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.

5 files changed, 190 insertions, 43 deletions

diff --git a/sub/dec_sub.c b/sub/dec_sub.c
index 27cab201d9..f98a478be4 100644
--- a/sub/dec_sub.c
+++ b/sub/dec_sub.c
@@ -413,6 +413,7 @@ int sub_control(struct dec_sub *sub, enum sd_ctrl cmd, void *arg)
 {
     int r = CONTROL_UNKNOWN;
     pthread_mutex_lock(&sub->lock);
+    bool propagate = false;
     switch (cmd) {
     case SD_CTRL_SET_VIDEO_DEF_FPS:
         sub->video_fps = *(double *)arg;
@@ -427,23 +428,21 @@ int sub_control(struct dec_sub *sub, enum sd_ctrl cmd, void *arg)
         if (r == CONTROL_OK)
             a[0] = pts_from_subtitle(sub, arg2[0]);
         break;
+    case SD_CTRL_UPDATE_OPTS:
+        if (m_config_cache_update(sub->opts_cache))
+            update_subtitle_speed(sub);
+        propagate = true;
+        break;
     }
     default:
-        if (sub->sd->driver->control)
-            r = sub->sd->driver->control(sub->sd, cmd, arg);
+        propagate = true;
     }
+    if (propagate && sub->sd->driver->control)
+        r = sub->sd->driver->control(sub->sd, cmd, arg);
     pthread_mutex_unlock(&sub->lock);
     return r;
 }
 
-void sub_update_opts(struct dec_sub *sub)
-{
-    pthread_mutex_lock(&sub->lock);
-    if (m_config_cache_update(sub->opts_cache))
-        update_subtitle_speed(sub);
-    pthread_mutex_unlock(&sub->lock);
-}
-
 void sub_set_recorder_sink(struct dec_sub *sub, struct mp_recorder_sink *sink)
 {
     pthread_mutex_lock(&sub->lock);
diff --git a/sub/dec_sub.h b/sub/dec_sub.h
index 5449b97ad0..030b8d21e1 100644
--- a/sub/dec_sub.h
+++ b/sub/dec_sub.h
@@ -19,6 +19,7 @@ enum sd_ctrl {
     SD_CTRL_SET_VIDEO_PARAMS,
     SD_CTRL_SET_TOP,
     SD_CTRL_SET_VIDEO_DEF_FPS,
+    SD_CTRL_UPDATE_OPTS,
 };
 
 struct sd_times {
@@ -46,7 +47,6 @@ char *sub_get_text(struct dec_sub *sub, double pts);
 struct sd_times sub_get_times(struct dec_sub *sub, double pts);
 void sub_reset(struct dec_sub *sub);
 void sub_select(struct dec_sub *sub, bool selected);
-void sub_update_opts(struct dec_sub *sub);
 void sub_set_recorder_sink(struct dec_sub *sub, struct mp_recorder_sink *sink);
 void sub_set_play_dir(struct dec_sub *sub, int dir);
 
diff --git a/sub/filter_sdh.c b/sub/filter_sdh.c
index 72744ec20e..2b544ea222 100644
--- a/sub/filter_sdh.c
+++ b/sub/filter_sdh.c
@@ -15,9 +15,11 @@
  * License along with mpv.  If not, see <http://www.gnu.org/licenses/>.
  */
 
-#include <stdlib.h>
 #include <assert.h>
+#include <stdlib.h>
 #include <string.h>
+#include <limits.h>
+
 #include "misc/ctype.h"
 #include "common/common.h"
 #include "common/msg.h"
@@ -43,7 +45,7 @@ static void init_buf(struct buffer *buf, int length)
     buf->length = length;
 }
 
-static inline int append(struct sd *sd, struct buffer *buf, char c)
+static inline int append(struct sd_filter *sd, struct buffer *buf, char c)
 {
     if (buf->pos >= 0 && buf->pos < buf->length) {
         buf->string[buf->pos++] = c;
@@ -66,7 +68,7 @@ static inline int append(struct sd *sd, struct buffer *buf, char c)
 //
 // on return the read pointer is updated to the position after
 // the tags.
-static void copy_ass(struct sd *sd, char **rpp, struct buffer *buf)
+static void copy_ass(struct sd_filter *sd, char **rpp, struct buffer *buf)
 {
     char *rp = *rpp;
 
@@ -83,7 +85,7 @@ static void copy_ass(struct sd *sd, char **rpp, struct buffer *buf)
     return;
 }
 
-static bool skip_bracketed(struct sd *sd, char **rpp, struct buffer *buf);
+static bool skip_bracketed(struct sd_filter *sd, char **rpp, struct buffer *buf);
 
 // check for speaker label, like MAN:
 // normal subtitles may include mixed case text with : after so
@@ -101,7 +103,7 @@ static bool skip_bracketed(struct sd *sd, char **rpp, struct buffer *buf);
 // if no label was found read pointer and write position in buffer
 // will be unchanged
 // otherwise they point to next position after label and next write position
-static void skip_speaker_label(struct sd *sd, char **rpp, struct buffer *buf)
+static void skip_speaker_label(struct sd_filter *sd, char **rpp, struct buffer *buf)
 {
     int filter_harder = sd->opts->sub_filter_SDH_harder;
     char *rp = *rpp;
@@ -187,7 +189,7 @@ static void skip_speaker_label(struct sd *sd, char **rpp, struct buffer *buf)
 // return true if bracketed text was removed.
 // if not valid SDH read pointer and write buffer position will be unchanged
 // otherwise they point to next position after text and next write position
-static bool skip_bracketed(struct sd *sd, char **rpp, struct buffer *buf)
+static bool skip_bracketed(struct sd_filter *sd, char **rpp, struct buffer *buf)
 {
     char *rp = *rpp;
     int old_pos = buf->pos;
@@ -235,7 +237,7 @@ static bool skip_bracketed(struct sd *sd, char **rpp, struct buffer *buf)
 // return true if paranthesed text was removed.
 // if not valid SDH read pointer and write buffer position will be unchanged
 // otherwise they point to next position after text and next write position
-static bool skip_parenthesed(struct sd *sd, char **rpp, struct buffer *buf)
+static bool skip_parenthesed(struct sd_filter *sd, char **rpp, struct buffer *buf)
 {
     int filter_harder = sd->opts->sub_filter_SDH_harder;
     char *rp = *rpp;
@@ -289,7 +291,8 @@ static bool skip_parenthesed(struct sd *sd, char **rpp, struct buffer *buf)
 //
 // when removing characters the following are moved back
 //
-static void remove_leading_hyphen_space(struct sd *sd, int start_pos, struct buffer *buf)
+static void remove_leading_hyphen_space(struct sd_filter *sd, int start_pos,
+                                        struct buffer *buf)
 {
     int old_pos = buf->pos;
     if (start_pos < 0 || start_pos >= old_pos)
@@ -340,7 +343,8 @@ static void remove_leading_hyphen_space(struct sd *sd, int start_pos, struct buf
 //
 // Returns NULL if filtering resulted in all of ASS data being removed so no
 // subtitle should be output
-char *filter_SDH(struct sd *sd, char *format, int n_ignored, char *data, int length)
+static char *filter_SDH(struct sd_filter *sd, char *format, int n_ignored,
+                        char *data, int length)
 {
     if (!format) {
         MP_VERBOSE(sd, "SDH filtering not possible - format missing\n");
@@ -462,3 +466,42 @@ char *filter_SDH(struct sd *sd, char *format, int n_ignored, char *data, int len
         return NULL;
     }
 }
+
+static bool sdh_init(struct sd_filter *ft)
+{
+    if (strcmp(ft->codec, "ass") != 0)
+        return false;
+
+    if (!ft->opts->sub_filter_SDH)
+        return false;
+
+    return true;
+}
+
+static struct demux_packet *sdh_filter(struct sd_filter *ft,
+                                       struct demux_packet *pkt)
+{
+    char *line = (char *)pkt->buffer;
+    size_t len = pkt->len;
+    if (len >= INT_MAX)
+        return NULL;
+
+    line = filter_SDH(ft, ft->event_format, 1, line, len);
+    if (!line)
+        return NULL;
+
+    // Stupidly, this copies it again. One could possibly allocate the packet
+    // for writing in the first place (new_demux_packet()) and use
+    // demux_packet_shorten(). Or not allocate anything on no change.
+    struct demux_packet *npkt = new_demux_packet_from(line, strlen(line));
+    if (npkt)
+        demux_packet_copy_attribs(npkt, pkt);
+
+    talloc_free(line);
+    return npkt;
+}
+
+const struct sd_filter_functions sd_filter_sdh = {
+    .init   = sdh_init,
+    .filter = sdh_filter,
+};
diff --git a/sub/sd.h b/sub/sd.h
index 6c3fc4e285..0d361edba6 100644
--- a/sub/sd.h
+++ b/sub/sd.h
@@ -3,6 +3,7 @@
 
 #include "dec_sub.h"
 #include "demux/packet.h"
+#include "misc/bstr.h"
 
 // up to 210 ms overlaps or gaps are removed
 #define SUB_GAP_THRESHOLD 0.210
@@ -43,6 +44,7 @@ struct sd_functions {
     struct sd_times (*get_times)(struct sd *sd, double pts);
 };
 
+// lavc_conv.c
 struct lavc_conv;
 struct lavc_conv *lavc_conv_create(struct mp_log *log, const char *codec_name,
                                    char *extradata, int extradata_len);
@@ -52,6 +54,38 @@ char **lavc_conv_decode(struct lavc_conv *priv, struct demux_packet *packet,
 void lavc_conv_reset(struct lavc_conv *priv);
 void lavc_conv_uninit(struct lavc_conv *priv);
 
-char *filter_SDH(struct sd *sd, char *format, int n_ignored, char *data, int length);
+struct sd_filter {
+    struct mpv_global *global;
+    struct mp_log *log;
+    struct mp_sub_filter_opts *opts;
+    const struct sd_filter_functions *driver;
+
+    void *priv;
+
+    // Static codec parameters. Set by sd; cannot be changed by filter.
+    char *codec;
+    char *event_format;
+};
+
+struct sd_filter_functions {
+    bool (*init)(struct sd_filter *ft);
+
+    // Filter an ASS event (usually in the Matroska format, but event_format
+    // can be used to determine details).
+    // Returning NULL is interpreted as dropping the event completely.
+    // Returning pkt makes it no-op.
+    // If the returned packet is not pkt or NULL, it must have been properly
+    // allocated.
+    // pkt is owned by the caller (and freed by the caller when needed).
+    // Note: as by normal demux_packet rules, you must not modify any fields in
+    //       it, or the data referenced by it. You must create a new demux_packet
+    //       when modifying data.
+    struct demux_packet *(*filter)(struct sd_filter *ft,
+                                   struct demux_packet *pkt);
+
+    void (*uninit)(struct sd_filter *ft);
+};
+
+extern const struct sd_filter_functions sd_filter_sdh;
 
 #endif
diff --git a/sub/sd_ass.c b/sub/sd_ass.c
index 5443a688d4..2b0cf13127 100644
--- a/sub/sd_ass.c
+++ b/sub/sd_ass.c
@@ -26,6 +26,8 @@
 
 #include "mpv_talloc.h"
 
+#include "config.h"
+#include "options/m_config.h"
 #include "options/options.h"
 #include "common/common.h"
 #include "common/msg.h"
@@ -43,6 +45,9 @@ struct sd_ass_priv {
     struct ass_track *shadow_track; // for --sub-ass=no rendering
     bool is_converted;
     struct lavc_conv *converter;
+    struct sd_filter **filters;
+    int num_filters;
+    bool clear_once;
     bool on_top;
     struct mp_ass_packer *packer;
     struct sub_bitmap *bs;
@@ -57,6 +62,12 @@ struct sd_ass_priv {
 static void mangle_colors(struct sd *sd, struct sub_bitmaps *parts);
 static void fill_plaintext(struct sd *sd, double pts);
 
+static const struct sd_filter_functions *const filters[] = {
+    // Note: list order defines filter order.
+    &sd_filter_sdh,
+    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)
@@ -130,6 +141,43 @@ static void add_subtitle_fonts(struct sd *sd)
     }
 }
 
+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 = ctx->ass_track->event_format,
+        };
+        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;
@@ -198,12 +246,37 @@ static int init(struct sd *sd)
 #endif
 
     enable_output(sd, true);
+    filters_init(sd);
 
     ctx->packer = mp_ass_packer_alloc(ctx);
 
     return 0;
 }
 
+// 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;
+
+    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));
+
+    if (pkt != orig_pkt)
+        talloc_free(pkt);
+}
+
 // Test if the packet with the given file position (used as unique ID) 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.
@@ -251,15 +324,13 @@ static void decode(struct sd *sd, struct demux_packet *packet)
         }
 
         for (int n = 0; r && r[n]; n++) {
-            char *ass_line = r[n];
-            if (sd->opts->sub_filter_SDH)
-                ass_line = filter_SDH(sd, track->event_format, 1, ass_line, 0);
-            if (ass_line)
-                ass_process_chunk(track, ass_line, strlen(ass_line),
-                                  llrint(sub_pts * 1000),
-                                  llrint(sub_duration * 1000));
-            if (sd->opts->sub_filter_SDH)
-                talloc_free(ass_line);
+            struct demux_packet pkt2 = {
+                .pts = sub_pts,
+                .duration = sub_duration,
+                .buffer = r[n],
+                .len = strlen(r[n]),
+            };
+            filter_and_add(sd, &pkt2);
         }
         if (ctx->duration_unknown) {
             for (int n = 0; n < track->n_events - 1; n++) {
@@ -272,18 +343,7 @@ static void decode(struct sd *sd, struct demux_packet *packet)
     } else {
         // Note that for this packet format, libass has an internal mechanism
         // for discarding duplicate (already seen) packets.
-        char *ass_line = packet->buffer;
-        int ass_len = packet->len;
-        if (sd->opts->sub_filter_SDH) {
-            ass_line = filter_SDH(sd, track->event_format, 1, ass_line, ass_len);
-            ass_len = ass_line ? strlen(ass_line) : 0;
-        }
-        if (ass_line)
-            ass_process_chunk(track, ass_line, ass_len,
-                              llrint(packet->pts * 1000),
-                              llrint(packet->duration * 1000));
-        if (sd->opts->sub_filter_SDH)
-            talloc_free(ass_line);
+        filter_and_add(sd, packet);
     }
 }
 
@@ -668,10 +728,11 @@ static void fill_plaintext(struct sd *sd, double pts)
 static void reset(struct sd *sd)
 {
     struct sd_ass_priv *ctx = sd->priv;
-    if (sd->opts->sub_clear_on_seek || ctx->duration_unknown) {
+    if (sd->opts->sub_clear_on_seek || ctx->duration_unknown || ctx->clear_once) {
         ass_flush_events(ctx->ass_track);
         ctx->num_seen_packets = 0;
         sd->preload_ok = false;
+        ctx->clear_once = false;
     }
     if (ctx->converter)
         lavc_conv_reset(ctx->converter);
@@ -681,6 +742,7 @@ static void uninit(struct sd *sd)
 {
     struct sd_ass_priv *ctx = sd->priv;
 
+    filters_destroy(sd);
     if (ctx->converter)
         lavc_conv_uninit(ctx->converter);
     ass_free_track(ctx->ass_track);
@@ -708,6 +770,15 @@ static int control(struct sd *sd, enum sd_ctrl cmd, void *arg)
     case SD_CTRL_SET_TOP:
         ctx->on_top = *(bool *)arg;
         return CONTROL_OK;
+    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
+        }
+        return CONTROL_OK;
+    }
     default:
         return CONTROL_UNKNOWN;
     }