[FFmpeg-devel] [PATCH 2/4] avformat/srtdec: UTF-16 support

[Clément Bœsch]

On Tue, Sep 02, 2014 at 08:56:10PM +0200, wm4 wrote:
> ---
>  libavformat/srtdec.c    | 22 +++++++++++----------
>  libavformat/subtitles.c | 51 +++++++++++++++++++++++++++++++++++++++++++++++--
>  libavformat/subtitles.h | 33 +++++++++++++++++++++++++++++---
>  3 files changed, 91 insertions(+), 15 deletions(-)
> 
> diff --git a/libavformat/srtdec.c b/libavformat/srtdec.c
> index 53182cd..a258cfe 100644
> --- a/libavformat/srtdec.c
> +++ b/libavformat/srtdec.c
> @@ -31,20 +31,20 @@ typedef struct {
>  
>  static int srt_probe(AVProbeData *p)
>  {
> -    const unsigned char *ptr = p->buf;
>      int i, v, num = 0;
> +    FFTextReader tr;
>  
> -    if (AV_RB24(ptr) == 0xEFBBBF)
> -        ptr += 3;  /* skip UTF-8 BOM */
> +    ff_text_init_buf(&tr, p->buf, p->buf_size);
>  
> -    while (*ptr == '\r' || *ptr == '\n')
> -        ptr++;
> +    while (ff_text_peek_r8(&tr) == '\r' || ff_text_peek_r8(&tr) == '\n')
> +        ff_text_r8(&tr);
>      for (i=0; i<2; i++) {
> +        char buf[128];
> +        ff_subtitles_read_line(&tr, buf, sizeof(buf));
>          if ((num == i || num + 1 == i)
> -            && sscanf(ptr, "%*d:%*2d:%*2d%*1[,.]%*3d --> %*d:%*2d:%*2d%*1[,.]%3d", &v) == 1)
> +            && sscanf(buf, "%*d:%*2d:%*2d%*1[,.]%*3d --> %*d:%*2d:%*2d%*1[,.]%3d", &v) == 1)
>              return AVPROBE_SCORE_MAX;
> -        num = atoi(ptr);
> -        ptr += ff_subtitles_next_line(ptr);
> +        num = atoi(buf);

So ff_subtitles_read_line() actually makes the probe too slow?

What if you try to:
  buf[0] >= '0' && buf[1] <= '9' && strstr(buf, " --> ") && sscanf(...)
to gain some little time by sometimes avoiding the sscanf call?

>      }
>      return 0;
>  }
> @@ -79,6 +79,8 @@ static int srt_read_header(AVFormatContext *s)
>      AVBPrint buf;
>      AVStream *st = avformat_new_stream(s, NULL);
>      int res = 0;
> +    FFTextReader tr;
> +    ff_text_init_avio(&tr, s->pb);
>  
>      if (!st)
>          return AVERROR(ENOMEM);
> @@ -88,8 +90,8 @@ static int srt_read_header(AVFormatContext *s)
>  
>      av_bprint_init(&buf, 0, AV_BPRINT_SIZE_UNLIMITED);
>  
> -    while (!avio_feof(s->pb)) {
> -        ff_subtitles_read_chunk(s->pb, &buf);
> +    while (!ff_text_eof(&tr)) {
> +        ff_subtitles_read_text_chunk(&tr, &buf);
>  
>          if (buf.len) {

>              int64_t pos = avio_tell(s->pb);

ff_text_pos()?

[...]
> +size_t ff_subtitles_read_line(FFTextReader *tr, char *buf, size_t size)
> +{
> +    size_t cur = 0;
> +    if (!size)
> +        return 0;
> +    while (cur + 1 < size) {
> +        unsigned char c = ff_text_r8(tr);
> +        if (!c && ff_text_eof(tr))
> +            return cur;
> +        if (c == '\r' || c == '\n')
> +            break;
> +        buf[cur++] = c;

> +        buf[cur] = '\0';

By reworking the logic, you might be able avoid double writing that end
character and gain some speed for the probing.

> +    }
> +    if (ff_text_peek_r8(tr) == '\r')
> +        ff_text_r8(tr);
> +    if (ff_text_peek_r8(tr) == '\n')
> +        ff_text_r8(tr);
> +    return cur;
> +}

LGTM otherwise

-- 
Clément B.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 473 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20140902/a1ee238b/attachment.asc>