuxer,
27.     &ff_aptx_hd_demuxer,
28.     &ff_aqtitle_demuxer,
29.     &ff_argo_asf_demuxer,
30.     &ff_argo_brp_demuxer,
31.     &ff_argo_cvg_demuxer,
32.     &ff_asf_demuxer,
33.     &ff_asf_o_demuxer,
34.     &ff_ass_demuxer,
35.     &ff_ast_demuxer,
36.     &ff_au_demuxer,
37.     &ff_av1_demuxer,
38.     &ff_avi_demuxer,
39.     &ff_avr_demuxer,
40.     &ff_avs_demuxer,
41.     &ff_avs2_demuxer,
42.     &ff_avs3_demuxer,
43.     &ff_bethsoftvid_demuxer,
44.     &ff_bfi_demuxer,
45.     &ff_bintext_demuxer,
46.     &ff_bink_demuxer,
47.     &ff_binka_demuxer,
48.     &ff_bit_demuxer,
49.     &ff_bitpacked_demuxer,
50.     &ff_bmv_demuxer,
51.     &ff_bfstm_demuxer,
52.     &ff_brstm_demuxer,
53.     &ff_boa_demuxer,
54.     &ff_c93_demuxer,
55.     &ff_caf_demuxer,
56.     &ff_cavsvideo_demuxer,
57.     &ff_cdg_demuxer,
58.     &ff_cdxl_demuxer,
59.     &ff_cine_demuxer,
60.     &ff_codec2_demuxer,
61.     &ff_codec2raw_demuxer,
62.     &ff_concat_demuxer,
63.     &ff_data_demuxer,
64.     &ff_daud_demuxer,
65.     &ff_dcstr_demuxer,
66.     &ff_derf_demuxer,
67.     &ff_dfa_demuxer,
68.     &ff_dhav_demuxer,
69.     &ff_dirac_demuxer,
70.     &ff_dnxhd_demuxer,
71.     &ff_dsf_demuxer,
72.     &ff_dsicin_demuxer,
73.     &ff_dss_demuxer,
74.     &ff_dts_demuxer,
75.     &ff_dtshd_demuxer,
76.     &ff_dv_demuxer,
77.     &ff_dvbsub_demuxer,
78.     &ff_dvbtxt_demuxer,
79.     &ff_dxa_demuxer,
80.     &ff_ea_demuxer,
81.     &ff_ea_cdata_demuxer,
82.     &ff_eac3_demuxer,
83.     &ff_epaf_demuxer,
84.     &ff_ffmetadata_demuxer,
85.     &ff_filmstrip_demuxer,
86.     &ff_fits_demuxer,
87.     &ff_flac_demuxer,
88.     &ff_flic_demuxer,
89.     &ff_flv_demuxer,
90.     &ff_live_flv_demuxer,
91.     &ff_fourxm_demuxer,
92.     &ff_frm_demuxer,
93.     &ff_fsb_demuxer,
94.     &ff_fwse_demuxer,
95.     &ff_g722_demuxer,
96.     &ff_g723_1_demuxer,
97.     &ff_g726_demuxer,
98.     &ff_g726le_demuxer,
99.     &ff_g729_demuxer,
100.     &ff_gdv_demuxer,
101.     &ff_genh_demuxer,
102.     &ff_gif_demuxer,
103.     &ff_gsm_demuxer,
104.     &ff_gxf_demuxer,
105.     &ff_h261_demuxer,
106.     &ff_h263_demuxer,
107.     &ff_h264_demuxer,
108.     &ff_hca_demuxer,
109.     &ff_hcom_demuxer,
110.     &ff_hevc_demuxer,
111.     &ff_hls_demuxer,
112.     &ff_hnm_demuxer,
113.     &ff_ico_demuxer,
114.     &ff_idcin_demuxer,
115.     &ff_idf_demuxer,
116.     &ff_iff_demuxer,
117.     &ff_ifv_demuxer,
118.     &ff_ilbc_demuxer,
119.     &ff_image2_demuxer,
120.     &ff_image2pipe_demuxer,
121.     &ff_image2_alias_pix_demuxer,
122.     &ff_image2_brender_pix_demuxer,
123.     &ff_ingenient_demuxer,
124.     &ff_ipmovie_demuxer,
125.     &ff_ipu_demuxer,
126.     &ff_ircam_demuxer,
127.     &ff_iss_demuxer,
128.     &ff_iv8_demuxer,
129.     &ff_ivf_demuxer,
130.     &ff_ivr_demuxer,
131.     &ff_jacosub_demuxer,
132.     &ff_jv_demuxer,
133.     &ff_kux_demuxer,
134.     &ff_kvag_demuxer,
135.     &ff_lmlm4_demuxer,
136.     &ff_loas_demuxer,
137.     &ff_luodat_demuxer,
138.     &ff_lrc_demuxer,
139.     &ff_lvf_demuxer,
140.     &ff_lxf_demuxer,
141.     &ff_m4v_demuxer,
142.     &ff_mca_demuxer,
143.     &ff_mcc_demuxer,
144.     &ff_matroska_demuxer,
145.     &ff_mgsts_demuxer,
146.     &ff_microdvd_demuxer,
147.     &ff_mjpeg_demuxer,
148.     &ff_mjpeg_2000_demuxer,
149.     &ff_mlp_demuxer,
150.     &ff_mlv_demuxer,
151.     &ff_mm_demuxer,
152.     &ff_mmf_demuxer,
153.     &ff_mods_demuxer,
154.     &ff_moflex_demuxer,
155.     &ff_mov_demuxer,
156.     &ff_mp3_demuxer,
157.     &ff_mpc_demuxer,
158.     &ff_mpc8_demuxer,
159.     &ff_mpegps_demuxer,
160.     &ff_mpegts_demuxer,
161.     &ff_mpegtsraw_demuxer,
162.     &ff_mpegvideo_demuxer,
163.     &ff_mpjpeg_demuxer,
164.     &ff_mpl2_demuxer,
165.     &ff_mpsub_demuxer,
166.     &ff_msf_demuxer,
167.     &ff_msnwc_tcp_demuxer,
168.     &ff_msp_demuxer,
169.     &ff_mtaf_demuxer,
170.     &ff_mtv_demuxer,
171.     &ff_musx_demuxer,
172.     &ff_mv_demuxer,
173.     &ff_mvi_demuxer,
174.     &ff_mxf_demuxer,
175.     &ff_mxg_demuxer,
176.     &ff_nc_demuxer,
177.     &ff_nistsphere_demuxer,
178.     &ff_nsp_demuxer,
179.     &ff_nsv_demuxer,
180.     &ff_nut_demuxer,
181.     &ff_nuv_demuxer,
182.     &ff_obu_demuxer,
183.     &ff_ogg_demuxer,
184.     &ff_oma_demuxer,
185.     &ff_paf_demuxer,
186.     &ff_pcm_alaw_demuxer,
187.     &ff_pcm_mulaw_demuxer,
188.     &ff_pcm_vidc_demuxer,
189.     &ff_pcm_f64be_demuxer,
190.     &ff_pcm_f64le_demuxer,
191.     &ff_pcm_f32be_demuxer,
192.     &ff_pcm_f32le_demuxer,
193.     &ff_pcm_s32be_demuxer,
194.     &ff_pcm_s32le_demuxer,
195.     &ff_pcm_s24be_demuxer,
196.     &ff_pcm_s24le_demuxer,
197.     &ff_pcm_s16be_demuxer,
198.     &ff_pcm_s16le_demuxer,
199.     &ff_pcm_s8_demuxer,
200.     &ff_pcm_u32be_demuxer,
201.     &ff_pcm_u32le_demuxer,
202.     &ff_pcm_u24be_demuxer,
203.     &ff_pcm_u24le_demuxer,
204.     &ff_pcm_u16be_demuxer,
205.     &ff_pcm_u16le_demuxer,
206.     &ff_pcm_u8_demuxer,
207.     &ff_pjs_demuxer,
208.     &ff_pmp_demuxer,
209.     &ff_pp_bnk_demuxer,
210.     &ff_pva_demuxer,
211.     &ff_pvf_demuxer,
212.     &ff_qcp_demuxer,
213.     &ff_r3d_demuxer,
214.     &ff_rawvideo_demuxer,
215.     &ff_realtext_demuxer,
216.     &ff_redspark_demuxer,
217.     &ff_rl2_demuxer,
218.     &ff_rm_demuxer,
219.     &ff_roq_demuxer,
220.     &ff_rpl_demuxer,
221.     &ff_rsd_demuxer,
222.     &ff_rso_demuxer,
223.     &ff_rtp_demuxer,
224.     &ff_rtsp_demuxer,
225.     &ff_s337m_demuxer,
226.     &ff_sami_demuxer,
227.     &ff_sap_demuxer,
228.     &ff_sbc_demuxer,
229.     &ff_sbg_demuxer,
230.     &ff_scc_demuxer,
231.     &ff_scd_demuxer,
232.     &ff_sdp_demuxer,
233.     &ff_sdr2_demuxer,
234.     &ff_sds_demuxer,
235.     &ff_sdx_demuxer,
236.     &ff_segafilm_demuxer,
237.     &ff_ser_demuxer,
238.     &ff_sga_demuxer,
239.     &ff_shorten_demuxer,
240.     &ff_siff_demuxer,
241.     &ff_simbiosis_imx_demuxer,
242.     &ff_sln_demuxer,
243.     &ff_smacker_demuxer,
244.     &ff_smjpeg_demuxer,
245.     &ff_smush_demuxer,
246.     &ff_sol_demuxer,
247.     &ff_sox_demuxer,
248.     &ff_spdif_demuxer,
249.     &ff_srt_demuxer,
250.     &ff_str_demuxer,
251.     &ff_stl_demuxer,
252.     &ff_subviewer1_demuxer,
253.     &ff_subviewer_demuxer,
254.     &ff_sup_demuxer,
255.     &ff_svag_demuxer,
256.     &ff_svs_demuxer,
257.     &ff_swf_demuxer,
258.     &ff_tak_demuxer,
259.     &ff_tedcaptions_demuxer,
260.     &ff_thp_demuxer,
261.     &ff_threedostr_demuxer,
262.     &ff_tiertexseq_demuxer,
263.     &ff_tmv_demuxer,
264.     &ff_truehd_demuxer,
265.     &ff_tta_demuxer,
266.     &ff_txd_demuxer,
267.     &ff_tty_demuxer,
268.     &ff_ty_demuxer,
269.     &ff_v210_demuxer,
270.     &ff_v210x_demuxer,
271.     &ff_vag_demuxer,
272.     &ff_vc1_demuxer,
273.     &ff_vc1t_demuxer,
274.     &ff_vividas_demuxer,
275.     &ff_vivo_demuxer,
276.     &ff_vmd_demuxer,
277.     &ff_vobsub_demuxer,
278.     &ff_voc_demuxer,
279.     &ff_vpk_demuxer,
280.     &ff_vplayer_demuxer,
281.     &ff_vqf_demuxer,
282.     &ff_w64_demuxer,
283.     &ff_wav_demuxer,
284.     &ff_wc3_demuxer,
285.     &ff_webm_dash_manifest_demuxer,
286.     &ff_webvtt_demuxer,
287.     &ff_wsaud_demuxer,
288.     &ff_wsd_demuxer,
289.     &ff_wsvqa_demuxer,
290.     &ff_wtv_demuxer,
291.     &ff_wve_demuxer,
292.     &ff_wv_demuxer,
293.     &ff_xa_demuxer,
294.     &ff_xbin_demuxer,
295.     &ff_xmv_demuxer,
296.     &ff_xvag_demuxer,
297.     &ff_xwma_demuxer,
298.     &ff_yop_demuxer,
299.     &ff_yuv4mpegpipe_demuxer,
300.     &ff_image_bmp_pipe_demuxer,
301.     &ff_image_cri_pipe_demuxer,
302.     &ff_image_dds_pipe_demuxer,
303.     &ff_image_dpx_pipe_demuxer,
304.     &ff_image_exr_pipe_demuxer,
305.     &ff_image_gem_pipe_demuxer,
306.     &ff_image_gif_pipe_demuxer,
307.     &ff_image_j2k_pipe_demuxer,
308.     &ff_image_jpeg_pipe_demuxer,
309.     &ff_image_jpegls_pipe_demuxer,
310.     &ff_image_pam_pipe_demuxer,
311.     &ff_image_pbm_pipe_demuxer,
312.     &ff_image_pcx_pipe_demuxer,
313.     &ff_image_pgmyuv_pipe_demuxer,
314.     &ff_image_pgm_pipe_demuxer,
315.     &ff_image_pgx_pipe_demuxer,
316.     &ff_image_photocd_pipe_demuxer,
317.     &ff_image_pictor_pipe_demuxer,
318.     &ff_image_png_pipe_demuxer,
319.     &ff_image_ppm_pipe_demuxer,
320.     &ff_image_psd_pipe_demuxer,
321.     &ff_image_qdraw_pipe_demuxer,
322.     &ff_image_sgi_pipe_demuxer,
323.     &ff_image_svg_pipe_demuxer,
324.     &ff_image_sunrast_pipe_demuxer,
325.     &ff_image_tiff_pipe_demuxer,
326.     &ff_image_webp_pipe_demuxer,
327.     &ff_image_xbm_pipe_demuxer,
328.     &ff_image_xpm_pipe_demuxer,
329.     &ff_image_xwd_pipe_demuxer,
330.     NULL };

复制代码

以是下面语句现实上是循环得到全局数组demuxer_list中的下标为i的元素，也就是得到demuxer_list对应的AVInputFormat结构。实行该语句后fmt1指向i对应的AVInputFormat结构，i的值会加1。通过该语句可以把FFmpeg中全部被注册的文件格式都遍历一遍：

1. while ((fmt1 = av_demuxer_iterate(&i))) {
2. //...
3. }

复制代码

在上述循环中通过下面语句实行差异文件格式对应的分析函数，得到分值。read_probe是函数指针，实行fmt1->read_probe语句会调用差异文件格式对应的分析函数。好比对于BMP图片，会实行bmp_probe函数，对于H.264裸流文件，会实行h264_probe函数。实行完后会得到分值，表现跟该文件格式的匹配水平：

1. score = fmt1->read_probe(&lpd);

复制代码

循环分析全部格式的文件的过程中，不绝将最高分存贮到变量score_max中，让指针fmt指向最高分对应的AVInputFormat结构。从而在实行完全部差异文件格式对应的分析函数后，可以得到最匹配的文件格式：

1. if (score > score_max) {
2.             score_max = score;
3.             fmt       = fmt1;
4.         } else if (score == score_max)
5.             fmt = NULL;

复制代码

四、av_probe_input_format3函数的调用堆栈

av_probe_input_format3函数的调用堆栈为：
int avformat_open_input(AVFormatContext **ps, const char *filename,const AVInputFormat *fmt, AVDictionary **options) ->
static int init_input(AVFormatContext *s, const char *filename,AVDictionary **options) ->
int av_probe_input_buffer2(AVIOContext *pb, const AVInputFormat **fmt,const char *filename, void *logctx,unsigned int offset, unsigned int max_probe_size) ->
const AVInputFormat *av_probe_input_format2(const AVProbeData *pd,int is_opened, int *score_max) ->
const AVInputFormat *av_probe_input_format3(const AVProbeData *pd,int is_opened, int *score_ret)


以是avformat_open_input函数内部调用了init_input函数，init_input函数内部又调用了av_probe_input_buffer2函数，av_probe_input_buffer2函数内部调用av_probe_input_format2函数，av_probe_input_format2函数内部调用av_probe_input_format3函数。


由于：

1. static int init_input(AVFormatContext *s, const char *filename,
2.                       AVDictionary **options)
3. {
4. //...
5.     return av_probe_input_buffer2(s->pb, &s->iformat, filename,
6.                                   s, 0, s->format_probesize);
7. }

复制代码

1. int av_probe_input_buffer2(AVIOContext *pb, const AVInputFormat **fmt,
2.                            const char *filename, void *logctx,
3.                            unsigned int offset, unsigned int max_probe_size)
4. {
5. //...
6.         *fmt = av_probe_input_format2(&pd, 1, &score);
7. //...
8. }

复制代码

1. const AVInputFormat *av_probe_input_format2(const AVProbeData *pd,
2.                                             int is_opened, int *score_max)
3. {
4. //...
5.     const AVInputFormat *fmt = av_probe_input_format3(pd, is_opened, &score_ret);
6. //...
7.     return fmt;
9. }

复制代码

以是调用完av_probe_input_format3函数后，s->iformat（AVFormatContext的iformat成员变量）会得到最匹配该文件格式的AVInputFormat结构体指针。



五、总结

av_probe_input_format3函数是FFmpeg中实现容器格式检测的函数，其内部通过循环while ((fmt1 = av_demuxer_iterate(&i))) 拿到全部容器格式对应的AVInputFormat结构，然后通过score = fmt1->read_probe(&lpd)语句实行差异容器格式对应的分析函数，根据是否能被分析，以及匹配水平，来判断出这是哪种容器格式。