mpv/options/m_config.c

1774 lines
57 KiB
C
Raw Permalink Normal View History

/*
* This file is part of mpv.
*
* 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
* GNU Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with mpv. If not, see <http://www.gnu.org/licenses/>.
*/
/// \file
/// \ingroup Config
#include "config.h"
#include <stdlib.h>
#include <stdio.h>
#include <errno.h>
#include <string.h>
#include <strings.h>
#include <assert.h>
#include <stdbool.h>
#include <pthread.h>
#include "libmpv/client.h"
#include "mpv_talloc.h"
#include "m_config.h"
#include "options/m_option.h"
#include "common/common.h"
#include "common/global.h"
#include "common/msg.h"
#include "common/msg_control.h"
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
#include "misc/dispatch.h"
#include "misc/node.h"
#include "osdep/atomic.h"
extern const char mp_help_text[];
static const union m_option_value default_value;
// Profiles allow to predefine some sets of options that can then
// be applied later on with the internal -profile option.
#define MAX_PROFILE_DEPTH 20
// Maximal include depth.
#define MAX_RECURSION_DEPTH 8
// Maximum possibly option name length (as it appears to the user).
#define MAX_OPT_NAME_LEN 80
// For use with m_config_cache.
struct m_config_shadow {
pthread_mutex_t lock;
// Incremented on every option change.
mp_atomic_uint64 ts;
// -- immutable after init
// List of m_sub_options instances.
// Index 0 is the top-level and is always present.
// Immutable after init.
// Invariant: a parent is always at a lower index than any of its children.
struct m_config_group *groups;
int num_groups;
// -- protected by lock
struct m_config_data *data; // protected shadow copy of the option data
struct config_cache **listeners;
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
int num_listeners;
};
// Represents a sub-struct (OPT_SUBSTRUCT()).
struct m_config_group {
const struct m_sub_options *group;
int group_count; // 1 + number of all sub groups owned by this (so
// m_config_shadow.groups[idx..idx+group_count] is used
// by the entire tree of sub groups included by this
// group)
int parent_group; // index of parent group into m_config_shadow.groups[],
// or -1 for group 0
int parent_ptr; // ptr offset in the parent group's data, or -1 if
// none
const char *prefix; // concat_name(_, prefix, opt->name) => full name
// (the parent names are already included in this)
};
// A copy of option data. Used for the main option struct, the shadow data,
// and copies for m_config_cache.
struct m_config_data {
struct m_config_shadow *shadow; // option definitions etc., main data copy
int group_index; // start index into m_config.groups[]
struct m_group_data *gdata; // user struct allocation (our copy of data)
int num_gdata; // (group_index+num_gdata = end index)
};
struct config_cache {
struct m_config_cache *public;
struct m_config_data *data; // public data
struct m_config_data *src; // global data (currently ==shadow->data)
struct m_config_shadow *shadow; // global metadata
uint64_t ts; // timestamp of this data copy
bool in_list; // part of m_config_shadow->listeners[]
2019-11-28 20:19:41 +00:00
int upd_group; // for "incremental" change notification
int upd_opt;
// --- Implicitly synchronized by setting/unsetting wakeup_cb.
struct mp_dispatch_queue *wakeup_dispatch_queue;
void (*wakeup_dispatch_cb)(void *ctx);
void *wakeup_dispatch_cb_ctx;
// --- Protected by shadow->lock
void (*wakeup_cb)(void *ctx);
void *wakeup_cb_ctx;
};
// Per m_config_data state for each m_config_group.
struct m_group_data {
char *udata; // pointer to group user option struct
uint64_t ts; // timestamp of the data copy
};
struct m_profile {
struct m_profile *next;
char *name;
char *desc;
int num_opts;
// Option/value pair array.
char **opts;
};
// In the file local case, this contains the old global value.
struct m_opt_backup {
struct m_opt_backup *next;
struct m_config_option *co;
void *backup;
};
player: change m_config to use new option handling mechanisms Instead of making m_config a special-case, it more or less uses the underlying m_config_cache/m_config_shadow APIs properly. This makes the player core a (relatively) equivalent user of the core option API. In particular, this means that other threads can change core options with m_config_cache_write_opt() calls (before this commit, this merely led to diverging option values). An important change is that before this commit, mpctx->opts contained the "master copy" of all option data. Now it's just another copy of the option data, and the shadow copy is considered the master. This is why whenever mpctx->opts is written, the change needs to be copied to the master (thus why this commits add a bunch of m_config_notify... calls). If another thread (e.g. a VO) changes an option, async_change_cb is now invoked, which funnels the change notification through the player's layers. The new self_notification parameter on mp_option_change_callback is so that m_config_notify... doesn't trigger recursion, and it's used in cases where the change was already "processed". It's still needed to trigger libmpv property updates. (I considered using an extra m_config_cache for that, but it'd only cause problems with no advantages.) I think the recent changes actually forgot to send libmpv property updates in some cases. This should fix this anyway. In some cases, property updates are reworked, and the potential for bugs should be lower (probably). The primary point of this change is to allow external updates, for example by a VO writing the fullscreen option if the window state is changed by the window manager (rather than mpv changing it). This is not used yet, but the following commits will.
2019-11-29 11:49:15 +00:00
static struct m_config_cache *m_config_cache_alloc_internal(void *ta_parent,
struct m_config_shadow *shadow,
const struct m_sub_options *group);
static void add_sub_group(struct m_config_shadow *shadow, const char *name_prefix,
int parent_group_index, int parent_ptr,
const struct m_sub_options *subopts);
static struct m_group_data *m_config_gdata(struct m_config_data *data,
int group_index)
{
if (group_index < data->group_index ||
group_index >= data->group_index + data->num_gdata)
return NULL;
return &data->gdata[group_index - data->group_index];
}
// Like concat_name(), but returns either a, b, or buf. buf/buf_size is used as
// target for snprintf(). (buf_size is recommended to be MAX_OPT_NAME_LEN.)
static const char *concat_name_buf(char *buf, size_t buf_size,
const char *a, const char *b)
{
assert(a);
assert(b);
if (!a[0])
return b;
if (!b[0])
return a;
snprintf(buf, buf_size, "%s-%s", a, b);
return buf;
}
// Return full option name from prefix (a) and option name (b). Returns either
// a, b, or a talloc'ed string under ta_parent.
static const char *concat_name(void *ta_parent, const char *a, const char *b)
{
char buf[MAX_OPT_NAME_LEN];
const char *r = concat_name_buf(buf, sizeof(buf), a, b);
return r == buf ? talloc_strdup(ta_parent, r) : r;
}
struct opt_iterate_state {
// User can read these fields.
int group_index;
int opt_index;
const struct m_option *opt;
const char *full_name; // may point to name_buf
// Internal.
int group_index_end;
char name_buf[MAX_OPT_NAME_LEN];
struct m_config_shadow *shadow;
};
// Start iterating all options and sub-options in the given group.
static void opt_iterate_init(struct opt_iterate_state *iter,
struct m_config_shadow *shadow, int group_index)
{
assert(group_index >= 0 && group_index < shadow->num_groups);
iter->group_index = group_index;
iter->group_index_end = group_index + shadow->groups[group_index].group_count;
iter->opt_index = -1;
iter->shadow = shadow;
}
// Get the first or next option. Returns false if end reached. If this returns
// true, most fields in *iter are valid.
// This does not return pseudo-option entries (like m_option_type_subconfig).
static bool opt_iterate_next(struct opt_iterate_state *iter)
{
if (iter->group_index < 0)
return false;
while (1) {
if (iter->group_index >= iter->group_index_end) {
iter->group_index = -1;
return false;
}
iter->opt_index += 1;
struct m_config_group *g = &iter->shadow->groups[iter->group_index];
const struct m_option *opts = g->group->opts;
if (!opts || !opts[iter->opt_index].name) {
iter->group_index += 1;
iter->opt_index = -1;
continue;
}
iter->opt = &opts[iter->opt_index];
if (iter->opt->type == &m_option_type_subconfig)
continue;
iter->full_name = concat_name_buf(iter->name_buf, sizeof(iter->name_buf),
g->prefix, iter->opt->name);
return true;
}
assert(0);
}
static void list_profiles(struct m_config *config)
{
MP_INFO(config, "Available profiles:\n");
for (struct m_profile *p = config->profiles; p; p = p->next)
MP_INFO(config, "\t%s\t%s\n", p->name, p->desc ? p->desc : "");
MP_INFO(config, "\n");
}
static int show_profile(struct m_config *config, bstr param)
{
struct m_profile *p;
if (!param.len) {
list_profiles(config);
return M_OPT_EXIT;
}
if (!(p = m_config_get_profile(config, param))) {
MP_ERR(config, "Unknown profile '%.*s'.\n", BSTR_P(param));
return M_OPT_EXIT;
}
if (!config->profile_depth)
MP_INFO(config, "Profile %s: %s\n", p->name,
p->desc ? p->desc : "");
config->profile_depth++;
for (int i = 0; i < p->num_opts; i++) {
MP_INFO(config, "%*s%s=%s\n", config->profile_depth, "",
p->opts[2 * i], p->opts[2 * i + 1]);
if (config->profile_depth < MAX_PROFILE_DEPTH
&& !strcmp(p->opts[2*i], "profile")) {
char *e, *list = p->opts[2 * i + 1];
while ((e = strchr(list, ','))) {
int l = e - list;
if (!l)
continue;
show_profile(config, (bstr){list, e - list});
list = e + 1;
}
if (list[0] != '\0')
show_profile(config, bstr0(list));
}
}
config->profile_depth--;
if (!config->profile_depth)
MP_INFO(config, "\n");
return M_OPT_EXIT;
}
// The memcpys are supposed to work around the strict aliasing violation,
// that would result if we just dereferenced a void** (where the void** is
// actually casted from struct some_type* ). The dummy struct type is in
// theory needed, because void* and struct pointers could have different
// representations, while pointers to different struct types don't.
static void *substruct_read_ptr(const void *ptr)
{
struct mp_dummy_ *res;
memcpy(&res, ptr, sizeof(res));
return res;
}
static void substruct_write_ptr(void *ptr, void *val)
mplayer: turn playtree into a list, and change per-file option handling Summary: - There is no playtree anymore. It's reduced to a simple list. - Options are now always global. You can still have per-file options, but these are optional and require special syntax. - The slave command pt_step has been removed, and playlist_next and playlist_prev added. (See etc/input.conf changes.) This is a user visible incompatible change, and will break slave-mode applications. - The pt_clear slave command is renamed to playlist_clear. - Playtree entries could have multiple files. This is not the case anymore, and playlist entries have always exactly one entry. Whenever something adds more than one file (like ASX playlists or dvd:// or dvdnav:// on the command line), all files are added as separate playlist entries. Note that some of the changes are quite deep and violent. Expect regressions. The playlist parsing code in particular is of low quality. I didn't try to improve it, and merely spent to least effort necessary to keep it somehow working. (Especially ASX playlist handling.) The playtree code was complicated and bloated. It was also barely used. Most users don't even know that mplayer manages the playlist as tree, or how to use it. The most obscure features was probably specifying a tree on command line (with '{' and '}' to create/close tree nodes). It filled the player code with complexity and confused users with weird slave commands like pt_up. Replace the playtree with a simple flat playlist. Playlist parsers that actually return trees are changed to append all files to the playlist pre-order. It used to be the responsibility of the playtree code to change per-file config options. Now this is done by the player core, and the playlist code is free of such details. Options are not per-file by default anymore. This was a very obscure and complicated feature that confused even experienced users. Consider the following command line: mplayer file1.mkv file2.mkv --no-audio file3.mkv This will disable the audio for file2.mkv only, because options are per-file by default. To make the option affect all files, you're supposed to put it before the first file. This is bad, because normally you don't need per-file options. They are very rarely needed, and the only reasonable use cases I can imagine are use of the encode backend (mplayer encode branch), or for debugging. The normal use case is made harder, and the feature is perceived as bug. Even worse, correct usage is hard to explain for users. Make all options global by default. The position of an option isn't significant anymore (except for options that compensate each other, consider --shuffle --no-shuffle). One other important change is that no options are reset anymore if a new file is started. If you change settings with slave mode commands, they will not be changed by playing a new file. (Exceptions include settings that are too file specific, like audio/subtitle stream selection.) There is still some need for per-file options. Debugging and encoding are use cases that profit from per-file options. Per-file profiles (as well as per-protocol and per-VO/AO options) need the implementation related mechanisms to backup and restore options when the playback file changes. Simplify the save-slot stuff, which is possible because there is no hierarchical play tree anymore. Now there's a simple backup field. Add a way to specify per-file options on command line. Example: mplayer f1.mkv -o0 --{ -o1 f2.mkv -o2 f3.mkv --} f4.mkv -o3 will have the following options per file set: f1.mkv, f4.mkv: -o0 -o3 f2.mkv, f3.mkv: -o0 -o3 -o1 -o2 The options --{ and --} start and end per-file options. All files inside the { } will be affected by the options equally (similar to how global options and multiple files are handled). When playback of a file starts, the per-file options are set according to the command line. When playback ends, the per-file options are restored to the values when playback started.
2012-07-31 19:33:26 +00:00
{
struct mp_dummy_ *src = val;
memcpy(ptr, &src, sizeof(src));
}
// Initialize a field with a given value. In case this is dynamic data, it has
// to be allocated and copied. src can alias dst.
static void init_opt_inplace(const struct m_option *opt, void *dst,
const void *src)
{
// The option will use dynamic memory allocation iff it has a free callback.
if (opt->type->free) {
union m_option_value temp;
memcpy(&temp, src, opt->type->size);
memset(dst, 0, opt->type->size);
m_option_copy(opt, dst, &temp);
} else if (src != dst) {
memcpy(dst, src, opt->type->size);
}
}
static void alloc_group(struct m_config_data *data, int group_index,
struct m_config_data *copy)
mplayer: turn playtree into a list, and change per-file option handling Summary: - There is no playtree anymore. It's reduced to a simple list. - Options are now always global. You can still have per-file options, but these are optional and require special syntax. - The slave command pt_step has been removed, and playlist_next and playlist_prev added. (See etc/input.conf changes.) This is a user visible incompatible change, and will break slave-mode applications. - The pt_clear slave command is renamed to playlist_clear. - Playtree entries could have multiple files. This is not the case anymore, and playlist entries have always exactly one entry. Whenever something adds more than one file (like ASX playlists or dvd:// or dvdnav:// on the command line), all files are added as separate playlist entries. Note that some of the changes are quite deep and violent. Expect regressions. The playlist parsing code in particular is of low quality. I didn't try to improve it, and merely spent to least effort necessary to keep it somehow working. (Especially ASX playlist handling.) The playtree code was complicated and bloated. It was also barely used. Most users don't even know that mplayer manages the playlist as tree, or how to use it. The most obscure features was probably specifying a tree on command line (with '{' and '}' to create/close tree nodes). It filled the player code with complexity and confused users with weird slave commands like pt_up. Replace the playtree with a simple flat playlist. Playlist parsers that actually return trees are changed to append all files to the playlist pre-order. It used to be the responsibility of the playtree code to change per-file config options. Now this is done by the player core, and the playlist code is free of such details. Options are not per-file by default anymore. This was a very obscure and complicated feature that confused even experienced users. Consider the following command line: mplayer file1.mkv file2.mkv --no-audio file3.mkv This will disable the audio for file2.mkv only, because options are per-file by default. To make the option affect all files, you're supposed to put it before the first file. This is bad, because normally you don't need per-file options. They are very rarely needed, and the only reasonable use cases I can imagine are use of the encode backend (mplayer encode branch), or for debugging. The normal use case is made harder, and the feature is perceived as bug. Even worse, correct usage is hard to explain for users. Make all options global by default. The position of an option isn't significant anymore (except for options that compensate each other, consider --shuffle --no-shuffle). One other important change is that no options are reset anymore if a new file is started. If you change settings with slave mode commands, they will not be changed by playing a new file. (Exceptions include settings that are too file specific, like audio/subtitle stream selection.) There is still some need for per-file options. Debugging and encoding are use cases that profit from per-file options. Per-file profiles (as well as per-protocol and per-VO/AO options) need the implementation related mechanisms to backup and restore options when the playback file changes. Simplify the save-slot stuff, which is possible because there is no hierarchical play tree anymore. Now there's a simple backup field. Add a way to specify per-file options on command line. Example: mplayer f1.mkv -o0 --{ -o1 f2.mkv -o2 f3.mkv --} f4.mkv -o3 will have the following options per file set: f1.mkv, f4.mkv: -o0 -o3 f2.mkv, f3.mkv: -o0 -o3 -o1 -o2 The options --{ and --} start and end per-file options. All files inside the { } will be affected by the options equally (similar to how global options and multiple files are handled). When playback of a file starts, the per-file options are set according to the command line. When playback ends, the per-file options are restored to the values when playback started.
2012-07-31 19:33:26 +00:00
{
assert(group_index == data->group_index + data->num_gdata);
assert(group_index < data->shadow->num_groups);
struct m_config_group *group = &data->shadow->groups[group_index];
const struct m_sub_options *opts = group->group;
MP_TARRAY_GROW(data, data->gdata, data->num_gdata);
struct m_group_data *gdata = &data->gdata[data->num_gdata++];
struct m_group_data *copy_gdata =
copy ? m_config_gdata(copy, group_index) : NULL;
*gdata = (struct m_group_data){
.udata = talloc_zero_size(data, opts->size),
.ts = copy_gdata ? copy_gdata->ts : 0,
};
if (opts->defaults)
memcpy(gdata->udata, opts->defaults, opts->size);
char *copy_src = copy_gdata ? copy_gdata->udata : NULL;
for (int n = 0; opts->opts && opts->opts[n].name; n++) {
const struct m_option *opt = &opts->opts[n];
if (opt->offset < 0 || opt->type->size == 0)
continue;
void *dst = gdata->udata + opt->offset;
const void *defptr = opt->defval ? opt->defval : dst;
if (copy_src)
defptr = copy_src + opt->offset;
init_opt_inplace(opt, dst, defptr);
}
// If there's a parent, update its pointer to the new struct.
if (group->parent_group >= data->group_index && group->parent_ptr >= 0) {
struct m_group_data *parent_gdata =
m_config_gdata(data, group->parent_group);
assert(parent_gdata);
substruct_write_ptr(parent_gdata->udata + group->parent_ptr, gdata->udata);
}
}
static void free_option_data(void *p)
{
struct m_config_data *data = p;
for (int i = 0; i < data->num_gdata; i++) {
struct m_group_data *gdata = &data->gdata[i];
struct m_config_group *group =
&data->shadow->groups[data->group_index + i];
const struct m_option *opts = group->group->opts;
for (int n = 0; opts && opts[n].name; n++) {
const struct m_option *opt = &opts[n];
if (opt->offset >= 0 && opt->type->size > 0)
m_option_free(opt, gdata->udata + opt->offset);
}
}
}
// Allocate data using the option description in shadow, starting at group_index
// (index into m_config.groups[]).
// If copy is not NULL, copy all data from there (for groups which are in both
// m_config_data instances), in all other cases init the data with the defaults.
static struct m_config_data *allocate_option_data(void *ta_parent,
struct m_config_shadow *shadow,
int group_index,
struct m_config_data *copy)
{
assert(group_index >= 0 && group_index < shadow->num_groups);
struct m_config_data *data = talloc_zero(ta_parent, struct m_config_data);
talloc_set_destructor(data, free_option_data);
data->shadow = shadow;
data->group_index = group_index;
struct m_config_group *root_group = &shadow->groups[group_index];
assert(root_group->group_count > 0);
for (int n = group_index; n < group_index + root_group->group_count; n++)
alloc_group(data, n, copy);
return data;
}
static void shadow_destroy(void *p)
{
struct m_config_shadow *shadow = p;
// must all have been unregistered
assert(shadow->num_listeners == 0);
talloc_free(shadow->data);
pthread_mutex_destroy(&shadow->lock);
}
static struct m_config_shadow *m_config_shadow_new(const struct m_sub_options *root)
{
struct m_config_shadow *shadow = talloc_zero(NULL, struct m_config_shadow);
talloc_set_destructor(shadow, shadow_destroy);
pthread_mutex_init(&shadow->lock, NULL);
add_sub_group(shadow, NULL, -1, -1, root);
if (!root->size)
return shadow;
shadow->data = allocate_option_data(shadow, shadow, 0, NULL);
return shadow;
}
static void config_destroy(void *p)
{
struct m_config *config = p;
config->option_change_callback = NULL;
m_config_restore_backups(config);
player: change m_config to use new option handling mechanisms Instead of making m_config a special-case, it more or less uses the underlying m_config_cache/m_config_shadow APIs properly. This makes the player core a (relatively) equivalent user of the core option API. In particular, this means that other threads can change core options with m_config_cache_write_opt() calls (before this commit, this merely led to diverging option values). An important change is that before this commit, mpctx->opts contained the "master copy" of all option data. Now it's just another copy of the option data, and the shadow copy is considered the master. This is why whenever mpctx->opts is written, the change needs to be copied to the master (thus why this commits add a bunch of m_config_notify... calls). If another thread (e.g. a VO) changes an option, async_change_cb is now invoked, which funnels the change notification through the player's layers. The new self_notification parameter on mp_option_change_callback is so that m_config_notify... doesn't trigger recursion, and it's used in cases where the change was already "processed". It's still needed to trigger libmpv property updates. (I considered using an extra m_config_cache for that, but it'd only cause problems with no advantages.) I think the recent changes actually forgot to send libmpv property updates in some cases. This should fix this anyway. In some cases, property updates are reworked, and the potential for bugs should be lower (probably). The primary point of this change is to allow external updates, for example by a VO writing the fullscreen option if the window state is changed by the window manager (rather than mpv changing it). This is not used yet, but the following commits will.
2019-11-29 11:49:15 +00:00
talloc_free(config->cache);
talloc_free(config->shadow);
mplayer: turn playtree into a list, and change per-file option handling Summary: - There is no playtree anymore. It's reduced to a simple list. - Options are now always global. You can still have per-file options, but these are optional and require special syntax. - The slave command pt_step has been removed, and playlist_next and playlist_prev added. (See etc/input.conf changes.) This is a user visible incompatible change, and will break slave-mode applications. - The pt_clear slave command is renamed to playlist_clear. - Playtree entries could have multiple files. This is not the case anymore, and playlist entries have always exactly one entry. Whenever something adds more than one file (like ASX playlists or dvd:// or dvdnav:// on the command line), all files are added as separate playlist entries. Note that some of the changes are quite deep and violent. Expect regressions. The playlist parsing code in particular is of low quality. I didn't try to improve it, and merely spent to least effort necessary to keep it somehow working. (Especially ASX playlist handling.) The playtree code was complicated and bloated. It was also barely used. Most users don't even know that mplayer manages the playlist as tree, or how to use it. The most obscure features was probably specifying a tree on command line (with '{' and '}' to create/close tree nodes). It filled the player code with complexity and confused users with weird slave commands like pt_up. Replace the playtree with a simple flat playlist. Playlist parsers that actually return trees are changed to append all files to the playlist pre-order. It used to be the responsibility of the playtree code to change per-file config options. Now this is done by the player core, and the playlist code is free of such details. Options are not per-file by default anymore. This was a very obscure and complicated feature that confused even experienced users. Consider the following command line: mplayer file1.mkv file2.mkv --no-audio file3.mkv This will disable the audio for file2.mkv only, because options are per-file by default. To make the option affect all files, you're supposed to put it before the first file. This is bad, because normally you don't need per-file options. They are very rarely needed, and the only reasonable use cases I can imagine are use of the encode backend (mplayer encode branch), or for debugging. The normal use case is made harder, and the feature is perceived as bug. Even worse, correct usage is hard to explain for users. Make all options global by default. The position of an option isn't significant anymore (except for options that compensate each other, consider --shuffle --no-shuffle). One other important change is that no options are reset anymore if a new file is started. If you change settings with slave mode commands, they will not be changed by playing a new file. (Exceptions include settings that are too file specific, like audio/subtitle stream selection.) There is still some need for per-file options. Debugging and encoding are use cases that profit from per-file options. Per-file profiles (as well as per-protocol and per-VO/AO options) need the implementation related mechanisms to backup and restore options when the playback file changes. Simplify the save-slot stuff, which is possible because there is no hierarchical play tree anymore. Now there's a simple backup field. Add a way to specify per-file options on command line. Example: mplayer f1.mkv -o0 --{ -o1 f2.mkv -o2 f3.mkv --} f4.mkv -o3 will have the following options per file set: f1.mkv, f4.mkv: -o0 -o3 f2.mkv, f3.mkv: -o0 -o3 -o1 -o2 The options --{ and --} start and end per-file options. All files inside the { } will be affected by the options equally (similar to how global options and multiple files are handled). When playback of a file starts, the per-file options are set according to the command line. When playback ends, the per-file options are restored to the values when playback started.
2012-07-31 19:33:26 +00:00
}
struct m_config *m_config_new(void *talloc_ctx, struct mp_log *log,
const struct m_sub_options *root)
{
struct m_config *config = talloc(talloc_ctx, struct m_config);
talloc_set_destructor(config, config_destroy);
2018-05-20 22:11:48 +00:00
*config = (struct m_config){.log = log,};
config->shadow = m_config_shadow_new(root);
if (root->size) {
player: change m_config to use new option handling mechanisms Instead of making m_config a special-case, it more or less uses the underlying m_config_cache/m_config_shadow APIs properly. This makes the player core a (relatively) equivalent user of the core option API. In particular, this means that other threads can change core options with m_config_cache_write_opt() calls (before this commit, this merely led to diverging option values). An important change is that before this commit, mpctx->opts contained the "master copy" of all option data. Now it's just another copy of the option data, and the shadow copy is considered the master. This is why whenever mpctx->opts is written, the change needs to be copied to the master (thus why this commits add a bunch of m_config_notify... calls). If another thread (e.g. a VO) changes an option, async_change_cb is now invoked, which funnels the change notification through the player's layers. The new self_notification parameter on mp_option_change_callback is so that m_config_notify... doesn't trigger recursion, and it's used in cases where the change was already "processed". It's still needed to trigger libmpv property updates. (I considered using an extra m_config_cache for that, but it'd only cause problems with no advantages.) I think the recent changes actually forgot to send libmpv property updates in some cases. This should fix this anyway. In some cases, property updates are reworked, and the potential for bugs should be lower (probably). The primary point of this change is to allow external updates, for example by a VO writing the fullscreen option if the window state is changed by the window manager (rather than mpv changing it). This is not used yet, but the following commits will.
2019-11-29 11:49:15 +00:00
config->cache =
m_config_cache_alloc_internal(config, config->shadow, root);
config->optstruct = config->cache->opts;
}
struct opt_iterate_state it;
opt_iterate_init(&it, config->shadow, 0);
while (opt_iterate_next(&it)) {
struct m_config_option co = {
.name = talloc_strdup(config, it.full_name),
.opt = it.opt,
.group_index = it.group_index,
};
player: change m_config to use new option handling mechanisms Instead of making m_config a special-case, it more or less uses the underlying m_config_cache/m_config_shadow APIs properly. This makes the player core a (relatively) equivalent user of the core option API. In particular, this means that other threads can change core options with m_config_cache_write_opt() calls (before this commit, this merely led to diverging option values). An important change is that before this commit, mpctx->opts contained the "master copy" of all option data. Now it's just another copy of the option data, and the shadow copy is considered the master. This is why whenever mpctx->opts is written, the change needs to be copied to the master (thus why this commits add a bunch of m_config_notify... calls). If another thread (e.g. a VO) changes an option, async_change_cb is now invoked, which funnels the change notification through the player's layers. The new self_notification parameter on mp_option_change_callback is so that m_config_notify... doesn't trigger recursion, and it's used in cases where the change was already "processed". It's still needed to trigger libmpv property updates. (I considered using an extra m_config_cache for that, but it'd only cause problems with no advantages.) I think the recent changes actually forgot to send libmpv property updates in some cases. This should fix this anyway. In some cases, property updates are reworked, and the potential for bugs should be lower (probably). The primary point of this change is to allow external updates, for example by a VO writing the fullscreen option if the window state is changed by the window manager (rather than mpv changing it). This is not used yet, but the following commits will.
2019-11-29 11:49:15 +00:00
struct m_group_data *gdata = config->cache
? m_config_gdata(config->cache->internal->data, it.group_index)
: NULL;
if (gdata && co.opt->offset >= 0)
co.data = gdata->udata + co.opt->offset;
MP_TARRAY_APPEND(config, config->opts, config->num_opts, co);
}
return config;
}
static struct m_config *m_config_from_obj_desc(void *talloc_ctx,
struct mp_log *log,
struct mpv_global *global,
struct m_obj_desc *desc)
{
struct m_sub_options *root = talloc_ptrtype(NULL, root);
*root = (struct m_sub_options){
.opts = desc->options,
// (global == NULL got repurposed to mean "no alloc")
.size = global ? desc->priv_size : 0,
.defaults = desc->priv_defaults,
};
struct m_config *c = m_config_new(talloc_ctx, log, root);
talloc_steal(c, root);
c->global = global;
return c;
}
struct m_config *m_config_from_obj_desc_noalloc(void *talloc_ctx,
struct mp_log *log,
struct m_obj_desc *desc)
{
return m_config_from_obj_desc(talloc_ctx, log, NULL, desc);
}
static const struct m_config_group *find_group(struct mpv_global *global,
const struct m_option *cfg)
{
struct m_config_shadow *shadow = global->config;
for (int n = 0; n < shadow->num_groups; n++) {
if (shadow->groups[n].group->opts == cfg)
return &shadow->groups[n];
}
return NULL;
}
void *m_config_group_from_desc(void *ta_parent, struct mp_log *log,
struct mpv_global *global, struct m_obj_desc *desc, const char *name)
{
const struct m_config_group *group = find_group(global, desc->options);
if (group) {
return mp_get_config_group(ta_parent, global, group->group);
} else {
void *d = talloc_zero_size(ta_parent, desc->priv_size);
if (desc->priv_defaults)
memcpy(d, desc->priv_defaults, desc->priv_size);
return d;
}
}
static int m_config_set_obj_params(struct m_config *config, struct mp_log *log,
struct mpv_global *global,
struct m_obj_desc *desc, char **args)
{
for (int n = 0; args && args[n * 2 + 0]; n++) {
bstr opt = bstr0(args[n * 2 + 0]);
bstr val = bstr0(args[n * 2 + 1]);
if (m_config_set_option_cli(config, opt, val, 0) < 0)
return -1;
}
return 0;
}
struct m_config *m_config_from_obj_desc_and_args(void *ta_parent,
struct mp_log *log, struct mpv_global *global, struct m_obj_desc *desc,
const char *name, struct m_obj_settings *defaults, char **args)
{
struct m_config *config = m_config_from_obj_desc(ta_parent, log, global, desc);
for (int n = 0; defaults && defaults[n].name; n++) {
struct m_obj_settings *entry = &defaults[n];
if (name && strcmp(entry->name, name) == 0) {
if (m_config_set_obj_params(config, log, global, desc, entry->attribs) < 0)
goto error;
}
}
if (m_config_set_obj_params(config, log, global, desc, args) < 0)
goto error;
return config;
error:
talloc_free(config);
return NULL;
}
mplayer: turn playtree into a list, and change per-file option handling Summary: - There is no playtree anymore. It's reduced to a simple list. - Options are now always global. You can still have per-file options, but these are optional and require special syntax. - The slave command pt_step has been removed, and playlist_next and playlist_prev added. (See etc/input.conf changes.) This is a user visible incompatible change, and will break slave-mode applications. - The pt_clear slave command is renamed to playlist_clear. - Playtree entries could have multiple files. This is not the case anymore, and playlist entries have always exactly one entry. Whenever something adds more than one file (like ASX playlists or dvd:// or dvdnav:// on the command line), all files are added as separate playlist entries. Note that some of the changes are quite deep and violent. Expect regressions. The playlist parsing code in particular is of low quality. I didn't try to improve it, and merely spent to least effort necessary to keep it somehow working. (Especially ASX playlist handling.) The playtree code was complicated and bloated. It was also barely used. Most users don't even know that mplayer manages the playlist as tree, or how to use it. The most obscure features was probably specifying a tree on command line (with '{' and '}' to create/close tree nodes). It filled the player code with complexity and confused users with weird slave commands like pt_up. Replace the playtree with a simple flat playlist. Playlist parsers that actually return trees are changed to append all files to the playlist pre-order. It used to be the responsibility of the playtree code to change per-file config options. Now this is done by the player core, and the playlist code is free of such details. Options are not per-file by default anymore. This was a very obscure and complicated feature that confused even experienced users. Consider the following command line: mplayer file1.mkv file2.mkv --no-audio file3.mkv This will disable the audio for file2.mkv only, because options are per-file by default. To make the option affect all files, you're supposed to put it before the first file. This is bad, because normally you don't need per-file options. They are very rarely needed, and the only reasonable use cases I can imagine are use of the encode backend (mplayer encode branch), or for debugging. The normal use case is made harder, and the feature is perceived as bug. Even worse, correct usage is hard to explain for users. Make all options global by default. The position of an option isn't significant anymore (except for options that compensate each other, consider --shuffle --no-shuffle). One other important change is that no options are reset anymore if a new file is started. If you change settings with slave mode commands, they will not be changed by playing a new file. (Exceptions include settings that are too file specific, like audio/subtitle stream selection.) There is still some need for per-file options. Debugging and encoding are use cases that profit from per-file options. Per-file profiles (as well as per-protocol and per-VO/AO options) need the implementation related mechanisms to backup and restore options when the playback file changes. Simplify the save-slot stuff, which is possible because there is no hierarchical play tree anymore. Now there's a simple backup field. Add a way to specify per-file options on command line. Example: mplayer f1.mkv -o0 --{ -o1 f2.mkv -o2 f3.mkv --} f4.mkv -o3 will have the following options per file set: f1.mkv, f4.mkv: -o0 -o3 f2.mkv, f3.mkv: -o0 -o3 -o1 -o2 The options --{ and --} start and end per-file options. All files inside the { } will be affected by the options equally (similar to how global options and multiple files are handled). When playback of a file starts, the per-file options are set according to the command line. When playback ends, the per-file options are restored to the values when playback started.
2012-07-31 19:33:26 +00:00
static void ensure_backup(struct m_config *config, struct m_config_option *co)
options: support parsing values into substructs Add an alternate mode for option parser objects (struct m_config) which is not inherently tied to any particular instance of an option value struct. Instead, this type or parsers can be used to initialize defaults in or parse values into a struct given as a parameter. They do not have the save slot functionality used for main player configuration. The new functionality will be used to replace the separate subopt_helper.c parsing code that is currently used to parse per-object suboptions in VOs etc. Previously, option default values were handled by initializing them in external code before creating a parser. This initialization was done with constants even for dynamically-allocated types like strings. Because trying to free a pointer to a constant would cause a crash when trying to replace the default with another value, parser initialization code then replaced all the original defaults with dynamically-allocated copies. This replace-with-copy behavior is no longer supported for new-style options; instead the option definition itself may contain a default value (new OPTDEF macros), and the new function m_config_initialize() is used to set all options to their default values. Convert the existing initialized dynamically allocated options in main config (the string options --dumpfile, --term-osd-esc, --input=conf) to use this. Other non-dynamic ones could be later converted to use this style of initialization too. There's currently no public call to free all dynamically allocated options in a given option struct because I intend to use talloc functionality for that (make them children of the struct and free with it).
2012-05-17 00:31:11 +00:00
{
if (!co->data)
mplayer: turn playtree into a list, and change per-file option handling Summary: - There is no playtree anymore. It's reduced to a simple list. - Options are now always global. You can still have per-file options, but these are optional and require special syntax. - The slave command pt_step has been removed, and playlist_next and playlist_prev added. (See etc/input.conf changes.) This is a user visible incompatible change, and will break slave-mode applications. - The pt_clear slave command is renamed to playlist_clear. - Playtree entries could have multiple files. This is not the case anymore, and playlist entries have always exactly one entry. Whenever something adds more than one file (like ASX playlists or dvd:// or dvdnav:// on the command line), all files are added as separate playlist entries. Note that some of the changes are quite deep and violent. Expect regressions. The playlist parsing code in particular is of low quality. I didn't try to improve it, and merely spent to least effort necessary to keep it somehow working. (Especially ASX playlist handling.) The playtree code was complicated and bloated. It was also barely used. Most users don't even know that mplayer manages the playlist as tree, or how to use it. The most obscure features was probably specifying a tree on command line (with '{' and '}' to create/close tree nodes). It filled the player code with complexity and confused users with weird slave commands like pt_up. Replace the playtree with a simple flat playlist. Playlist parsers that actually return trees are changed to append all files to the playlist pre-order. It used to be the responsibility of the playtree code to change per-file config options. Now this is done by the player core, and the playlist code is free of such details. Options are not per-file by default anymore. This was a very obscure and complicated feature that confused even experienced users. Consider the following command line: mplayer file1.mkv file2.mkv --no-audio file3.mkv This will disable the audio for file2.mkv only, because options are per-file by default. To make the option affect all files, you're supposed to put it before the first file. This is bad, because normally you don't need per-file options. They are very rarely needed, and the only reasonable use cases I can imagine are use of the encode backend (mplayer encode branch), or for debugging. The normal use case is made harder, and the feature is perceived as bug. Even worse, correct usage is hard to explain for users. Make all options global by default. The position of an option isn't significant anymore (except for options that compensate each other, consider --shuffle --no-shuffle). One other important change is that no options are reset anymore if a new file is started. If you change settings with slave mode commands, they will not be changed by playing a new file. (Exceptions include settings that are too file specific, like audio/subtitle stream selection.) There is still some need for per-file options. Debugging and encoding are use cases that profit from per-file options. Per-file profiles (as well as per-protocol and per-VO/AO options) need the implementation related mechanisms to backup and restore options when the playback file changes. Simplify the save-slot stuff, which is possible because there is no hierarchical play tree anymore. Now there's a simple backup field. Add a way to specify per-file options on command line. Example: mplayer f1.mkv -o0 --{ -o1 f2.mkv -o2 f3.mkv --} f4.mkv -o3 will have the following options per file set: f1.mkv, f4.mkv: -o0 -o3 f2.mkv, f3.mkv: -o0 -o3 -o1 -o2 The options --{ and --} start and end per-file options. All files inside the { } will be affected by the options equally (similar to how global options and multiple files are handled). When playback of a file starts, the per-file options are set according to the command line. When playback ends, the per-file options are restored to the values when playback started.
2012-07-31 19:33:26 +00:00
return;
for (struct m_opt_backup *cur = config->backup_opts; cur; cur = cur->next) {
if (cur->co->data == co->data) // comparing data ptr catches aliases
return;
}
struct m_opt_backup *bc = talloc_ptrtype(NULL, bc);
*bc = (struct m_opt_backup) {
.co = co,
.backup = talloc_zero_size(bc, co->opt->type->size),
};
m_option_copy(co->opt, bc->backup, co->data);
bc->next = config->backup_opts;
config->backup_opts = bc;
co->is_set_locally = true;
options: support parsing values into substructs Add an alternate mode for option parser objects (struct m_config) which is not inherently tied to any particular instance of an option value struct. Instead, this type or parsers can be used to initialize defaults in or parse values into a struct given as a parameter. They do not have the save slot functionality used for main player configuration. The new functionality will be used to replace the separate subopt_helper.c parsing code that is currently used to parse per-object suboptions in VOs etc. Previously, option default values were handled by initializing them in external code before creating a parser. This initialization was done with constants even for dynamically-allocated types like strings. Because trying to free a pointer to a constant would cause a crash when trying to replace the default with another value, parser initialization code then replaced all the original defaults with dynamically-allocated copies. This replace-with-copy behavior is no longer supported for new-style options; instead the option definition itself may contain a default value (new OPTDEF macros), and the new function m_config_initialize() is used to set all options to their default values. Convert the existing initialized dynamically allocated options in main config (the string options --dumpfile, --term-osd-esc, --input=conf) to use this. Other non-dynamic ones could be later converted to use this style of initialization too. There's currently no public call to free all dynamically allocated options in a given option struct because I intend to use talloc functionality for that (make them children of the struct and free with it).
2012-05-17 00:31:11 +00:00
}
void m_config_restore_backups(struct m_config *config)
{
while (config->backup_opts) {
struct m_opt_backup *bc = config->backup_opts;
config->backup_opts = bc->next;
m_config_set_option_raw(config, bc->co, bc->backup, 0);
m_option_free(bc->co->opt, bc->backup);
bc->co->is_set_locally = false;
talloc_free(bc);
}
}
void m_config_backup_opt(struct m_config *config, const char *opt)
{
struct m_config_option *co = m_config_get_co(config, bstr0(opt));
if (co) {
ensure_backup(config, co);
} else {
MP_ERR(config, "Option %s not found.\n", opt);
}
}
void m_config_backup_all_opts(struct m_config *config)
{
for (int n = 0; n < config->num_opts; n++)
ensure_backup(config, &config->opts[n]);
}
static void init_obj_settings_list(struct m_config_shadow *shadow,
int parent_group_index,
const struct m_obj_list *list)
{
struct m_obj_desc desc;
for (int n = 0; ; n++) {
if (!list->get_desc(&desc, n))
break;
if (desc.global_opts) {
add_sub_group(shadow, NULL, parent_group_index, -1,
desc.global_opts);
}
if (list->use_global_options && desc.options) {
struct m_sub_options *conf = talloc_ptrtype(shadow, conf);
*conf = (struct m_sub_options){
.prefix = desc.options_prefix,
.opts = desc.options,
.defaults = desc.priv_defaults,
.size = desc.priv_size,
};
add_sub_group(shadow, NULL, parent_group_index, -1, conf);
}
}
}
static void add_sub_group(struct m_config_shadow *shadow, const char *name_prefix,
int parent_group_index, int parent_ptr,
const struct m_sub_options *subopts)
{
// Can't be used multiple times.
for (int n = 0; n < shadow->num_groups; n++)
assert(shadow->groups[n].group != subopts);
if (!name_prefix)
name_prefix = "";
if (subopts->prefix && subopts->prefix[0]) {
assert(!name_prefix[0]);
name_prefix = subopts->prefix;
}
// You can only use UPDATE_ flags here.
assert(!(subopts->change_flags & ~(unsigned)UPDATE_OPTS_MASK));
assert(parent_group_index >= -1 && parent_group_index < shadow->num_groups);
int group_index = shadow->num_groups++;
MP_TARRAY_GROW(shadow, shadow->groups, group_index);
shadow->groups[group_index] = (struct m_config_group){
.group = subopts,
.parent_group = parent_group_index,
.parent_ptr = parent_ptr,
.prefix = name_prefix,
};
for (int i = 0; subopts->opts && subopts->opts[i].name; i++) {
const struct m_option *opt = &subopts->opts[i];
if (opt->type == &m_option_type_subconfig) {
const struct m_sub_options *new_subopts = opt->priv;
// Providing default structs in-place is not allowed.
if (opt->offset >= 0 && subopts->defaults) {
void *ptr = (char *)subopts->defaults + opt->offset;
assert(!substruct_read_ptr(ptr));
}
const char *prefix = concat_name(shadow, name_prefix, opt->name);
add_sub_group(shadow, prefix, group_index, opt->offset, new_subopts);
} else if (opt->type == &m_option_type_obj_settings_list) {
const struct m_obj_list *objlist = opt->priv;
init_obj_settings_list(shadow, group_index, objlist);
}
}
if (subopts->get_sub_options) {
for (int i = 0; ; i++) {
const struct m_sub_options *sub = NULL;
if (!subopts->get_sub_options(i, &sub))
break;
if (sub)
add_sub_group(shadow, NULL, group_index, -1, sub);
}
}
shadow->groups[group_index].group_count = shadow->num_groups - group_index;
}
2019-11-28 20:19:41 +00:00
static uint64_t get_option_change_mask(struct m_config_shadow *shadow,
int group_index, int group_root,
const struct m_option *opt)
{
uint64_t changed = opt->flags & UPDATE_OPTS_MASK;
while (group_index != group_root) {
struct m_config_group *g = &shadow->groups[group_index];
changed |= g->group->change_flags;
group_index = g->parent_group;
}
return changed;
}
struct m_config_option *m_config_get_co_raw(const struct m_config *config,
struct bstr name)
{
if (!name.len)
return NULL;
for (int n = 0; n < config->num_opts; n++) {
struct m_config_option *co = &config->opts[n];
struct bstr coname = bstr0(co->name);
if (bstrcmp(coname, name) == 0)
return co;
}
return NULL;
}
// Like m_config_get_co_raw(), but resolve aliases.
static struct m_config_option *m_config_get_co_any(const struct m_config *config,
struct bstr name)
{
struct m_config_option *co = m_config_get_co_raw(config, name);
if (!co)
return NULL;
const char *prefix = config->is_toplevel ? "--" : "";
if (co->opt->type == &m_option_type_alias) {
const char *alias = (const char *)co->opt->priv;
if (co->opt->deprecation_message && !co->warning_was_printed) {
if (co->opt->deprecation_message[0]) {
MP_WARN(config, "Warning: option %s%s was replaced with "
"%s%s: %s\n", prefix, co->name, prefix, alias,
co->opt->deprecation_message);
} else {
MP_WARN(config, "Warning: option %s%s was replaced with "
"%s%s and might be removed in the future.\n",
prefix, co->name, prefix, alias);
}
co->warning_was_printed = true;
}
return m_config_get_co_any(config, bstr0(alias));
} else if (co->opt->type == &m_option_type_removed) {
if (!co->warning_was_printed) {
char *msg = co->opt->priv;
if (msg) {
MP_FATAL(config, "Option %s%s was removed: %s\n",
prefix, co->name, msg);
} else {
MP_FATAL(config, "Option %s%s was removed.\n",
prefix, co->name);
}
co->warning_was_printed = true;
}
return NULL;
} else if (co->opt->deprecation_message) {
if (!co->warning_was_printed) {
MP_WARN(config, "Warning: option %s%s is deprecated "
"and might be removed in the future (%s).\n",
prefix, co->name, co->opt->deprecation_message);
co->warning_was_printed = true;
}
}
return co;
}
struct m_config_option *m_config_get_co(const struct m_config *config,
struct bstr name)
{
struct m_config_option *co = m_config_get_co_any(config, name);
// CLI aliases should not be real options, and are explicitly handled by
2018-05-20 10:30:49 +00:00
// m_config_set_option_cli(). So pretend it does not exist.
if (co && co->opt->type == &m_option_type_cli_alias)
co = NULL;
return co;
}
int m_config_get_co_count(struct m_config *config)
{
return config->num_opts;
}
struct m_config_option *m_config_get_co_index(struct m_config *config, int index)
{
return &config->opts[index];
}
const void *m_config_get_co_default(const struct m_config *config,
struct m_config_option *co)
{
if (co->opt->defval)
return co->opt->defval;
const struct m_sub_options *subopt =
config->shadow->groups[co->group_index].group;
if (co->opt->offset >= 0 && subopt->defaults)
return (char *)subopt->defaults + co->opt->offset;
return NULL;
}
const char *m_config_get_positional_option(const struct m_config *config, int p)
{
int pos = 0;
for (int n = 0; n < config->num_opts; n++) {
struct m_config_option *co = &config->opts[n];
if (!co->opt->deprecation_message) {
if (pos == p)
return co->name;
pos++;
}
}
return NULL;
}
// return: <0: M_OPT_ error, 0: skip, 1: check, 2: set
static int handle_set_opt_flags(struct m_config *config,
struct m_config_option *co, int flags)
{
int optflags = co->opt->flags;
bool set = !(flags & M_SETOPT_CHECK_ONLY);
if ((flags & M_SETOPT_PRE_PARSE_ONLY) && !(optflags & M_OPT_PRE_PARSE))
return 0;
if ((flags & M_SETOPT_PRESERVE_CMDLINE) && co->is_set_from_cmdline)
set = false;
if ((flags & M_SETOPT_NO_OVERWRITE) &&
(co->is_set_from_cmdline || co->is_set_from_config))
set = false;
if ((flags & M_SETOPT_NO_PRE_PARSE) && (optflags & M_OPT_PRE_PARSE))
return M_OPT_INVALID;
// Check if this option isn't forbidden in the current mode
if ((flags & M_SETOPT_FROM_CONFIG_FILE) && (optflags & M_OPT_NOCFG)) {
MP_ERR(config, "The %s option can't be used in a config file.\n",
co->name);
return M_OPT_INVALID;
}
if ((flags & M_SETOPT_BACKUP) && set)
ensure_backup(config, co);
return set ? 2 : 1;
}
void m_config_mark_co_flags(struct m_config_option *co, int flags)
{
if (flags & M_SETOPT_FROM_CMDLINE)
co->is_set_from_cmdline = true;
if (flags & M_SETOPT_FROM_CONFIG_FILE)
co->is_set_from_config = true;
}
2019-07-06 00:27:15 +00:00
// Special options that don't really fit into the option handling model. They
options: make mess to allow setting profile option with libmpv Certain options, such as --profile, --help, and many others require special-handling, because they don't fit conceptually into the option and property model. They don't store data, but perform actions. This caused the situation that profiles could not be set when using libmpv in encoding mode (although you should probably not used libmpv in encoding mode). Using libmpv always ends up in calling m_config_set_option_raw_direct(), while --profile was handled in m_config_parse_option(). Solve this by moving the handling of this from m_config_parse_option() to m_config_set_option_raw_direct(). Actually we just stuff most of this into m_config_handle_special_options(), which is only called by the aforementioned function. Strangely this also means that the --h/--help option declarations need to be changed, because they used OPT_PRINT, and now the option "parser" is always invoked before the special code. Thus, make them a string. Them being OPT_PRINT was apparently always redundant. (The other option declarations are moved for cosmetic purposes only.) The most weird change is how co->data==NULL is handled. We now allow passing down involved options to m_config_set_option_raw_direct(). The thing is that we don't want them to error if the command line parser is using them (with special handling done there), while all other code paths should raise an error. We try using M_SETOPT_FROM_CMDLINE to distinguish these cases. Note that normal libmpv users are supposed to use the "apply-profile" command instead. This probably contains a bunch of bugs, which you should report.
2017-06-15 13:15:05 +00:00
// usually store no data, but trigger actions. Caller is assumed to have called
// handle_set_opt_flags() to make sure the option can be set.
// Returns M_OPT_UNKNOWN if the option is not a special option.
static int m_config_handle_special_options(struct m_config *config,
struct m_config_option *co,
void *data, int flags)
{
if (config->use_profiles && strcmp(co->name, "profile") == 0) {
char **list = *(char ***)data;
if (list && list[0] && !list[1] && strcmp(list[0], "help") == 0) {
if (!config->profiles) {
MP_INFO(config, "No profiles have been defined.\n");
return M_OPT_EXIT;
}
list_profiles(config);
options: make mess to allow setting profile option with libmpv Certain options, such as --profile, --help, and many others require special-handling, because they don't fit conceptually into the option and property model. They don't store data, but perform actions. This caused the situation that profiles could not be set when using libmpv in encoding mode (although you should probably not used libmpv in encoding mode). Using libmpv always ends up in calling m_config_set_option_raw_direct(), while --profile was handled in m_config_parse_option(). Solve this by moving the handling of this from m_config_parse_option() to m_config_set_option_raw_direct(). Actually we just stuff most of this into m_config_handle_special_options(), which is only called by the aforementioned function. Strangely this also means that the --h/--help option declarations need to be changed, because they used OPT_PRINT, and now the option "parser" is always invoked before the special code. Thus, make them a string. Them being OPT_PRINT was apparently always redundant. (The other option declarations are moved for cosmetic purposes only.) The most weird change is how co->data==NULL is handled. We now allow passing down involved options to m_config_set_option_raw_direct(). The thing is that we don't want them to error if the command line parser is using them (with special handling done there), while all other code paths should raise an error. We try using M_SETOPT_FROM_CMDLINE to distinguish these cases. Note that normal libmpv users are supposed to use the "apply-profile" command instead. This probably contains a bunch of bugs, which you should report.
2017-06-15 13:15:05 +00:00
return M_OPT_EXIT;
}
for (int n = 0; list && list[n]; n++) {
int r = m_config_set_profile(config, list[n], flags);
if (r < 0)
return r;
}
return 0;
}
if (config->includefunc && strcmp(co->name, "include") == 0) {
char *param = *(char **)data;
if (!param || !param[0])
return M_OPT_MISSING_PARAM;
if (config->recursion_depth >= MAX_RECURSION_DEPTH) {
MP_ERR(config, "Maximum 'include' nesting depth exceeded.\n");
return M_OPT_INVALID;
}
config->recursion_depth += 1;
config->includefunc(config->includefunc_ctx, param, flags);
config->recursion_depth -= 1;
if (config->recursion_depth == 0 && config->profile_depth == 0)
m_config_finish_default_profile(config, flags);
options: make mess to allow setting profile option with libmpv Certain options, such as --profile, --help, and many others require special-handling, because they don't fit conceptually into the option and property model. They don't store data, but perform actions. This caused the situation that profiles could not be set when using libmpv in encoding mode (although you should probably not used libmpv in encoding mode). Using libmpv always ends up in calling m_config_set_option_raw_direct(), while --profile was handled in m_config_parse_option(). Solve this by moving the handling of this from m_config_parse_option() to m_config_set_option_raw_direct(). Actually we just stuff most of this into m_config_handle_special_options(), which is only called by the aforementioned function. Strangely this also means that the --h/--help option declarations need to be changed, because they used OPT_PRINT, and now the option "parser" is always invoked before the special code. Thus, make them a string. Them being OPT_PRINT was apparently always redundant. (The other option declarations are moved for cosmetic purposes only.) The most weird change is how co->data==NULL is handled. We now allow passing down involved options to m_config_set_option_raw_direct(). The thing is that we don't want them to error if the command line parser is using them (with special handling done there), while all other code paths should raise an error. We try using M_SETOPT_FROM_CMDLINE to distinguish these cases. Note that normal libmpv users are supposed to use the "apply-profile" command instead. This probably contains a bunch of bugs, which you should report.
2017-06-15 13:15:05 +00:00
return 1;
}
if (config->use_profiles && strcmp(co->name, "show-profile") == 0)
return show_profile(config, bstr0(*(char **)data));
if (config->is_toplevel && (strcmp(co->name, "h") == 0 ||
strcmp(co->name, "help") == 0))
{
char *h = *(char **)data;
mp_info(config->log, "%s", mp_help_text);
if (h && h[0])
m_config_print_option_list(config, h);
return M_OPT_EXIT;
}
if (strcmp(co->name, "list-options") == 0) {
m_config_print_option_list(config, "*");
return M_OPT_EXIT;
}
return M_OPT_UNKNOWN;
}
player: change m_config to use new option handling mechanisms Instead of making m_config a special-case, it more or less uses the underlying m_config_cache/m_config_shadow APIs properly. This makes the player core a (relatively) equivalent user of the core option API. In particular, this means that other threads can change core options with m_config_cache_write_opt() calls (before this commit, this merely led to diverging option values). An important change is that before this commit, mpctx->opts contained the "master copy" of all option data. Now it's just another copy of the option data, and the shadow copy is considered the master. This is why whenever mpctx->opts is written, the change needs to be copied to the master (thus why this commits add a bunch of m_config_notify... calls). If another thread (e.g. a VO) changes an option, async_change_cb is now invoked, which funnels the change notification through the player's layers. The new self_notification parameter on mp_option_change_callback is so that m_config_notify... doesn't trigger recursion, and it's used in cases where the change was already "processed". It's still needed to trigger libmpv property updates. (I considered using an extra m_config_cache for that, but it'd only cause problems with no advantages.) I think the recent changes actually forgot to send libmpv property updates in some cases. This should fix this anyway. In some cases, property updates are reworked, and the potential for bugs should be lower (probably). The primary point of this change is to allow external updates, for example by a VO writing the fullscreen option if the window state is changed by the window manager (rather than mpv changing it). This is not used yet, but the following commits will.
2019-11-29 11:49:15 +00:00
// This notification happens when anyone other than m_config->cache (i.e. not
// through m_config_set_option_raw() or related) changes any options.
static void async_change_cb(void *p)
{
struct m_config *config = p;
void *ptr;
while (m_config_cache_get_next_changed(config->cache, &ptr)) {
// Regrettable linear search, might degenerate to quadratic.
for (int n = 0; n < config->num_opts; n++) {
struct m_config_option *co = &config->opts[n];
if (co->data == ptr) {
if (config->option_change_callback) {
config->option_change_callback(
config->option_change_callback_ctx, co,
config->cache->change_flags, false);
}
break;
}
}
config->cache->change_flags = 0;
}
}
void m_config_set_update_dispatch_queue(struct m_config *config,
struct mp_dispatch_queue *dispatch)
{
m_config_cache_set_dispatch_change_cb(config->cache, dispatch,
async_change_cb, config);
}
// Normally m_config_cache will not send notifications when _we_ change our
// own stuff. For whatever funny reasons, we need that, though.
static void force_self_notify_change_opt(struct m_config *config,
struct m_config_option *co,
bool self_notification)
{
int changed =
get_option_change_mask(config->shadow, co->group_index, 0, co->opt);
if (config->option_change_callback) {
config->option_change_callback(config->option_change_callback_ctx, co,
changed, self_notification);
}
}
static void notify_opt(struct m_config *config, void *ptr, bool self_notification)
player: change m_config to use new option handling mechanisms Instead of making m_config a special-case, it more or less uses the underlying m_config_cache/m_config_shadow APIs properly. This makes the player core a (relatively) equivalent user of the core option API. In particular, this means that other threads can change core options with m_config_cache_write_opt() calls (before this commit, this merely led to diverging option values). An important change is that before this commit, mpctx->opts contained the "master copy" of all option data. Now it's just another copy of the option data, and the shadow copy is considered the master. This is why whenever mpctx->opts is written, the change needs to be copied to the master (thus why this commits add a bunch of m_config_notify... calls). If another thread (e.g. a VO) changes an option, async_change_cb is now invoked, which funnels the change notification through the player's layers. The new self_notification parameter on mp_option_change_callback is so that m_config_notify... doesn't trigger recursion, and it's used in cases where the change was already "processed". It's still needed to trigger libmpv property updates. (I considered using an extra m_config_cache for that, but it'd only cause problems with no advantages.) I think the recent changes actually forgot to send libmpv property updates in some cases. This should fix this anyway. In some cases, property updates are reworked, and the potential for bugs should be lower (probably). The primary point of this change is to allow external updates, for example by a VO writing the fullscreen option if the window state is changed by the window manager (rather than mpv changing it). This is not used yet, but the following commits will.
2019-11-29 11:49:15 +00:00
{
for (int n = 0; n < config->num_opts; n++) {
struct m_config_option *co = &config->opts[n];
if (co->data == ptr) {
if (m_config_cache_write_opt(config->cache, co->data))
force_self_notify_change_opt(config, co, self_notification);
player: change m_config to use new option handling mechanisms Instead of making m_config a special-case, it more or less uses the underlying m_config_cache/m_config_shadow APIs properly. This makes the player core a (relatively) equivalent user of the core option API. In particular, this means that other threads can change core options with m_config_cache_write_opt() calls (before this commit, this merely led to diverging option values). An important change is that before this commit, mpctx->opts contained the "master copy" of all option data. Now it's just another copy of the option data, and the shadow copy is considered the master. This is why whenever mpctx->opts is written, the change needs to be copied to the master (thus why this commits add a bunch of m_config_notify... calls). If another thread (e.g. a VO) changes an option, async_change_cb is now invoked, which funnels the change notification through the player's layers. The new self_notification parameter on mp_option_change_callback is so that m_config_notify... doesn't trigger recursion, and it's used in cases where the change was already "processed". It's still needed to trigger libmpv property updates. (I considered using an extra m_config_cache for that, but it'd only cause problems with no advantages.) I think the recent changes actually forgot to send libmpv property updates in some cases. This should fix this anyway. In some cases, property updates are reworked, and the potential for bugs should be lower (probably). The primary point of this change is to allow external updates, for example by a VO writing the fullscreen option if the window state is changed by the window manager (rather than mpv changing it). This is not used yet, but the following commits will.
2019-11-29 11:49:15 +00:00
return;
}
}
// ptr doesn't point to any config->optstruct field declared in the
// option list?
assert(false);
}
void m_config_notify_change_opt_ptr(struct m_config *config, void *ptr)
{
notify_opt(config, ptr, true);
}
void m_config_notify_change_opt_ptr_notify(struct m_config *config, void *ptr)
{
// (the notify bool is inverted: by not marking it as self-notification,
// the mpctx option change handler actually applies it)
notify_opt(config, ptr, false);
}
int m_config_set_option_raw(struct m_config *config,
struct m_config_option *co,
void *data, int flags)
{
if (!co)
return M_OPT_UNKNOWN;
int r = handle_set_opt_flags(config, co, flags);
if (r <= 1)
return r;
options: make mess to allow setting profile option with libmpv Certain options, such as --profile, --help, and many others require special-handling, because they don't fit conceptually into the option and property model. They don't store data, but perform actions. This caused the situation that profiles could not be set when using libmpv in encoding mode (although you should probably not used libmpv in encoding mode). Using libmpv always ends up in calling m_config_set_option_raw_direct(), while --profile was handled in m_config_parse_option(). Solve this by moving the handling of this from m_config_parse_option() to m_config_set_option_raw_direct(). Actually we just stuff most of this into m_config_handle_special_options(), which is only called by the aforementioned function. Strangely this also means that the --h/--help option declarations need to be changed, because they used OPT_PRINT, and now the option "parser" is always invoked before the special code. Thus, make them a string. Them being OPT_PRINT was apparently always redundant. (The other option declarations are moved for cosmetic purposes only.) The most weird change is how co->data==NULL is handled. We now allow passing down involved options to m_config_set_option_raw_direct(). The thing is that we don't want them to error if the command line parser is using them (with special handling done there), while all other code paths should raise an error. We try using M_SETOPT_FROM_CMDLINE to distinguish these cases. Note that normal libmpv users are supposed to use the "apply-profile" command instead. This probably contains a bunch of bugs, which you should report.
2017-06-15 13:15:05 +00:00
r = m_config_handle_special_options(config, co, data, flags);
if (r != M_OPT_UNKNOWN)
return r;
// This affects some special options like "playlist", "v". Maybe these
// should work, or maybe not. For now they would require special code.
if (!co->data)
return flags & M_SETOPT_FROM_CMDLINE ? 0 : M_OPT_UNKNOWN;
m_config_mark_co_flags(co, flags);
player: change m_config to use new option handling mechanisms Instead of making m_config a special-case, it more or less uses the underlying m_config_cache/m_config_shadow APIs properly. This makes the player core a (relatively) equivalent user of the core option API. In particular, this means that other threads can change core options with m_config_cache_write_opt() calls (before this commit, this merely led to diverging option values). An important change is that before this commit, mpctx->opts contained the "master copy" of all option data. Now it's just another copy of the option data, and the shadow copy is considered the master. This is why whenever mpctx->opts is written, the change needs to be copied to the master (thus why this commits add a bunch of m_config_notify... calls). If another thread (e.g. a VO) changes an option, async_change_cb is now invoked, which funnels the change notification through the player's layers. The new self_notification parameter on mp_option_change_callback is so that m_config_notify... doesn't trigger recursion, and it's used in cases where the change was already "processed". It's still needed to trigger libmpv property updates. (I considered using an extra m_config_cache for that, but it'd only cause problems with no advantages.) I think the recent changes actually forgot to send libmpv property updates in some cases. This should fix this anyway. In some cases, property updates are reworked, and the potential for bugs should be lower (probably). The primary point of this change is to allow external updates, for example by a VO writing the fullscreen option if the window state is changed by the window manager (rather than mpv changing it). This is not used yet, but the following commits will.
2019-11-29 11:49:15 +00:00
m_option_copy(co->opt, co->data, data);
if (m_config_cache_write_opt(config->cache, co->data))
force_self_notify_change_opt(config, co, false);
return 0;
}
// Handle CLI exceptions to option handling.
// Used to turn "--no-foo" into "--foo=no".
// It also handles looking up "--vf-add" as "--vf".
static struct m_config_option *m_config_mogrify_cli_opt(struct m_config *config,
struct bstr *name,
bool *out_negate,
int *out_add_flags)
{
*out_negate = false;
*out_add_flags = 0;
struct m_config_option *co = m_config_get_co(config, *name);
if (co)
return co;
// Turn "--no-foo" into "foo" + set *out_negate.
bstr no_name = *name;
if (!co && bstr_eatstart0(&no_name, "no-")) {
co = m_config_get_co(config, no_name);
// Not all choice types have this value - if they don't, then parsing
// them will simply result in an error. Good enough.
if (!co || !(co->opt->type->flags & M_OPT_TYPE_CHOICE))
return NULL;
*name = no_name;
*out_negate = true;
return co;
}
// Resolve CLI alias. (We don't allow you to combine them with "--no-".)
co = m_config_get_co_any(config, *name);
if (co && co->opt->type == &m_option_type_cli_alias)
*name = bstr0((char *)co->opt->priv);
// Might be a suffix "action", like "--vf-add". Expensively check for
// matches. (We don't allow you to combine them with "--no-".)
for (int n = 0; n < config->num_opts; n++) {
co = &config->opts[n];
2017-09-22 09:31:23 +00:00
struct bstr basename = bstr0(co->name);
2017-09-22 09:31:23 +00:00
if (!bstr_startswith(*name, basename))
continue;
// Aliased option + a suffix action, e.g. --opengl-shaders-append
if (co->opt->type == &m_option_type_alias)
2017-09-22 09:31:23 +00:00
co = m_config_get_co_any(config, basename);
if (!co)
continue;
const struct m_option_type *type = co->opt->type;
for (int i = 0; type->actions && type->actions[i].name; i++) {
const struct m_option_action *action = &type->actions[i];
bstr suffix = bstr0(action->name);
if (bstr_endswith(*name, suffix) &&
2017-09-22 09:31:23 +00:00
(name->len == basename.len + 1 + suffix.len) &&
name->start[basename.len] == '-')
{
*out_add_flags = action->flags;
return co;
}
}
}
return NULL;
}
int m_config_set_option_cli(struct m_config *config, struct bstr name,
struct bstr param, int flags)
{
int r;
assert(config != NULL);
bool negate;
struct m_config_option *co =
m_config_mogrify_cli_opt(config, &name, &negate, &(int){0});
if (!co) {
r = M_OPT_UNKNOWN;
goto done;
}
if (negate) {
if (param.len) {
r = M_OPT_DISALLOW_PARAM;
goto done;
}
param = bstr0("no");
}
// This is the only mandatory function
assert(co->opt->type->parse);
r = handle_set_opt_flags(config, co, flags);
if (r <= 0)
goto done;
if (r == 2) {
MP_VERBOSE(config, "Setting option '%.*s' = '%.*s' (flags = %d)\n",
BSTR_P(name), BSTR_P(param), flags);
}
union m_option_value val = {0};
// Some option types are "impure" and work on the existing data.
// (Prime examples: --vf-add, --sub-file)
if (co->data)
m_option_copy(co->opt, &val, co->data);
r = m_option_parse(config->log, co->opt, name, param, &val);
options: make mess to allow setting profile option with libmpv Certain options, such as --profile, --help, and many others require special-handling, because they don't fit conceptually into the option and property model. They don't store data, but perform actions. This caused the situation that profiles could not be set when using libmpv in encoding mode (although you should probably not used libmpv in encoding mode). Using libmpv always ends up in calling m_config_set_option_raw_direct(), while --profile was handled in m_config_parse_option(). Solve this by moving the handling of this from m_config_parse_option() to m_config_set_option_raw_direct(). Actually we just stuff most of this into m_config_handle_special_options(), which is only called by the aforementioned function. Strangely this also means that the --h/--help option declarations need to be changed, because they used OPT_PRINT, and now the option "parser" is always invoked before the special code. Thus, make them a string. Them being OPT_PRINT was apparently always redundant. (The other option declarations are moved for cosmetic purposes only.) The most weird change is how co->data==NULL is handled. We now allow passing down involved options to m_config_set_option_raw_direct(). The thing is that we don't want them to error if the command line parser is using them (with special handling done there), while all other code paths should raise an error. We try using M_SETOPT_FROM_CMDLINE to distinguish these cases. Note that normal libmpv users are supposed to use the "apply-profile" command instead. This probably contains a bunch of bugs, which you should report.
2017-06-15 13:15:05 +00:00
if (r >= 0)
r = m_config_set_option_raw(config, co, &val, flags);
m_option_free(co->opt, &val);
done:
if (r < 0 && r != M_OPT_EXIT) {
MP_ERR(config, "Error parsing option %.*s (%s)\n",
BSTR_P(name), m_option_strerror(r));
r = M_OPT_INVALID;
}
return r;
}
int m_config_set_option_node(struct m_config *config, bstr name,
struct mpv_node *data, int flags)
{
int r;
struct m_config_option *co = m_config_get_co(config, name);
if (!co)
return M_OPT_UNKNOWN;
// Do this on an "empty" type to make setting the option strictly overwrite
// the old value, as opposed to e.g. appending to lists.
union m_option_value val = {0};
if (data->format == MPV_FORMAT_STRING) {
bstr param = bstr0(data->u.string);
r = m_option_parse(mp_null_log, co->opt, name, param, &val);
} else {
r = m_option_set_node(co->opt, &val, data);
}
if (r >= 0)
r = m_config_set_option_raw(config, co, &val, flags);
if (mp_msg_test(config->log, MSGL_V)) {
char *s = m_option_type_node.print(NULL, data);
MP_DBG(config, "Setting option '%.*s' = %s (flags = %d) -> %d\n",
BSTR_P(name), s ? s : "?", flags, r);
talloc_free(s);
}
m_option_free(co->opt, &val);
return r;
}
int m_config_option_requires_param(struct m_config *config, bstr name)
{
bool negate;
int flags;
struct m_config_option *co =
m_config_mogrify_cli_opt(config, &name, &negate, &flags);
if (!co)
return M_OPT_UNKNOWN;
if (negate || (flags & M_OPT_TYPE_OPTIONAL_PARAM))
return 0;
return m_option_required_params(co->opt);
}
static int sort_opt_compare(const void *pa, const void *pb)
{
const struct m_config_option *a = pa;
const struct m_config_option *b = pb;
return strcasecmp(a->name, b->name);
}
void m_config_print_option_list(const struct m_config *config, const char *name)
{
char min[50], max[50];
int count = 0;
const char *prefix = config->is_toplevel ? "--" : "";
struct m_config_option *sorted =
talloc_memdup(NULL, config->opts, config->num_opts * sizeof(sorted[0]));
if (config->is_toplevel)
qsort(sorted, config->num_opts, sizeof(sorted[0]), sort_opt_compare);
MP_INFO(config, "Options:\n\n");
for (int i = 0; i < config->num_opts; i++) {
struct m_config_option *co = &sorted[i];
const struct m_option *opt = co->opt;
if (strcmp(name, "*") != 0 && !strstr(co->name, name))
continue;
MP_INFO(config, " %s%-30s", prefix, co->name);
if (opt->type == &m_option_type_choice) {
MP_INFO(config, " Choices:");
struct m_opt_choice_alternatives *alt = opt->priv;
for (int n = 0; alt[n].name; n++)
MP_INFO(config, " %s", alt[n].name);
if (opt->flags & (M_OPT_MIN | M_OPT_MAX))
MP_INFO(config, " (or an integer)");
} else {
MP_INFO(config, " %s", opt->type->name);
}
if (opt->flags & (M_OPT_MIN | M_OPT_MAX)) {
snprintf(min, sizeof(min), "any");
snprintf(max, sizeof(max), "any");
if (opt->flags & M_OPT_MIN)
snprintf(min, sizeof(min), "%.14g", opt->min);
if (opt->flags & M_OPT_MAX)
snprintf(max, sizeof(max), "%.14g", opt->max);
MP_INFO(config, " (%s to %s)", min, max);
}
char *def = NULL;
const void *defptr = m_config_get_co_default(config, co);
if (!defptr)
defptr = &default_value;
if (defptr)
def = m_option_pretty_print(opt, defptr);
if (def) {
MP_INFO(config, " (default: %s)", def);
talloc_free(def);
}
if (opt->flags & M_OPT_NOCFG)
MP_INFO(config, " [not in config files]");
if (opt->flags & M_OPT_FILE)
MP_INFO(config, " [file]");
if (opt->deprecation_message)
MP_INFO(config, " [deprecated]");
if (opt->type == &m_option_type_alias)
MP_INFO(config, " for %s", (char *)opt->priv);
if (opt->type == &m_option_type_cli_alias)
MP_INFO(config, " for --%s (CLI/config files only)", (char *)opt->priv);
MP_INFO(config, "\n");
for (int n = 0; opt->type->actions && opt->type->actions[n].name; n++) {
const struct m_option_action *action = &opt->type->actions[n];
MP_INFO(config, " %s%s-%s\n", prefix, co->name, action->name);
count++;
}
count++;
}
MP_INFO(config, "\nTotal: %d options\n", count);
talloc_free(sorted);
}
char **m_config_list_options(void *ta_parent, const struct m_config *config)
{
char **list = talloc_new(ta_parent);
int count = 0;
for (int i = 0; i < config->num_opts; i++) {
struct m_config_option *co = &config->opts[i];
// For use with CONF_TYPE_STRING_LIST, it's important not to set list
// as allocation parent.
char *s = talloc_strdup(ta_parent, co->name);
MP_TARRAY_APPEND(ta_parent, list, count, s);
}
MP_TARRAY_APPEND(ta_parent, list, count, NULL);
return list;
}
struct m_profile *m_config_get_profile(const struct m_config *config, bstr name)
{
for (struct m_profile *p = config->profiles; p; p = p->next) {
if (bstr_equals0(name, p->name))
return p;
}
return NULL;
}
struct m_profile *m_config_get_profile0(const struct m_config *config,
char *name)
{
return m_config_get_profile(config, bstr0(name));
}
struct m_profile *m_config_add_profile(struct m_config *config, char *name)
{
if (!name || !name[0])
name = "default";
struct m_profile *p = m_config_get_profile0(config, name);
if (p)
return p;
p = talloc_zero(config, struct m_profile);
p->name = talloc_strdup(p, name);
p->next = config->profiles;
config->profiles = p;
return p;
}
void m_profile_set_desc(struct m_profile *p, bstr desc)
{
2008-04-25 12:59:30 +00:00
talloc_free(p->desc);
p->desc = bstrto0(p, desc);
}
int m_config_set_profile_option(struct m_config *config, struct m_profile *p,
bstr name, bstr val)
{
int i = m_config_set_option_cli(config, name, val,
M_SETOPT_CHECK_ONLY |
M_SETOPT_FROM_CONFIG_FILE);
if (i < 0)
return i;
p->opts = talloc_realloc(p, p->opts, char *, 2 * (p->num_opts + 2));
p->opts[p->num_opts * 2] = bstrto0(p, name);
p->opts[p->num_opts * 2 + 1] = bstrto0(p, val);
p->num_opts++;
p->opts[p->num_opts * 2] = p->opts[p->num_opts * 2 + 1] = NULL;
return 1;
}
int m_config_set_profile(struct m_config *config, char *name, int flags)
{
struct m_profile *p = m_config_get_profile0(config, name);
if (!p) {
MP_WARN(config, "Unknown profile '%s'.\n", name);
return M_OPT_INVALID;
}
2019-11-01 00:29:44 +00:00
MP_VERBOSE(config, "Applying profile '%s'...\n", name);
if (config->profile_depth > MAX_PROFILE_DEPTH) {
MP_WARN(config, "WARNING: Profile inclusion too deep.\n");
options: make mess to allow setting profile option with libmpv Certain options, such as --profile, --help, and many others require special-handling, because they don't fit conceptually into the option and property model. They don't store data, but perform actions. This caused the situation that profiles could not be set when using libmpv in encoding mode (although you should probably not used libmpv in encoding mode). Using libmpv always ends up in calling m_config_set_option_raw_direct(), while --profile was handled in m_config_parse_option(). Solve this by moving the handling of this from m_config_parse_option() to m_config_set_option_raw_direct(). Actually we just stuff most of this into m_config_handle_special_options(), which is only called by the aforementioned function. Strangely this also means that the --h/--help option declarations need to be changed, because they used OPT_PRINT, and now the option "parser" is always invoked before the special code. Thus, make them a string. Them being OPT_PRINT was apparently always redundant. (The other option declarations are moved for cosmetic purposes only.) The most weird change is how co->data==NULL is handled. We now allow passing down involved options to m_config_set_option_raw_direct(). The thing is that we don't want them to error if the command line parser is using them (with special handling done there), while all other code paths should raise an error. We try using M_SETOPT_FROM_CMDLINE to distinguish these cases. Note that normal libmpv users are supposed to use the "apply-profile" command instead. This probably contains a bunch of bugs, which you should report.
2017-06-15 13:15:05 +00:00
return M_OPT_INVALID;
}
config->profile_depth++;
for (int i = 0; i < p->num_opts; i++) {
m_config_set_option_cli(config,
bstr0(p->opts[2 * i]),
bstr0(p->opts[2 * i + 1]),
flags | M_SETOPT_FROM_CONFIG_FILE);
}
config->profile_depth--;
return 0;
}
void m_config_finish_default_profile(struct m_config *config, int flags)
{
struct m_profile *p = m_config_add_profile(config, NULL);
m_config_set_profile(config, p->name, flags);
p->num_opts = 0;
}
struct mpv_node m_config_get_profiles(struct m_config *config)
{
struct mpv_node root;
node_init(&root, MPV_FORMAT_NODE_ARRAY, NULL);
for (m_profile_t *profile = config->profiles; profile; profile = profile->next)
{
struct mpv_node *entry = node_array_add(&root, MPV_FORMAT_NODE_MAP);
node_map_add_string(entry, "name", profile->name);
if (profile->desc)
node_map_add_string(entry, "profile-desc", profile->desc);
struct mpv_node *opts =
node_map_add(entry, "options", MPV_FORMAT_NODE_ARRAY);
for (int n = 0; n < profile->num_opts; n++) {
struct mpv_node *opt_entry = node_array_add(opts, MPV_FORMAT_NODE_MAP);
node_map_add_string(opt_entry, "key", profile->opts[n * 2 + 0]);
node_map_add_string(opt_entry, "value", profile->opts[n * 2 + 1]);
}
}
return root;
}
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
static void cache_destroy(void *p)
{
struct m_config_cache *cache = p;
// (technically speaking, being able to call them both without anything
// breaking is a feature provided by these functions)
m_config_cache_set_wakeup_cb(cache, NULL, NULL);
m_config_cache_set_dispatch_change_cb(cache, NULL, NULL, NULL);
}
player: change m_config to use new option handling mechanisms Instead of making m_config a special-case, it more or less uses the underlying m_config_cache/m_config_shadow APIs properly. This makes the player core a (relatively) equivalent user of the core option API. In particular, this means that other threads can change core options with m_config_cache_write_opt() calls (before this commit, this merely led to diverging option values). An important change is that before this commit, mpctx->opts contained the "master copy" of all option data. Now it's just another copy of the option data, and the shadow copy is considered the master. This is why whenever mpctx->opts is written, the change needs to be copied to the master (thus why this commits add a bunch of m_config_notify... calls). If another thread (e.g. a VO) changes an option, async_change_cb is now invoked, which funnels the change notification through the player's layers. The new self_notification parameter on mp_option_change_callback is so that m_config_notify... doesn't trigger recursion, and it's used in cases where the change was already "processed". It's still needed to trigger libmpv property updates. (I considered using an extra m_config_cache for that, but it'd only cause problems with no advantages.) I think the recent changes actually forgot to send libmpv property updates in some cases. This should fix this anyway. In some cases, property updates are reworked, and the potential for bugs should be lower (probably). The primary point of this change is to allow external updates, for example by a VO writing the fullscreen option if the window state is changed by the window manager (rather than mpv changing it). This is not used yet, but the following commits will.
2019-11-29 11:49:15 +00:00
static struct m_config_cache *m_config_cache_alloc_internal(void *ta_parent,
struct m_config_shadow *shadow,
const struct m_sub_options *group)
{
int group_index = -1;
for (int n = 0; n < shadow->num_groups; n++) {
if (shadow->groups[n].group == group) {
group_index = n;
break;
}
}
assert(group_index >= 0); // invalid group (or not in option tree)
struct cache_alloc {
struct m_config_cache a;
struct config_cache b;
};
struct cache_alloc *alloc = talloc_zero(ta_parent, struct cache_alloc);
assert((void *)&alloc->a == (void *)alloc);
struct m_config_cache *cache = &alloc->a;
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
talloc_set_destructor(cache, cache_destroy);
cache->internal = &alloc->b;
struct config_cache *in = cache->internal;
in->shadow = shadow;
in->src = shadow->data;
pthread_mutex_lock(&shadow->lock);
in->data = allocate_option_data(cache, shadow, group_index, in->src);
pthread_mutex_unlock(&shadow->lock);
cache->opts = in->data->gdata[0].udata;
2019-11-28 20:19:41 +00:00
in->upd_group = -1;
return cache;
}
player: change m_config to use new option handling mechanisms Instead of making m_config a special-case, it more or less uses the underlying m_config_cache/m_config_shadow APIs properly. This makes the player core a (relatively) equivalent user of the core option API. In particular, this means that other threads can change core options with m_config_cache_write_opt() calls (before this commit, this merely led to diverging option values). An important change is that before this commit, mpctx->opts contained the "master copy" of all option data. Now it's just another copy of the option data, and the shadow copy is considered the master. This is why whenever mpctx->opts is written, the change needs to be copied to the master (thus why this commits add a bunch of m_config_notify... calls). If another thread (e.g. a VO) changes an option, async_change_cb is now invoked, which funnels the change notification through the player's layers. The new self_notification parameter on mp_option_change_callback is so that m_config_notify... doesn't trigger recursion, and it's used in cases where the change was already "processed". It's still needed to trigger libmpv property updates. (I considered using an extra m_config_cache for that, but it'd only cause problems with no advantages.) I think the recent changes actually forgot to send libmpv property updates in some cases. This should fix this anyway. In some cases, property updates are reworked, and the potential for bugs should be lower (probably). The primary point of this change is to allow external updates, for example by a VO writing the fullscreen option if the window state is changed by the window manager (rather than mpv changing it). This is not used yet, but the following commits will.
2019-11-29 11:49:15 +00:00
struct m_config_cache *m_config_cache_alloc(void *ta_parent,
struct mpv_global *global,
const struct m_sub_options *group)
{
return m_config_cache_alloc_internal(ta_parent, global->config, group);
}
2019-11-28 20:19:41 +00:00
static void update_next_option(struct m_config_cache *cache, void **p_opt)
{
2019-11-28 20:19:41 +00:00
struct config_cache *in = cache->internal;
struct m_config_data *dst = in->data;
struct m_config_data *src = in->src;
2019-11-28 20:19:41 +00:00
assert(src->group_index == 0); // must be the option root currently
2019-11-28 20:19:41 +00:00
*p_opt = NULL;
2019-11-28 20:19:41 +00:00
while (in->upd_group < dst->group_index + dst->num_gdata) {
struct m_group_data *gsrc = m_config_gdata(src, in->upd_group);
struct m_group_data *gdst = m_config_gdata(dst, in->upd_group);
assert(gsrc && gdst);
2019-11-28 20:19:41 +00:00
if (gdst->ts < gsrc->ts) {
struct m_config_group *g = &dst->shadow->groups[in->upd_group];
const struct m_option *opts = g->group->opts;
while (opts && opts[in->upd_opt].name) {
const struct m_option *opt = &opts[in->upd_opt];
if (opt->offset >= 0 && opt->type->size) {
void *dsrc = gsrc->udata + opt->offset;
void *ddst = gdst->udata + opt->offset;
if (!m_option_equal(opt, ddst, dsrc)) {
uint64_t ch = get_option_change_mask(dst->shadow,
in->upd_group, dst->group_index, opt);
if (cache->debug) {
char *vdst = m_option_print(opt, ddst);
char *vsrc = m_option_print(opt, dsrc);
mp_warn(cache->debug, "Option '%s' changed from "
"'%s' to' %s' (flags = 0x%"PRIx64")\n",
opt->name, vdst, vsrc, ch);
talloc_free(vdst);
talloc_free(vsrc);
}
m_option_copy(opt, ddst, dsrc);
cache->change_flags |= ch;
in->upd_opt++; // skip this next time
*p_opt = ddst;
return;
}
}
in->upd_opt++;
}
2019-11-28 20:19:41 +00:00
gdst->ts = gsrc->ts;
}
2019-11-28 20:19:41 +00:00
in->upd_group++;
in->upd_opt = 0;
}
2019-11-28 20:19:41 +00:00
in->upd_group = -1;
}
2019-11-28 20:19:41 +00:00
static bool cache_check_update(struct m_config_cache *cache)
{
struct config_cache *in = cache->internal;
struct m_config_shadow *shadow = in->shadow;
// Using atomics and checking outside of the lock - it's unknown whether
// this makes it faster or slower. Just cargo culting it.
uint64_t new_ts = atomic_load(&shadow->ts);
if (in->ts >= new_ts)
return false;
in->ts = new_ts;
2019-11-28 20:19:41 +00:00
in->upd_group = in->data->group_index;
in->upd_opt = 0;
return true;
}
bool m_config_cache_update(struct m_config_cache *cache)
{
struct config_cache *in = cache->internal;
struct m_config_shadow *shadow = in->shadow;
if (!cache_check_update(cache))
return false;
pthread_mutex_lock(&shadow->lock);
2019-11-28 20:19:41 +00:00
bool res = false;
while (1) {
void *p;
update_next_option(cache, &p);
if (!p)
break;
res = true;
}
pthread_mutex_unlock(&shadow->lock);
return res;
}
2019-11-28 20:19:41 +00:00
bool m_config_cache_get_next_changed(struct m_config_cache *cache, void **opt)
{
struct config_cache *in = cache->internal;
struct m_config_shadow *shadow = in->shadow;
*opt = NULL;
if (!cache_check_update(cache) && in->upd_group < 0)
return false;
pthread_mutex_lock(&shadow->lock);
update_next_option(cache, opt);
pthread_mutex_unlock(&shadow->lock);
return !!*opt;
}
static void find_opt(struct m_config_shadow *shadow, struct m_config_data *data,
void *ptr, int *group_idx, int *opt_idx)
{
*group_idx = -1;
*opt_idx = -1;
for (int n = data->group_index; n < data->group_index + data->num_gdata; n++)
{
struct m_group_data *gd = m_config_gdata(data, n);
struct m_config_group *g = &shadow->groups[n];
const struct m_option *opts = g->group->opts;
for (int i = 0; opts && opts[i].name; i++) {
const struct m_option *opt = &opts[i];
if (opt->offset >= 0 && opt->type->size &&
ptr == gd->udata + opt->offset)
{
*group_idx = n;
*opt_idx = i;
return;
}
}
}
}
bool m_config_cache_write_opt(struct m_config_cache *cache, void *ptr)
{
struct config_cache *in = cache->internal;
struct m_config_shadow *shadow = in->shadow;
int group_idx = -1;
int opt_idx = -1;
find_opt(shadow, in->data, ptr, &group_idx, &opt_idx);
// ptr was not in cache->opts, or no option declaration matching it.
assert(group_idx >= 0);
struct m_config_group *g = &shadow->groups[group_idx];
const struct m_option *opt = &g->group->opts[opt_idx];
pthread_mutex_lock(&shadow->lock);
struct m_group_data *gdst = m_config_gdata(in->data, group_idx);
struct m_group_data *gsrc = m_config_gdata(in->src, group_idx);
assert(gdst && gsrc);
bool changed = !m_option_equal(opt, gsrc->udata + opt->offset, ptr);
if (changed) {
m_option_copy(opt, gsrc->udata + opt->offset, ptr);
gsrc->ts = atomic_fetch_add(&shadow->ts, 1) + 1;
for (int n = 0; n < shadow->num_listeners; n++) {
struct config_cache *listener = shadow->listeners[n];
if (listener->wakeup_cb && m_config_gdata(listener->data, group_idx))
listener->wakeup_cb(listener->wakeup_cb_ctx);
}
}
pthread_mutex_unlock(&shadow->lock);
return changed;
}
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
void m_config_cache_set_wakeup_cb(struct m_config_cache *cache,
void (*cb)(void *ctx), void *cb_ctx)
{
struct config_cache *in = cache->internal;
struct m_config_shadow *shadow = in->shadow;
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
pthread_mutex_lock(&shadow->lock);
if (in->in_list) {
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
for (int n = 0; n < shadow->num_listeners; n++) {
if (shadow->listeners[n] == in) {
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
MP_TARRAY_REMOVE_AT(shadow->listeners, shadow->num_listeners, n);
break;
}
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
}
for (int n = 0; n < shadow->num_listeners; n++)
assert(shadow->listeners[n] != in); // only 1 wakeup_cb per cache
// (The deinitialization path relies on this to free all memory.)
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
if (!shadow->num_listeners) {
talloc_free(shadow->listeners);
shadow->listeners = NULL;
}
}
if (cb) {
MP_TARRAY_APPEND(NULL, shadow->listeners, shadow->num_listeners, in);
in->in_list = true;
in->wakeup_cb = cb;
in->wakeup_cb_ctx = cb_ctx;
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
}
pthread_mutex_unlock(&shadow->lock);
}
static void dispatch_notify(void *p)
{
struct config_cache *in = p;
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
assert(in->wakeup_dispatch_queue);
mp_dispatch_enqueue_notify(in->wakeup_dispatch_queue,
in->wakeup_dispatch_cb,
in->wakeup_dispatch_cb_ctx);
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
}
void m_config_cache_set_dispatch_change_cb(struct m_config_cache *cache,
struct mp_dispatch_queue *dispatch,
void (*cb)(void *ctx), void *cb_ctx)
{
struct config_cache *in = cache->internal;
2018-05-20 10:30:49 +00:00
// Removing the old one is tricky. First make sure no new notifications will
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
// come.
m_config_cache_set_wakeup_cb(cache, NULL, NULL);
// Remove any pending notifications (assume we're on the same thread as
// any potential mp_dispatch_queue_process() callers).
if (in->wakeup_dispatch_queue) {
mp_dispatch_cancel_fn(in->wakeup_dispatch_queue,
in->wakeup_dispatch_cb,
in->wakeup_dispatch_cb_ctx);
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
}
in->wakeup_dispatch_queue = NULL;
in->wakeup_dispatch_cb = NULL;
in->wakeup_dispatch_cb_ctx = NULL;
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
if (cb) {
in->wakeup_dispatch_queue = dispatch;
in->wakeup_dispatch_cb = cb;
in->wakeup_dispatch_cb_ctx = cb_ctx;
m_config_cache_set_wakeup_cb(cache, dispatch_notify, in);
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
}
}
void *mp_get_config_group(void *ta_parent, struct mpv_global *global,
const struct m_sub_options *group)
{
struct m_config_cache *cache = m_config_cache_alloc(NULL, global, group);
// Make talloc_free(cache->opts) free the entire cache.
ta_set_parent(cache->opts, ta_parent);
ta_set_parent(cache, cache->opts);
return cache->opts;
}
void mp_read_option_raw(struct mpv_global *global, const char *name,
const struct m_option_type *type, void *dst)
{
struct m_config_shadow *shadow = global->config;
struct opt_iterate_state it;
opt_iterate_init(&it, shadow, 0);
while (opt_iterate_next(&it)) {
if (strcmp(name, it.full_name) == 0) {
struct m_group_data *gdata =
m_config_gdata(shadow->data, it.group_index);
assert(gdata);
assert(it.opt->offset >= 0);
assert(it.opt->type == type);
memset(dst, 0, it.opt->type->size);
m_option_copy(it.opt, dst, gdata->udata + it.opt->offset);
return;
}
}
assert(0); // not found
}