mpv/sub/sd_ass.c

1168 lines
39 KiB
C
Raw Normal View History

/*
* This file is part of mpv.
*
sd_ass: change license to LGPL All contributors of the code used for sd_ass.c agreed to the LGPL relicensing. Some code has a very chaotic history, due to MPlayer subtitle handling being awful, chaotic, and having been refactored a dozen of times. Most of the subtitle code was actually rewritten from scratch (a few times), and the initial sd_ass.c was pretty tiny. So we should be fine, but it's still a good idea to look at this closely. Potentially problematic cases of old code leaking into sd_ass.c are mentioned below. Some code originates from demux_mkv. Most of this was added by eugeni, and later moved into mplayer.c or mpcommon.c. The old demux_mkv ASS/SSA subtitle code is somewhat dangerous from a legal perspective, because it involves 2 patches by a certain Tristan/z80, who disagreed with LGPL, and who told us to "rewrite" parts we need. These patches were for converting the ASS packet data to the old MPlayer text subtitle data structures. None of that survived in the current code base. Moving the subtitle handling out of demux_mkv happened in the following commits: bdb6a07d2a712c, de73d4dd978cb2, 61e4a801913f76. The code by z80 was removed in b44202b69fc4a1. At this time, the z80 code was located in mplayer.c and subreader.c. The code was fully removed, being unnecessary due to the entire old subtitle rendering code being removed. This adds a ass_to_plaintext(), function, which replaces the old ASS tag stripping code in sub_add_text(), which was based on the z80 code. The new function was intended to strip ASS tags in a correct way, instead of somehow dealing with other subtitle types (like HTML-style SRT tags), so it was written from scratch. Another potential issue is the --sub-fix-timing option is based on -overlapsub added in d459e644632cb9b. But the implementation is new, and no code from that commit is used in sd_ass.c. The new implementation started out in 64b1374a4456435. (The following commit, bd45eb468ced22 removes the original code that was replaced.) The code was later moved into sd_ass.c. The --sub-fps option has a similar history.
2017-04-20 08:52:57 +00:00
* mpv is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* mpv is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
sd_ass: change license to LGPL All contributors of the code used for sd_ass.c agreed to the LGPL relicensing. Some code has a very chaotic history, due to MPlayer subtitle handling being awful, chaotic, and having been refactored a dozen of times. Most of the subtitle code was actually rewritten from scratch (a few times), and the initial sd_ass.c was pretty tiny. So we should be fine, but it's still a good idea to look at this closely. Potentially problematic cases of old code leaking into sd_ass.c are mentioned below. Some code originates from demux_mkv. Most of this was added by eugeni, and later moved into mplayer.c or mpcommon.c. The old demux_mkv ASS/SSA subtitle code is somewhat dangerous from a legal perspective, because it involves 2 patches by a certain Tristan/z80, who disagreed with LGPL, and who told us to "rewrite" parts we need. These patches were for converting the ASS packet data to the old MPlayer text subtitle data structures. None of that survived in the current code base. Moving the subtitle handling out of demux_mkv happened in the following commits: bdb6a07d2a712c, de73d4dd978cb2, 61e4a801913f76. The code by z80 was removed in b44202b69fc4a1. At this time, the z80 code was located in mplayer.c and subreader.c. The code was fully removed, being unnecessary due to the entire old subtitle rendering code being removed. This adds a ass_to_plaintext(), function, which replaces the old ASS tag stripping code in sub_add_text(), which was based on the z80 code. The new function was intended to strip ASS tags in a correct way, instead of somehow dealing with other subtitle types (like HTML-style SRT tags), so it was written from scratch. Another potential issue is the --sub-fix-timing option is based on -overlapsub added in d459e644632cb9b. But the implementation is new, and no code from that commit is used in sd_ass.c. The new implementation started out in 64b1374a4456435. (The following commit, bd45eb468ced22 removes the original code that was replaced.) The code was later moved into sd_ass.c. The --sub-fps option has a similar history.
2017-04-20 08:52:57 +00:00
* GNU Lesser General Public License for more details.
*
sd_ass: change license to LGPL All contributors of the code used for sd_ass.c agreed to the LGPL relicensing. Some code has a very chaotic history, due to MPlayer subtitle handling being awful, chaotic, and having been refactored a dozen of times. Most of the subtitle code was actually rewritten from scratch (a few times), and the initial sd_ass.c was pretty tiny. So we should be fine, but it's still a good idea to look at this closely. Potentially problematic cases of old code leaking into sd_ass.c are mentioned below. Some code originates from demux_mkv. Most of this was added by eugeni, and later moved into mplayer.c or mpcommon.c. The old demux_mkv ASS/SSA subtitle code is somewhat dangerous from a legal perspective, because it involves 2 patches by a certain Tristan/z80, who disagreed with LGPL, and who told us to "rewrite" parts we need. These patches were for converting the ASS packet data to the old MPlayer text subtitle data structures. None of that survived in the current code base. Moving the subtitle handling out of demux_mkv happened in the following commits: bdb6a07d2a712c, de73d4dd978cb2, 61e4a801913f76. The code by z80 was removed in b44202b69fc4a1. At this time, the z80 code was located in mplayer.c and subreader.c. The code was fully removed, being unnecessary due to the entire old subtitle rendering code being removed. This adds a ass_to_plaintext(), function, which replaces the old ASS tag stripping code in sub_add_text(), which was based on the z80 code. The new function was intended to strip ASS tags in a correct way, instead of somehow dealing with other subtitle types (like HTML-style SRT tags), so it was written from scratch. Another potential issue is the --sub-fix-timing option is based on -overlapsub added in d459e644632cb9b. But the implementation is new, and no code from that commit is used in sd_ass.c. The new implementation started out in 64b1374a4456435. (The following commit, bd45eb468ced22 removes the original code that was replaced.) The code was later moved into sd_ass.c. The --sub-fps option has a similar history.
2017-04-20 08:52:57 +00:00
* You should have received a copy of the GNU Lesser General Public
* License along with mpv. If not, see <http://www.gnu.org/licenses/>.
*/
#include <stdlib.h>
#include <assert.h>
#include <string.h>
#include <math.h>
#include <limits.h>
#include <libavutil/common.h>
#include <ass/ass.h>
#include "mpv_talloc.h"
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
#include "config.h"
#include "options/m_config.h"
#include "options/options.h"
#include "common/common.h"
#include "common/msg.h"
#include "demux/demux.h"
#include "video/csputils.h"
#include "video/mp_image.h"
#include "dec_sub.h"
#include "ass_mp.h"
#include "sd.h"
struct sd_ass_priv {
struct ass_library *ass_library;
struct ass_renderer *ass_renderer;
struct ass_track *ass_track;
struct ass_track *shadow_track; // for --sub-ass=no rendering
bool ass_configured;
bool is_converted;
struct lavc_conv *converter;
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
struct sd_filter **filters;
int num_filters;
bool clear_once;
struct mp_ass_packer *packer;
video: make OSD/subtitle bitmaps refcounted (sort of) Making OSD/subtitle bitmaps refcounted was planend a longer time ago, e.g. the sub_bitmaps.packed field (which refcounts the subtitle bitmap data) was added in 2016. But nothing benefited much from it, because struct sub_bitmaps was usually stack allocated, and there was this weird callback stuff through osd_draw(). Make it possible to get actually refcounted subtitle bitmaps on the OSD API level. For this, we just copy all subtitle data other than the bitmaps with sub_bitmaps_copy(). At first, I had planned some fancy refcount shit, but when that was a big mess and hard to debug and just boiled to emulating malloc(), I made it a full allocation+copy. This affects mostly the parts array. With crazy ASS subtitles, this parts array can get pretty big (thousands of elements or more), in which case the extra alloc/copy could become performance relevant. But then again this is just pure bullshit, and I see no need to care. In practice, this extra work most likely gets drowned out by libass murdering a single core (while mpv is waiting for it) anyway. So fuck it. I just wanted this so draw_bmp.c requires only a single call to render everything. VOs also can benefit from this, because the weird callback shit isn't necessary anymore (simpler code), but I haven't done anything about it yet. In general I'd hope this will work towards simplifying the OSD layer, which is prerequisite for making actual further improvements. I haven't tested some cases such as the "overlay-add" command. Maybe it crashes now? Who knows, who cares. In addition, it might be worthwhile to reduce the code duplication between all the things that output subtitle bitmaps (with repacking, image allocation, etc.), but that's orthogonal.
2020-04-26 21:34:32 +00:00
struct sub_bitmap_copy_cache *copy_cache;
bstr last_text;
struct mp_image_params video_params;
struct mp_image_params last_params;
struct mp_osd_res osd;
struct seen_packet *seen_packets;
2015-12-17 23:51:57 +00:00
int num_seen_packets;
int *packets_animated;
int num_packets_animated;
bool check_animated;
};
struct seen_packet {
int64_t pos;
double pts;
};
static void mangle_colors(struct sd *sd, struct sub_bitmaps *parts);
static void fill_plaintext(struct sd *sd, double pts);
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
static const struct sd_filter_functions *const filters[] = {
// Note: list order defines filter order.
&sd_filter_sdh,
#if HAVE_POSIX
&sd_filter_regex,
#endif
#if HAVE_JAVASCRIPT
&sd_filter_jsre,
#endif
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
NULL,
};
// Add default styles, if the track does not have any styles yet.
// Apply style overrides if the user provides any.
static void mp_ass_add_default_styles(ASS_Track *track, struct mp_subtitle_opts *opts,
struct mp_subtitle_shared_opts *shared_opts, int order)
{
if (opts->ass_styles_file && shared_opts->ass_style_override[order])
ass_read_styles(track, opts->ass_styles_file, NULL);
if (track->n_styles == 0) {
if (!track->PlayResY) {
track->PlayResX = MP_ASS_FONT_PLAYRESX;
track->PlayResY = MP_ASS_FONT_PLAYRESY;
}
track->Kerning = true;
int sid = ass_alloc_style(track);
track->default_style = sid;
ASS_Style *style = track->styles + sid;
style->Name = strdup("Default");
mp_ass_set_style(style, track->PlayResY, opts->sub_style);
}
if (shared_opts->ass_style_override[order])
ass_process_force_style(track);
}
static const char *const font_mimetypes[] = {
"application/x-truetype-font",
"application/vnd.ms-opentype",
"application/x-font-ttf",
"application/x-font", // probably incorrect
"application/font-sfnt",
"font/collection",
"font/otf",
"font/sfnt",
"font/ttf",
NULL
};
static const char *const font_exts[] = {".ttf", ".ttc", ".otf", ".otc", NULL};
static bool attachment_is_font(struct mp_log *log, struct demux_attachment *f)
{
if (!f->name || !f->type || !f->data || !f->data_size)
return false;
for (int n = 0; font_mimetypes[n]; n++) {
if (strcmp(font_mimetypes[n], f->type) == 0)
return true;
}
// fallback: match against file extension
char *ext = strlen(f->name) > 4 ? f->name + strlen(f->name) - 4 : "";
for (int n = 0; font_exts[n]; n++) {
if (strcasecmp(ext, font_exts[n]) == 0) {
mp_warn(log, "Loading font attachment '%s' with MIME type %s. "
"Assuming this is a broken Matroska file, which was "
"muxed without setting a correct font MIME type.\n",
f->name, f->type);
return true;
}
}
return false;
}
static void add_subtitle_fonts(struct sd *sd)
{
struct sd_ass_priv *ctx = sd->priv;
struct mp_subtitle_opts *opts = sd->opts;
2016-04-30 12:26:08 +00:00
if (!opts->ass_enabled || !opts->use_embedded_fonts || !sd->attachments)
return;
for (int i = 0; i < sd->attachments->num_entries; i++) {
struct demux_attachment *f = &sd->attachments->entries[i];
2016-04-30 12:26:08 +00:00
if (attachment_is_font(sd->log, f))
ass_add_font(ctx->ass_library, f->name, f->data, f->data_size);
}
}
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
static void filters_destroy(struct sd *sd)
{
struct sd_ass_priv *ctx = sd->priv;
for (int n = 0; n < ctx->num_filters; n++) {
struct sd_filter *ft = ctx->filters[n];
if (ft->driver->uninit)
ft->driver->uninit(ft);
talloc_free(ft);
}
ctx->num_filters = 0;
}
static void filters_init(struct sd *sd)
{
struct sd_ass_priv *ctx = sd->priv;
filters_destroy(sd);
for (int n = 0; filters[n]; n++) {
struct sd_filter *ft = talloc_ptrtype(ctx, ft);
*ft = (struct sd_filter){
.global = sd->global,
.log = sd->log,
.opts = mp_get_config_group(ft, sd->global, &mp_sub_filter_opts),
.driver = filters[n],
.codec = "ass",
.event_format = talloc_strdup(ft, ctx->ass_track->event_format),
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
};
if (ft->driver->init(ft)) {
MP_TARRAY_APPEND(ctx, ctx->filters, ctx->num_filters, ft);
} else {
talloc_free(ft);
}
}
}
static void enable_output(struct sd *sd, bool enable)
{
struct sd_ass_priv *ctx = sd->priv;
if (enable == !!ctx->ass_renderer)
return;
if (ctx->ass_renderer) {
ass_renderer_done(ctx->ass_renderer);
ctx->ass_renderer = NULL;
} else {
ctx->ass_renderer = ass_renderer_init(ctx->ass_library);
mp_ass_configure_fonts(ctx->ass_renderer, sd->opts->sub_style,
sd->global, sd->log);
}
}
static void assobjects_init(struct sd *sd)
{
struct sd_ass_priv *ctx = sd->priv;
struct mp_subtitle_opts *opts = sd->opts;
struct mp_subtitle_shared_opts *shared_opts = sd->shared_opts;
ctx->ass_library = mp_ass_init(sd->global, sd->opts->sub_style, sd->log);
ass_set_extract_fonts(ctx->ass_library, opts->use_embedded_fonts);
add_subtitle_fonts(sd);
if (shared_opts->ass_style_override[sd->order])
ass_set_style_overrides(ctx->ass_library, opts->ass_style_override_list);
ctx->ass_track = ass_new_track(ctx->ass_library);
ctx->ass_track->track_type = TRACK_TYPE_ASS;
ctx->shadow_track = ass_new_track(ctx->ass_library);
ctx->shadow_track->PlayResX = MP_ASS_FONT_PLAYRESX;
ctx->shadow_track->PlayResY = MP_ASS_FONT_PLAYRESY;
mp_ass_add_default_styles(ctx->shadow_track, opts, shared_opts, sd->order);
char *extradata = sd->codec->extradata;
int extradata_size = sd->codec->extradata_size;
if (ctx->converter) {
extradata = lavc_conv_get_extradata(ctx->converter);
extradata_size = extradata ? strlen(extradata) : 0;
}
if (extradata)
ass_process_codec_private(ctx->ass_track, extradata, extradata_size);
mp_ass_add_default_styles(ctx->ass_track, opts, shared_opts, sd->order);
#if LIBASS_VERSION >= 0x01302000
ass_set_check_readorder(ctx->ass_track, sd->opts->sub_clear_on_seek ? 0 : 1);
#endif
enable_output(sd, true);
}
static void assobjects_destroy(struct sd *sd)
{
struct sd_ass_priv *ctx = sd->priv;
ass_free_track(ctx->ass_track);
ass_free_track(ctx->shadow_track);
enable_output(sd, false);
ass_library_done(ctx->ass_library);
}
static int init(struct sd *sd)
{
struct sd_ass_priv *ctx = talloc_zero(sd, struct sd_ass_priv);
sd->priv = ctx;
// Note: accept "null" as alias for "ass", so EDL delay_open subtitle
// streams work.
if (strcmp(sd->codec->codec, "ass") != 0 &&
strcmp(sd->codec->codec, "null") != 0)
{
ctx->is_converted = true;
ctx->converter = lavc_conv_create(sd);
if (!ctx->converter)
return -1;
}
assobjects_init(sd);
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
filters_init(sd);
ctx->packer = mp_ass_packer_alloc(ctx);
2024-04-12 16:25:30 +00:00
// Subtitles does not have any profile value, so put the converted type as a profile.
const char **desc = ctx->converter ? &sd->codec->codec_profile : &sd->codec->codec_desc;
switch (ctx->ass_track->track_type) {
case TRACK_TYPE_ASS:
*desc = "Advanced Sub Station Alpha";
break;
case TRACK_TYPE_SSA:
*desc = "Sub Station Alpha";
break;
}
return 0;
}
// Check if subtitle has events that would cause it to be animated inside {}
static bool is_animated(const char *str)
{
const char *begin = str;
while ((str = strchr(str, '{'))) {
if (str++ > begin && str[-2] == '\\')
continue;
const char *end = strchr(str, '}');
if (!end)
return false;
while ((str = memchr(str, '\\', end - str))) {
while (str[0] == '\\')
++str;
while (str[0] == ' ' || str[0] == '\t')
++str;
if (str[0] == 'k' || str[0] == 'K' || str[0] == 't' ||
(str[0] == 'f' && str[1] == 'a' && str[2] == 'd') ||
(str[0] == 'm' && str[1] == 'o' && str[2] == 'v' && str[3] == 'e'))
{
return true;
}
}
str = end + 1;
}
return false;
}
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
// Note: pkt is not necessarily a fully valid refcounted packet.
static void filter_and_add(struct sd *sd, struct demux_packet *pkt)
{
struct sd_ass_priv *ctx = sd->priv;
struct demux_packet *orig_pkt = pkt;
ASS_Track *track = ctx->ass_track;
int old_n_events = track->n_events;
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
for (int n = 0; n < ctx->num_filters; n++) {
struct sd_filter *ft = ctx->filters[n];
struct demux_packet *npkt = ft->driver->filter(ft, pkt);
if (pkt != npkt && pkt != orig_pkt)
talloc_free(pkt);
pkt = npkt;
if (!pkt)
return;
}
ass_process_chunk(ctx->ass_track, pkt->buffer, pkt->len,
llrint(pkt->pts * 1000),
llrint(pkt->duration * 1000));
// This bookkeeping only has any practical use for ASS subs
// over a VO with no video.
if (!ctx->is_converted) {
if (!pkt->seen) {
for (int n = track->n_events - 1; n >= 0; n--) {
if (n + 1 == old_n_events || pkt->animated == 1)
break;
ASS_Event *event = &track->events[n];
// Might as well mark pkt->animated here with effects if we can.
pkt->animated = (event->Effect && event->Effect[0]) ? 1 : -1;
if (ctx->check_animated && pkt->animated != 1)
pkt->animated = is_animated(event->Text);
}
MP_TARRAY_APPEND(ctx, ctx->packets_animated, ctx->num_packets_animated, pkt->animated);
} else {
if (ctx->check_animated && ctx->packets_animated[pkt->seen_pos] == -1) {
for (int n = track->n_events - 1; n >= 0; n--) {
if (n + 1 == old_n_events || pkt->animated == 1)
break;
ASS_Event *event = &track->events[n];
ctx->packets_animated[pkt->seen_pos] = is_animated(event->Text);
pkt->animated = ctx->packets_animated[pkt->seen_pos];
}
} else {
pkt->animated = ctx->packets_animated[pkt->seen_pos];
}
}
}
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
if (pkt != orig_pkt)
talloc_free(pkt);
}
// Test if the packet with the given file position and pts was already consumed.
// Return false if the packet is new (and add it to the internal list), and
// return true if it was already seen.
static bool check_packet_seen(struct sd *sd, struct demux_packet *packet)
2015-12-17 23:51:57 +00:00
{
struct sd_ass_priv *priv = sd->priv;
int a = 0;
int b = priv->num_seen_packets;
while (a < b) {
int mid = a + (b - a) / 2;
struct seen_packet *seen_packet = &priv->seen_packets[mid];
if (packet->pos == seen_packet->pos && packet->pts == seen_packet->pts) {
packet->seen_pos = mid;
2015-12-17 23:51:57 +00:00
return true;
}
if (packet->pos > seen_packet->pos ||
(packet->pos == seen_packet->pos && packet->pts > seen_packet->pts)) {
2015-12-17 23:51:57 +00:00
a = mid + 1;
} else {
b = mid;
}
}
packet->seen_pos = a;
MP_TARRAY_INSERT_AT(priv, priv->seen_packets, priv->num_seen_packets, a,
(struct seen_packet){packet->pos, packet->pts});
2015-12-17 23:51:57 +00:00
return false;
}
#define UNKNOWN_DURATION (INT_MAX / 1000)
static void decode(struct sd *sd, struct demux_packet *packet)
{
struct sd_ass_priv *ctx = sd->priv;
ASS_Track *track = ctx->ass_track;
packet->sub_duration = packet->duration;
if (ctx->converter) {
if (!sd->opts->sub_clear_on_seek && packet->pos >= 0 &&
check_packet_seen(sd, packet))
return;
double sub_pts = 0;
double sub_duration = 0;
char **r = lavc_conv_decode(ctx->converter, packet, &sub_pts,
&sub_duration);
if (sd->opts->sub_stretch_durations ||
packet->duration < 0 || sub_duration == UINT32_MAX) {
MP_VERBOSE(sd, "Subtitle with unknown duration.\n");
sub_duration = UNKNOWN_DURATION;
}
for (int n = 0; r && r[n]; n++) {
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
struct demux_packet pkt2 = {
.pts = sub_pts,
.duration = sub_duration,
.buffer = r[n],
.len = strlen(r[n]),
};
filter_and_add(sd, &pkt2);
}
if (sub_duration == UNKNOWN_DURATION) {
for (int n = track->n_events - 2; n >= 0; n--) {
if (track->events[n].Duration == UNKNOWN_DURATION * 1000) {
if (track->events[n].Start != track->events[n + 1].Start) {
track->events[n].Duration = track->events[n + 1].Start -
track->events[n].Start;
} else {
track->events[n].Duration = track->events[n + 1].Duration;
}
}
}
}
} else {
2015-12-17 23:51:57 +00:00
// Note that for this packet format, libass has an internal mechanism
// for discarding duplicate (already seen) packets but we check this
// anyways for our purposes for ASS subtitles.
packet->seen = check_packet_seen(sd, packet);
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
filter_and_add(sd, packet);
}
}
// Calculate the height used for scaling subtitle text size so --sub-scale-with-window
// can undo this scale and use frame size instead. The algorithm used is the following:
// - If use_margins is disabled, the text is scaled with the visual size of the video.
// - If use_margins is enabled, the text is scaled with the size of the video
// as if the video is resized to "fit" the size of the frame.
static float get_libass_scale_height(struct mp_osd_res *dim, bool use_margins)
{
float vidw = dim->w - (dim->ml + dim->mr);
float vidh = dim->h - (dim->mt + dim->mb);
if (!use_margins || vidw < 1.0)
return vidh;
else
return MPMIN(dim->h, dim->w / vidw * vidh);
}
static void configure_ass(struct sd *sd, struct mp_osd_res *dim,
bool converted, ASS_Track *track)
{
struct mp_subtitle_opts *opts = sd->opts;
struct mp_subtitle_shared_opts *shared_opts = sd->shared_opts;
struct sd_ass_priv *ctx = sd->priv;
ASS_Renderer *priv = ctx->ass_renderer;
ass_set_frame_size(priv, dim->w, dim->h);
ass_set_margins(priv, dim->mt, dim->mb, dim->ml, dim->mr);
bool set_use_margins = false;
float set_sub_pos = 0.0f;
float set_line_spacing = 0;
float set_font_scale = 1;
int set_hinting = 0;
bool set_scale_with_window = false;
bool set_scale_by_window = true;
bool total_override = false;
// With forced overrides, apply the --sub-* specific options
if (converted || shared_opts->ass_style_override[sd->order] == ASS_STYLE_OVERRIDE_FORCE) {
set_scale_with_window = opts->sub_scale_with_window;
set_use_margins = opts->sub_use_margins;
set_scale_by_window = opts->sub_scale_by_window;
total_override = true;
} else {
set_scale_with_window = opts->ass_scale_with_window;
set_use_margins = opts->ass_use_margins;
}
if (converted || shared_opts->ass_style_override[sd->order]) {
set_sub_pos = 100.0f - shared_opts->sub_pos[sd->order];
set_line_spacing = opts->ass_line_spacing;
set_hinting = opts->ass_hinting;
}
if (total_override || shared_opts->ass_style_override[sd->order] == ASS_STYLE_OVERRIDE_SCALE) {
set_font_scale = opts->sub_scale;
}
if (set_scale_with_window) {
set_font_scale *= dim->h / MPMAX(get_libass_scale_height(dim, set_use_margins), 1);
}
if (!set_scale_by_window) {
double factor = dim->h / 720.0;
if (factor != 0.0)
set_font_scale /= factor;
}
ass_set_use_margins(priv, set_use_margins);
ass_set_line_position(priv, set_sub_pos);
ass_set_shaper(priv, opts->ass_shaper);
int set_force_flags = 0;
if (total_override) {
set_force_flags |= ASS_OVERRIDE_BIT_FONT_NAME
| ASS_OVERRIDE_BIT_FONT_SIZE_FIELDS
| ASS_OVERRIDE_BIT_COLORS
| ASS_OVERRIDE_BIT_BORDER
| ASS_OVERRIDE_BIT_SELECTIVE_FONT_SCALE;
}
if (shared_opts->ass_style_override[sd->order] == ASS_STYLE_OVERRIDE_SCALE)
set_force_flags |= ASS_OVERRIDE_BIT_SELECTIVE_FONT_SCALE;
if (converted)
set_force_flags |= ASS_OVERRIDE_BIT_ALIGNMENT;
#if LIBASS_VERSION >= 0x01306000
if ((converted || shared_opts->ass_style_override[sd->order]) && opts->ass_justify)
set_force_flags |= ASS_OVERRIDE_BIT_JUSTIFY;
#endif
ass_set_selective_style_override_enabled(priv, set_force_flags);
ASS_Style style = {0};
mp_ass_set_style(&style, MP_ASS_FONT_PLAYRESY, opts->sub_style);
ass_set_selective_style_override(priv, &style);
free(style.FontName);
if (converted && track->default_style < track->n_styles) {
mp_ass_set_style(track->styles + track->default_style,
track->PlayResY, opts->sub_style);
}
ass_set_font_scale(priv, set_font_scale);
ass_set_hinting(priv, set_hinting);
ass_set_line_spacing(priv, set_line_spacing);
#if LIBASS_VERSION >= 0x01600010
if (converted) {
ass_track_set_feature(track, ASS_FEATURE_WRAP_UNICODE, 1);
if (!opts->sub_vsfilter_bidi_compat) {
for (int n = 0; n < track->n_styles; n++) {
track->styles[n].Encoding = -1;
}
ass_track_set_feature(track, ASS_FEATURE_BIDI_BRACKETS, 1);
ass_track_set_feature(track, ASS_FEATURE_WHOLE_TEXT_LAYOUT, 1);
}
}
#endif
sd_ass: fix converted subtitles having too-wide borders Libass commit f08f8ea5 (between 0.16 and 0.17) changed how PlayResX affects some aspects of rendering. The libass change fixes a VSFilter compatibility issue which existed for about two decades, and there are no libass plans to support the previous behavior, so ultimately we have to adjust the mpv code, and we can't guarantee to restore the old behavior in all cases. Starting at this commit, vector drawing coords, font spacing, border and shadow widths are all affected by PlayResX (specifically, by the aspect), while previously they were unaffected by PlayResX. This changed converted sub border and shadow widths in mpv, because ffmpeg generates the ass with fixed PlayResX of 384 (aspect of 4:3), and with libass 0.17, if this doesn't match the display aspect, then borders and shadow were too wide - because most clips aspect is more than 4:3. The fact that ffmpeg uses fixed PlayResX of 384 could be considered an issue, but for now we have no control over it, and ffmpeg doesn't have the video resolution when it converts an srt source to ass. So here we adjust PlayResX accordingly so that border/shadows are now rendered with correct width. However, regardless of that commit, changing PlayResX also affects the margin value, so to compensate, we adjust sub-margins-x too. According to libass devs, this should cover basic srt-to-ass conversion by ffmpeg to work correctly with libass 0.17. However, there could be srt extensions which use more complex ass, and/or ffmpeg conversion of other sub formats (such as aribb24, aribcaption and movtext), where more things need adjustments. As of now we don't know what these are, and so we don't really know what else might remain broken or get broken.
2023-06-25 00:40:25 +00:00
if (converted) {
bool override_playres = true;
char **ass_style_override_list = opts->ass_style_override_list;
for (int i = 0; ass_style_override_list && ass_style_override_list[i]; i++) {
if (bstr_find0(bstr0(ass_style_override_list[i]), "PlayResX") >= 0)
sd_ass: fix converted subtitles having too-wide borders Libass commit f08f8ea5 (between 0.16 and 0.17) changed how PlayResX affects some aspects of rendering. The libass change fixes a VSFilter compatibility issue which existed for about two decades, and there are no libass plans to support the previous behavior, so ultimately we have to adjust the mpv code, and we can't guarantee to restore the old behavior in all cases. Starting at this commit, vector drawing coords, font spacing, border and shadow widths are all affected by PlayResX (specifically, by the aspect), while previously they were unaffected by PlayResX. This changed converted sub border and shadow widths in mpv, because ffmpeg generates the ass with fixed PlayResX of 384 (aspect of 4:3), and with libass 0.17, if this doesn't match the display aspect, then borders and shadow were too wide - because most clips aspect is more than 4:3. The fact that ffmpeg uses fixed PlayResX of 384 could be considered an issue, but for now we have no control over it, and ffmpeg doesn't have the video resolution when it converts an srt source to ass. So here we adjust PlayResX accordingly so that border/shadows are now rendered with correct width. However, regardless of that commit, changing PlayResX also affects the margin value, so to compensate, we adjust sub-margins-x too. According to libass devs, this should cover basic srt-to-ass conversion by ffmpeg to work correctly with libass 0.17. However, there could be srt extensions which use more complex ass, and/or ffmpeg conversion of other sub formats (such as aribb24, aribcaption and movtext), where more things need adjustments. As of now we don't know what these are, and so we don't really know what else might remain broken or get broken.
2023-06-25 00:40:25 +00:00
override_playres = false;
}
// srt to ass conversion from ffmpeg has fixed PlayResX of 384 with an
// aspect of 4:3. Starting with libass f08f8ea5 (pre 0.17) PlayResX
// affects shadow and border widths, among others, so to render borders
// and shadows correctly, we adjust PlayResX according to the DAR.
// But PlayResX also affects margins, so we adjust those too.
// This should ensure basic srt-to-ass ffmpeg conversion has correct
// borders, but there could be other issues with some srt extensions
// and/or different source formats which would be exposed over time.
// Make these adjustments only if the user didn't set PlayResX.
if (override_playres) {
int vidw = dim->w - (dim->ml + dim->mr);
int vidh = dim->h - (dim->mt + dim->mb);
int old_playresx = track->PlayResX;
sd_ass: fix converted subtitles having too-wide borders Libass commit f08f8ea5 (between 0.16 and 0.17) changed how PlayResX affects some aspects of rendering. The libass change fixes a VSFilter compatibility issue which existed for about two decades, and there are no libass plans to support the previous behavior, so ultimately we have to adjust the mpv code, and we can't guarantee to restore the old behavior in all cases. Starting at this commit, vector drawing coords, font spacing, border and shadow widths are all affected by PlayResX (specifically, by the aspect), while previously they were unaffected by PlayResX. This changed converted sub border and shadow widths in mpv, because ffmpeg generates the ass with fixed PlayResX of 384 (aspect of 4:3), and with libass 0.17, if this doesn't match the display aspect, then borders and shadow were too wide - because most clips aspect is more than 4:3. The fact that ffmpeg uses fixed PlayResX of 384 could be considered an issue, but for now we have no control over it, and ffmpeg doesn't have the video resolution when it converts an srt source to ass. So here we adjust PlayResX accordingly so that border/shadows are now rendered with correct width. However, regardless of that commit, changing PlayResX also affects the margin value, so to compensate, we adjust sub-margins-x too. According to libass devs, this should cover basic srt-to-ass conversion by ffmpeg to work correctly with libass 0.17. However, there could be srt extensions which use more complex ass, and/or ffmpeg conversion of other sub formats (such as aribb24, aribcaption and movtext), where more things need adjustments. As of now we don't know what these are, and so we don't really know what else might remain broken or get broken.
2023-06-25 00:40:25 +00:00
track->PlayResX = track->PlayResY * (double)vidw / MPMAX(vidh, 1);
double fix_margins = track->PlayResX / (double)old_playresx;
for (int n = 0; n < track->n_styles; n++) {
track->styles[n].MarginL = lrint(track->styles[n].MarginL * fix_margins);
track->styles[n].MarginR = lrint(track->styles[n].MarginR * fix_margins);
track->styles[n].MarginV = lrint(track->styles[n].MarginV * set_font_scale);
}
sd_ass: fix converted subtitles having too-wide borders Libass commit f08f8ea5 (between 0.16 and 0.17) changed how PlayResX affects some aspects of rendering. The libass change fixes a VSFilter compatibility issue which existed for about two decades, and there are no libass plans to support the previous behavior, so ultimately we have to adjust the mpv code, and we can't guarantee to restore the old behavior in all cases. Starting at this commit, vector drawing coords, font spacing, border and shadow widths are all affected by PlayResX (specifically, by the aspect), while previously they were unaffected by PlayResX. This changed converted sub border and shadow widths in mpv, because ffmpeg generates the ass with fixed PlayResX of 384 (aspect of 4:3), and with libass 0.17, if this doesn't match the display aspect, then borders and shadow were too wide - because most clips aspect is more than 4:3. The fact that ffmpeg uses fixed PlayResX of 384 could be considered an issue, but for now we have no control over it, and ffmpeg doesn't have the video resolution when it converts an srt source to ass. So here we adjust PlayResX accordingly so that border/shadows are now rendered with correct width. However, regardless of that commit, changing PlayResX also affects the margin value, so to compensate, we adjust sub-margins-x too. According to libass devs, this should cover basic srt-to-ass conversion by ffmpeg to work correctly with libass 0.17. However, there could be srt extensions which use more complex ass, and/or ffmpeg conversion of other sub formats (such as aribb24, aribcaption and movtext), where more things need adjustments. As of now we don't know what these are, and so we don't really know what else might remain broken or get broken.
2023-06-25 00:40:25 +00:00
}
}
}
static bool has_overrides(char *s)
{
if (!s)
return false;
return strstr(s, "\\pos") || strstr(s, "\\move") || strstr(s, "\\clip") ||
strstr(s, "\\iclip") || strstr(s, "\\org") || strstr(s, "\\p");
}
#define END(ev) ((ev)->Start + (ev)->Duration)
static long long find_timestamp(struct sd *sd, double pts)
{
struct sd_ass_priv *priv = sd->priv;
if (pts == MP_NOPTS_VALUE)
return 0;
long long ts = llrint(pts * 1000);
if (!sd->opts->sub_fix_timing ||
sd->shared_opts->ass_style_override[sd->order] == ASS_STYLE_OVERRIDE_NONE)
return ts;
// Try to fix small gaps and overlaps.
ASS_Track *track = priv->ass_track;
int threshold = SUB_GAP_THRESHOLD * 1000;
int keep = SUB_GAP_KEEP * 1000;
// Find the "current" event.
ASS_Event *ev[2] = {0};
int n_ev = 0;
for (int n = 0; n < track->n_events; n++) {
ASS_Event *event = &track->events[n];
if (ts >= event->Start - threshold && ts <= END(event) + threshold) {
if (n_ev >= MP_ARRAY_SIZE(ev))
return ts; // multiple overlaps - give up (probably complex subs)
ev[n_ev++] = event;
}
}
if (n_ev != 2)
return ts;
// Simple/minor heuristic against destroying typesetting.
if (ev[0]->Style != ev[1]->Style || has_overrides(ev[0]->Text) ||
has_overrides(ev[1]->Text))
return ts;
// Sort by start timestamps.
if (ev[0]->Start > ev[1]->Start)
MPSWAP(ASS_Event*, ev[0], ev[1]);
// We want to fix partial overlaps only.
if (END(ev[0]) >= END(ev[1]))
return ts;
if (ev[0]->Duration < keep || ev[1]->Duration < keep)
return ts;
// Gap between the events -> move ts to show the end of the first event.
if (ts >= END(ev[0]) && ts < ev[1]->Start && END(ev[0]) < ev[1]->Start &&
END(ev[0]) + threshold >= ev[1]->Start)
return END(ev[0]) - 1;
// Overlap -> move ts to the (exclusive) end of the first event.
// Relies on the fact that the ASS_Renderer has no overlap registered, even
// if there is one. This happens to work because we never render the
// overlapped state, and libass never resolves a collision.
if (ts >= ev[1]->Start && ts <= END(ev[0]) && END(ev[0]) > ev[1]->Start &&
END(ev[0]) <= ev[1]->Start + threshold)
return END(ev[0]);
return ts;
}
#undef END
video: make OSD/subtitle bitmaps refcounted (sort of) Making OSD/subtitle bitmaps refcounted was planend a longer time ago, e.g. the sub_bitmaps.packed field (which refcounts the subtitle bitmap data) was added in 2016. But nothing benefited much from it, because struct sub_bitmaps was usually stack allocated, and there was this weird callback stuff through osd_draw(). Make it possible to get actually refcounted subtitle bitmaps on the OSD API level. For this, we just copy all subtitle data other than the bitmaps with sub_bitmaps_copy(). At first, I had planned some fancy refcount shit, but when that was a big mess and hard to debug and just boiled to emulating malloc(), I made it a full allocation+copy. This affects mostly the parts array. With crazy ASS subtitles, this parts array can get pretty big (thousands of elements or more), in which case the extra alloc/copy could become performance relevant. But then again this is just pure bullshit, and I see no need to care. In practice, this extra work most likely gets drowned out by libass murdering a single core (while mpv is waiting for it) anyway. So fuck it. I just wanted this so draw_bmp.c requires only a single call to render everything. VOs also can benefit from this, because the weird callback shit isn't necessary anymore (simpler code), but I haven't done anything about it yet. In general I'd hope this will work towards simplifying the OSD layer, which is prerequisite for making actual further improvements. I haven't tested some cases such as the "overlay-add" command. Maybe it crashes now? Who knows, who cares. In addition, it might be worthwhile to reduce the code duplication between all the things that output subtitle bitmaps (with repacking, image allocation, etc.), but that's orthogonal.
2020-04-26 21:34:32 +00:00
static struct sub_bitmaps *get_bitmaps(struct sd *sd, struct mp_osd_res dim,
int format, double pts)
{
struct sd_ass_priv *ctx = sd->priv;
struct mp_subtitle_opts *opts = sd->opts;
struct mp_subtitle_shared_opts *shared_opts = sd->shared_opts;
bool no_ass = !opts->ass_enabled ||
shared_opts->ass_style_override[sd->order] == ASS_STYLE_OVERRIDE_STRIP;
bool converted = (ctx->is_converted && !lavc_conv_is_styled(ctx->converter)) || no_ass;
ASS_Track *track = no_ass ? ctx->shadow_track : ctx->ass_track;
ASS_Renderer *renderer = ctx->ass_renderer;
video: make OSD/subtitle bitmaps refcounted (sort of) Making OSD/subtitle bitmaps refcounted was planend a longer time ago, e.g. the sub_bitmaps.packed field (which refcounts the subtitle bitmap data) was added in 2016. But nothing benefited much from it, because struct sub_bitmaps was usually stack allocated, and there was this weird callback stuff through osd_draw(). Make it possible to get actually refcounted subtitle bitmaps on the OSD API level. For this, we just copy all subtitle data other than the bitmaps with sub_bitmaps_copy(). At first, I had planned some fancy refcount shit, but when that was a big mess and hard to debug and just boiled to emulating malloc(), I made it a full allocation+copy. This affects mostly the parts array. With crazy ASS subtitles, this parts array can get pretty big (thousands of elements or more), in which case the extra alloc/copy could become performance relevant. But then again this is just pure bullshit, and I see no need to care. In practice, this extra work most likely gets drowned out by libass murdering a single core (while mpv is waiting for it) anyway. So fuck it. I just wanted this so draw_bmp.c requires only a single call to render everything. VOs also can benefit from this, because the weird callback shit isn't necessary anymore (simpler code), but I haven't done anything about it yet. In general I'd hope this will work towards simplifying the OSD layer, which is prerequisite for making actual further improvements. I haven't tested some cases such as the "overlay-add" command. Maybe it crashes now? Who knows, who cares. In addition, it might be worthwhile to reduce the code duplication between all the things that output subtitle bitmaps (with repacking, image allocation, etc.), but that's orthogonal.
2020-04-26 21:34:32 +00:00
struct sub_bitmaps *res = &(struct sub_bitmaps){0};
// Always update the osd_res
struct mp_osd_res old_osd = ctx->osd;
ctx->osd = dim;
if (pts == MP_NOPTS_VALUE || !renderer)
video: make OSD/subtitle bitmaps refcounted (sort of) Making OSD/subtitle bitmaps refcounted was planend a longer time ago, e.g. the sub_bitmaps.packed field (which refcounts the subtitle bitmap data) was added in 2016. But nothing benefited much from it, because struct sub_bitmaps was usually stack allocated, and there was this weird callback stuff through osd_draw(). Make it possible to get actually refcounted subtitle bitmaps on the OSD API level. For this, we just copy all subtitle data other than the bitmaps with sub_bitmaps_copy(). At first, I had planned some fancy refcount shit, but when that was a big mess and hard to debug and just boiled to emulating malloc(), I made it a full allocation+copy. This affects mostly the parts array. With crazy ASS subtitles, this parts array can get pretty big (thousands of elements or more), in which case the extra alloc/copy could become performance relevant. But then again this is just pure bullshit, and I see no need to care. In practice, this extra work most likely gets drowned out by libass murdering a single core (while mpv is waiting for it) anyway. So fuck it. I just wanted this so draw_bmp.c requires only a single call to render everything. VOs also can benefit from this, because the weird callback shit isn't necessary anymore (simpler code), but I haven't done anything about it yet. In general I'd hope this will work towards simplifying the OSD layer, which is prerequisite for making actual further improvements. I haven't tested some cases such as the "overlay-add" command. Maybe it crashes now? Who knows, who cares. In addition, it might be worthwhile to reduce the code duplication between all the things that output subtitle bitmaps (with repacking, image allocation, etc.), but that's orthogonal.
2020-04-26 21:34:32 +00:00
goto done;
// Currently no supported text sub formats support a distinction between forced
// and unforced lines, so we just assume everything's unforced and discard everything.
// If we ever see a format that makes this distinction, we can add support here.
if (opts->sub_forced_events_only)
goto done;
double scale = dim.display_par;
if (!converted && (!shared_opts->ass_style_override[sd->order] ||
opts->ass_use_video_data >= 1))
{
// Let's factor in video PAR for vsfilter compatibility:
double par = opts->ass_video_aspect > 0 ?
opts->ass_video_aspect :
ctx->video_params.p_w / (double)ctx->video_params.p_h;
if (isnormal(par))
scale *= par;
}
if (!ctx->ass_configured || !osd_res_equals(old_osd, ctx->osd)) {
configure_ass(sd, &dim, converted, track);
ctx->ass_configured = true;
}
ass_set_pixel_aspect(renderer, scale);
if (!converted && (!shared_opts->ass_style_override[sd->order] ||
opts->ass_use_video_data >= 2))
{
ass_set_storage_size(renderer, ctx->video_params.w, ctx->video_params.h);
} else {
ass_set_storage_size(renderer, 0, 0);
}
long long ts = find_timestamp(sd, pts);
if (no_ass)
fill_plaintext(sd, pts);
int changed;
ASS_Image *imgs = ass_render_frame(renderer, track, ts, &changed);
mp_ass_packer_pack(ctx->packer, &imgs, 1, changed, !converted, format, res);
video: make OSD/subtitle bitmaps refcounted (sort of) Making OSD/subtitle bitmaps refcounted was planend a longer time ago, e.g. the sub_bitmaps.packed field (which refcounts the subtitle bitmap data) was added in 2016. But nothing benefited much from it, because struct sub_bitmaps was usually stack allocated, and there was this weird callback stuff through osd_draw(). Make it possible to get actually refcounted subtitle bitmaps on the OSD API level. For this, we just copy all subtitle data other than the bitmaps with sub_bitmaps_copy(). At first, I had planned some fancy refcount shit, but when that was a big mess and hard to debug and just boiled to emulating malloc(), I made it a full allocation+copy. This affects mostly the parts array. With crazy ASS subtitles, this parts array can get pretty big (thousands of elements or more), in which case the extra alloc/copy could become performance relevant. But then again this is just pure bullshit, and I see no need to care. In practice, this extra work most likely gets drowned out by libass murdering a single core (while mpv is waiting for it) anyway. So fuck it. I just wanted this so draw_bmp.c requires only a single call to render everything. VOs also can benefit from this, because the weird callback shit isn't necessary anymore (simpler code), but I haven't done anything about it yet. In general I'd hope this will work towards simplifying the OSD layer, which is prerequisite for making actual further improvements. I haven't tested some cases such as the "overlay-add" command. Maybe it crashes now? Who knows, who cares. In addition, it might be worthwhile to reduce the code duplication between all the things that output subtitle bitmaps (with repacking, image allocation, etc.), but that's orthogonal.
2020-04-26 21:34:32 +00:00
done:
// mangle_colors() modifies the color field, so copy the thing _before_.
res = sub_bitmaps_copy(&ctx->copy_cache, res);
video: make OSD/subtitle bitmaps refcounted (sort of) Making OSD/subtitle bitmaps refcounted was planend a longer time ago, e.g. the sub_bitmaps.packed field (which refcounts the subtitle bitmap data) was added in 2016. But nothing benefited much from it, because struct sub_bitmaps was usually stack allocated, and there was this weird callback stuff through osd_draw(). Make it possible to get actually refcounted subtitle bitmaps on the OSD API level. For this, we just copy all subtitle data other than the bitmaps with sub_bitmaps_copy(). At first, I had planned some fancy refcount shit, but when that was a big mess and hard to debug and just boiled to emulating malloc(), I made it a full allocation+copy. This affects mostly the parts array. With crazy ASS subtitles, this parts array can get pretty big (thousands of elements or more), in which case the extra alloc/copy could become performance relevant. But then again this is just pure bullshit, and I see no need to care. In practice, this extra work most likely gets drowned out by libass murdering a single core (while mpv is waiting for it) anyway. So fuck it. I just wanted this so draw_bmp.c requires only a single call to render everything. VOs also can benefit from this, because the weird callback shit isn't necessary anymore (simpler code), but I haven't done anything about it yet. In general I'd hope this will work towards simplifying the OSD layer, which is prerequisite for making actual further improvements. I haven't tested some cases such as the "overlay-add" command. Maybe it crashes now? Who knows, who cares. In addition, it might be worthwhile to reduce the code duplication between all the things that output subtitle bitmaps (with repacking, image allocation, etc.), but that's orthogonal.
2020-04-26 21:34:32 +00:00
if (!converted && res)
mangle_colors(sd, res);
video: make OSD/subtitle bitmaps refcounted (sort of) Making OSD/subtitle bitmaps refcounted was planend a longer time ago, e.g. the sub_bitmaps.packed field (which refcounts the subtitle bitmap data) was added in 2016. But nothing benefited much from it, because struct sub_bitmaps was usually stack allocated, and there was this weird callback stuff through osd_draw(). Make it possible to get actually refcounted subtitle bitmaps on the OSD API level. For this, we just copy all subtitle data other than the bitmaps with sub_bitmaps_copy(). At first, I had planned some fancy refcount shit, but when that was a big mess and hard to debug and just boiled to emulating malloc(), I made it a full allocation+copy. This affects mostly the parts array. With crazy ASS subtitles, this parts array can get pretty big (thousands of elements or more), in which case the extra alloc/copy could become performance relevant. But then again this is just pure bullshit, and I see no need to care. In practice, this extra work most likely gets drowned out by libass murdering a single core (while mpv is waiting for it) anyway. So fuck it. I just wanted this so draw_bmp.c requires only a single call to render everything. VOs also can benefit from this, because the weird callback shit isn't necessary anymore (simpler code), but I haven't done anything about it yet. In general I'd hope this will work towards simplifying the OSD layer, which is prerequisite for making actual further improvements. I haven't tested some cases such as the "overlay-add" command. Maybe it crashes now? Who knows, who cares. In addition, it might be worthwhile to reduce the code duplication between all the things that output subtitle bitmaps (with repacking, image allocation, etc.), but that's orthogonal.
2020-04-26 21:34:32 +00:00
return res;
}
#define MAX_BUF_SIZE 1024 * 1024
#define MIN_EXPAND_SIZE 4096
static void append(bstr *b, char c)
{
bstr_xappend(NULL, b, (bstr){&c, 1});
}
static void ass_to_plaintext(bstr *b, const char *in)
{
const char *open_tag_pos = NULL;
bool in_drawing = false;
while (*in) {
if (open_tag_pos) {
if (in[0] == '}') {
in += 1;
open_tag_pos = NULL;
} else if (in[0] == '\\' && in[1] == 'p' && in[2] != 'o') {
in += 2;
// Skip text between \pN and \p0 tags. A \p without a number
// is the same as \p0, and leading 0s are also allowed.
in_drawing = false;
while (in[0] >= '0' && in[0] <= '9') {
if (in[0] != '0')
in_drawing = true;
in += 1;
}
} else {
in += 1;
}
} else {
if (in[0] == '\\' && (in[1] == 'N' || in[1] == 'n')) {
in += 2;
append(b, '\n');
} else if (in[0] == '\\' && in[1] == 'h') {
in += 2;
append(b, ' ');
} else if (in[0] == '{') {
open_tag_pos = in;
in += 1;
} else {
if (!in_drawing)
append(b, in[0]);
in += 1;
}
}
}
// A '{' without a closing '}' is always visible.
if (open_tag_pos) {
bstr_xappend(NULL, b, bstr0(open_tag_pos));
}
}
// Empty string counts as whitespace.
static bool is_whitespace_only(bstr b)
{
for (int n = 0; n < b.len; n++) {
if (b.start[n] != ' ' && b.start[n] != '\t')
return false;
}
return true;
}
static bstr get_text_buf(struct sd *sd, double pts, enum sd_text_type type)
{
struct sd_ass_priv *ctx = sd->priv;
ASS_Track *track = ctx->ass_track;
if (pts == MP_NOPTS_VALUE)
return (bstr){0};
long long ipts = find_timestamp(sd, pts);
bstr *b = &ctx->last_text;
if (!b->start)
b->start = talloc_size(ctx, 4096);
b->len = 0;
for (int i = 0; i < track->n_events; ++i) {
ASS_Event *event = track->events + i;
if (ipts >= event->Start && ipts < event->Start + event->Duration) {
if (event->Text) {
int start = b->len;
if (type == SD_TEXT_TYPE_PLAIN) {
ass_to_plaintext(b, event->Text);
} else if (type == SD_TEXT_TYPE_ASS_FULL) {
long long s = event->Start;
long long e = s + event->Duration;
ASS_Style *style = (event->Style < 0 || event->Style >= track->n_styles) ? NULL : &track->styles[event->Style];
int sh = (s / 60 / 60 / 1000);
int sm = (s / 60 / 1000) % 60;
int ss = (s / 1000) % 60;
int sc = (s / 10) % 100;
int eh = (e / 60 / 60 / 1000);
int em = (e / 60 / 1000) % 60;
int es = (e / 1000) % 60;
int ec = (e / 10) % 100;
bstr_xappend_asprintf(NULL, b, "Dialogue: %d,%d:%02d:%02d.%02d,%d:%02d:%02d.%02d,%s,%s,%04d,%04d,%04d,%s,%s",
event->Layer,
sh, sm, ss, sc,
eh, em, es, ec,
(style && style->Name) ? style->Name : "", event->Name,
event->MarginL, event->MarginR, event->MarginV,
event->Effect, event->Text);
} else {
bstr_xappend(NULL, b, bstr0(event->Text));
}
if (is_whitespace_only(bstr_cut(*b, start))) {
b->len = start;
} else {
append(b, '\n');
}
}
}
}
bstr_eatend(b, (bstr)bstr0_lit("\n"));
return *b;
}
static char *get_text(struct sd *sd, double pts, enum sd_text_type type)
{
return bstrto0(NULL, get_text_buf(sd, pts, type));
}
static struct sd_times get_times(struct sd *sd, double pts)
{
struct sd_ass_priv *ctx = sd->priv;
ASS_Track *track = ctx->ass_track;
struct sd_times res = { .start = MP_NOPTS_VALUE, .end = MP_NOPTS_VALUE };
if (pts == MP_NOPTS_VALUE)
return res;
long long ipts = find_timestamp(sd, pts);
for (int i = 0; i < track->n_events; ++i) {
ASS_Event *event = track->events + i;
if (ipts >= event->Start && ipts < event->Start + event->Duration) {
double start = event->Start / 1000.0;
double end = event->Duration == UNKNOWN_DURATION ?
MP_NOPTS_VALUE : (event->Start + event->Duration) / 1000.0;
if (res.start == MP_NOPTS_VALUE || res.start > start)
res.start = start;
if (res.end == MP_NOPTS_VALUE || res.end < end)
res.end = end;
}
}
return res;
}
static void fill_plaintext(struct sd *sd, double pts)
{
struct sd_ass_priv *ctx = sd->priv;
ASS_Track *track = ctx->shadow_track;
ass_flush_events(track);
bstr text = get_text_buf(sd, pts, SD_TEXT_TYPE_PLAIN);
if (!text.len)
return;
bstr dst = {0};
while (text.len) {
if (*text.start == '{') {
bstr_xappend(NULL, &dst, bstr0("\\{"));
text = bstr_cut(text, 1);
} else if (*text.start == '\\') {
bstr_xappend(NULL, &dst, bstr0("\\"));
// Break ASS escapes with U+2060 WORD JOINER
mp_append_utf8_bstr(NULL, &dst, 0x2060);
text = bstr_cut(text, 1);
}
int i = bstrcspn(text, "{\\");
bstr_xappend(NULL, &dst, (bstr){text.start, i});
text = bstr_cut(text, i);
}
if (!dst.start)
return;
int n = ass_alloc_event(track);
ASS_Event *event = track->events + n;
event->Start = 0;
event->Duration = INT_MAX;
event->Style = track->default_style;
event->Text = strdup(dst.start);
talloc_free(dst.start);
}
static void reset(struct sd *sd)
{
struct sd_ass_priv *ctx = sd->priv;
if (sd->opts->sub_clear_on_seek || ctx->clear_once) {
ass_flush_events(ctx->ass_track);
ctx->num_seen_packets = 0;
sd->preload_ok = false;
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
ctx->clear_once = false;
}
if (ctx->converter)
lavc_conv_reset(ctx->converter);
}
static void uninit(struct sd *sd)
{
struct sd_ass_priv *ctx = sd->priv;
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
filters_destroy(sd);
if (ctx->converter)
lavc_conv_uninit(ctx->converter);
assobjects_destroy(sd);
video: make OSD/subtitle bitmaps refcounted (sort of) Making OSD/subtitle bitmaps refcounted was planend a longer time ago, e.g. the sub_bitmaps.packed field (which refcounts the subtitle bitmap data) was added in 2016. But nothing benefited much from it, because struct sub_bitmaps was usually stack allocated, and there was this weird callback stuff through osd_draw(). Make it possible to get actually refcounted subtitle bitmaps on the OSD API level. For this, we just copy all subtitle data other than the bitmaps with sub_bitmaps_copy(). At first, I had planned some fancy refcount shit, but when that was a big mess and hard to debug and just boiled to emulating malloc(), I made it a full allocation+copy. This affects mostly the parts array. With crazy ASS subtitles, this parts array can get pretty big (thousands of elements or more), in which case the extra alloc/copy could become performance relevant. But then again this is just pure bullshit, and I see no need to care. In practice, this extra work most likely gets drowned out by libass murdering a single core (while mpv is waiting for it) anyway. So fuck it. I just wanted this so draw_bmp.c requires only a single call to render everything. VOs also can benefit from this, because the weird callback shit isn't necessary anymore (simpler code), but I haven't done anything about it yet. In general I'd hope this will work towards simplifying the OSD layer, which is prerequisite for making actual further improvements. I haven't tested some cases such as the "overlay-add" command. Maybe it crashes now? Who knows, who cares. In addition, it might be worthwhile to reduce the code duplication between all the things that output subtitle bitmaps (with repacking, image allocation, etc.), but that's orthogonal.
2020-04-26 21:34:32 +00:00
talloc_free(ctx->copy_cache);
}
static int control(struct sd *sd, enum sd_ctrl cmd, void *arg)
{
struct sd_ass_priv *ctx = sd->priv;
switch (cmd) {
case SD_CTRL_SUB_STEP: {
double *a = arg;
long long ts = llrint(a[0] * 1000.0);
long long res = ass_step_sub(ctx->ass_track, ts, a[1]);
if (!res)
return false;
// Try to account for overlapping durations
a[0] += res / 1000.0 + SUB_SEEK_OFFSET;
return true;
}
case SD_CTRL_SET_ANIMATED_CHECK:
ctx->check_animated = *(bool *)arg;
return CONTROL_OK;
case SD_CTRL_SET_VIDEO_PARAMS:
ctx->video_params = *(struct mp_image_params *)arg;
return CONTROL_OK;
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
case SD_CTRL_UPDATE_OPTS: {
int flags = (uintptr_t)arg;
if (flags & UPDATE_SUB_FILT) {
filters_destroy(sd);
filters_init(sd);
ctx->clear_once = true; // allow reloading on seeks
reset(sd);
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
}
if (flags & UPDATE_SUB_HARD) {
// ass_track will be recreated, so clear duplicate cache
ctx->clear_once = true;
reset(sd);
assobjects_destroy(sd);
assobjects_init(sd);
}
ctx->ass_configured = false; // ass always needs to be reconfigured
sub: make filter_sdh a "proper" filter, allow runtime changes Until now, filter_sdh was simply a function that was called by sd_ass directly (if enabled). I want to add another filter, so it's time to turn this into a somewhat more general subtitle filtering infrastructure. I pondered whether to reuse the audio/video filtering stuff - but better not. Also, since subtitles are horrible and tend to refuse proper abstraction, it's still messed into sd_ass, instead of working on the dec_sub.c level. Actually mpv used to have subtitle "filters" and even made subtitle converters part of it, but it was fairly horrible, so don't do that again. In addition, make runtime changes possible. Since this was supposed to be a quick hack, I just decided to put all subtitle filter options into a separate option group (=> simpler change notification), to manually push the change through the playloop (like it was sort of before for OSD options), and to recreate the sub filter chain completely in every change. Should be good enough. One strangeness is that due to prefetching and such, most subtitle packets (or those some time ahead) are actually done filtering when we change, so the user still needs to manually seek to actually refresh everything. And since subtitle data is usually cached in ASS_Track (for other terrible but user-friendly reasons), we also must clear the subtitle data, but of course only on seek, since otherwise all subtitles would just disappear. What a fucking mess, but such is life. We could trigger a "refresh seek" to make this more automatic, but I don't feel like it currently. This is slightly inefficient (lots of allocations and copying), but I decided that it doesn't matter. Could matter slightly for crazy ASS subtitles that render with thousands of events. Not very well tested. Still seems to work, but I didn't have many test cases.
2020-02-16 00:02:17 +00:00
return CONTROL_OK;
}
default:
return CONTROL_UNKNOWN;
}
}
const struct sd_functions sd_ass = {
.name = "ass",
.accept_packets_in_advance = true,
.init = init,
.decode = decode,
.get_bitmaps = get_bitmaps,
.get_text = get_text,
.get_times = get_times,
.control = control,
.reset = reset,
.select = enable_output,
.uninit = uninit,
};
// Disgusting hack for (xy-)vsfilter color compatibility.
static void mangle_colors(struct sd *sd, struct sub_bitmaps *parts)
{
struct mp_subtitle_opts *opts = sd->opts;
struct sd_ass_priv *ctx = sd->priv;
enum pl_color_system csp = 0;
enum pl_color_levels levels = 0;
if (opts->ass_vsfilter_color_compat == 0) // "no"
return;
bool force_601 = opts->ass_vsfilter_color_compat == 3;
ASS_Track *track = ctx->ass_track;
static const int ass_csp[] = {
[YCBCR_BT601_TV] = PL_COLOR_SYSTEM_BT_601,
[YCBCR_BT601_PC] = PL_COLOR_SYSTEM_BT_601,
[YCBCR_BT709_TV] = PL_COLOR_SYSTEM_BT_709,
[YCBCR_BT709_PC] = PL_COLOR_SYSTEM_BT_709,
[YCBCR_SMPTE240M_TV] = PL_COLOR_SYSTEM_SMPTE_240M,
[YCBCR_SMPTE240M_PC] = PL_COLOR_SYSTEM_SMPTE_240M,
};
static const int ass_levels[] = {
[YCBCR_BT601_TV] = PL_COLOR_LEVELS_LIMITED,
[YCBCR_BT601_PC] = PL_COLOR_LEVELS_FULL,
[YCBCR_BT709_TV] = PL_COLOR_LEVELS_LIMITED,
[YCBCR_BT709_PC] = PL_COLOR_LEVELS_FULL,
[YCBCR_SMPTE240M_TV] = PL_COLOR_LEVELS_LIMITED,
[YCBCR_SMPTE240M_PC] = PL_COLOR_LEVELS_FULL,
};
int trackcsp = track->YCbCrMatrix;
if (force_601)
trackcsp = YCBCR_BT601_TV;
// NONE is a bit random, but the intention is: don't modify colors.
if (trackcsp == YCBCR_NONE)
return;
if (trackcsp < sizeof(ass_csp) / sizeof(ass_csp[0]))
csp = ass_csp[trackcsp];
if (trackcsp < sizeof(ass_levels) / sizeof(ass_levels[0]))
levels = ass_levels[trackcsp];
if (trackcsp == YCBCR_DEFAULT) {
csp = PL_COLOR_SYSTEM_BT_601;
levels = PL_COLOR_LEVELS_LIMITED;
}
// Unknown colorspace (either YCBCR_UNKNOWN, or a valid value unknown to us)
if (!csp || !levels)
return;
struct mp_image_params params = ctx->video_params;
if (force_601) {
params.repr = (struct pl_color_repr){
.sys = PL_COLOR_SYSTEM_BT_709,
.levels = PL_COLOR_LEVELS_LIMITED,
};
}
if ((csp == params.repr.sys && levels == params.repr.levels) ||
params.repr.sys == PL_COLOR_SYSTEM_RGB) // Even VSFilter doesn't mangle on RGB video
return;
bool basic_conv = params.repr.sys == PL_COLOR_SYSTEM_BT_709 &&
params.repr.levels == PL_COLOR_LEVELS_LIMITED &&
csp == PL_COLOR_SYSTEM_BT_601 &&
levels == PL_COLOR_LEVELS_LIMITED;
// With "basic", only do as much as needed for basic compatibility.
if (opts->ass_vsfilter_color_compat == 1 && !basic_conv)
return;
if (params.repr.sys != ctx->last_params.repr.sys ||
params.repr.levels != ctx->last_params.repr.levels)
{
int msgl = basic_conv ? MSGL_V : MSGL_WARN;
ctx->last_params = params;
2013-12-21 18:06:37 +00:00
MP_MSG(sd, msgl, "mangling colors like vsfilter: "
"RGB -> %s %s -> %s %s -> RGB\n",
m_opt_choice_str(pl_csp_names, csp),
m_opt_choice_str(pl_csp_levels_names, levels),
m_opt_choice_str(pl_csp_names, params.repr.sys),
m_opt_choice_str(pl_csp_names, params.repr.levels));
}
// Conversion that VSFilter would use
struct mp_csp_params vs_params = MP_CSP_PARAMS_DEFAULTS;
vs_params.repr.sys = csp;
vs_params.repr.levels = levels;
struct pl_transform3x3 vs_yuv2rgb;
mp_get_csp_matrix(&vs_params, &vs_yuv2rgb);
pl_transform3x3_invert(&vs_yuv2rgb);
// Proper conversion to RGB
struct mp_csp_params rgb_params = MP_CSP_PARAMS_DEFAULTS;
rgb_params.repr = params.repr;
rgb_params.color = params.color;
struct pl_transform3x3 vs2rgb;
mp_get_csp_matrix(&rgb_params, &vs2rgb);
for (int n = 0; n < parts->num_parts; n++) {
struct sub_bitmap *sb = &parts->parts[n];
uint32_t color = sb->libass.color;
int r = (color >> 24u) & 0xff;
int g = (color >> 16u) & 0xff;
int b = (color >> 8u) & 0xff;
int a = 0xff - (color & 0xff);
int rgb[3] = {r, g, b}, yuv[3];
mp_map_fixp_color(&vs_yuv2rgb, 8, rgb, 8, yuv);
mp_map_fixp_color(&vs2rgb, 8, yuv, 8, rgb);
sb->libass.color = MP_ASS_RGBA(rgb[0], rgb[1], rgb[2], a);
}
}
int sd_ass_fmt_offset(const char *evt_fmt)
{
// "Text" is always last (as it's arbitrary content in buf), e.g. format:
// "Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text"
int n = 0;
while (evt_fmt && (evt_fmt = strchr(evt_fmt, ',')))
evt_fmt++, n++;
return n-1; // buffer is without the format's Start/End, with ReadOrder
}
bstr sd_ass_pkt_text(struct sd_filter *ft, struct demux_packet *pkt, int offset)
{
// e.g. pkt->buffer ("4" is ReadOrder): "4,0,Default,,0,0,0,,fifth line"
bstr txt = {(char *)pkt->buffer, pkt->len}, t0 = txt;
while (offset-- > 0) {
int n = bstrchr(txt, ',');
if (n < 0) { // shouldn't happen
MP_WARN(ft, "Malformed event '%.*s'\n", BSTR_P(t0));
return (bstr){NULL, 0};
}
txt = bstr_cut(txt, n+1);
}
return txt;
}
bstr sd_ass_to_plaintext(char **out, const char *in)
{
bstr b = {*out};
ass_to_plaintext(&b, in);
*out = b.start;
return b;
}