mirror of
https://github.com/swaywm/sway.git
synced 2024-11-29 19:31:29 +00:00
446 lines
11 KiB
C
446 lines
11 KiB
C
#define _XOPEN_SOURCE 700
|
|
#include <stdlib.h>
|
|
#include <stdio.h>
|
|
#include <stdbool.h>
|
|
#include <pcre.h>
|
|
#include "sway/criteria.h"
|
|
#include "sway/tree/container.h"
|
|
#include "sway/config.h"
|
|
#include "sway/tree/view.h"
|
|
#include "stringop.h"
|
|
#include "list.h"
|
|
#include "log.h"
|
|
|
|
enum criteria_type { // *must* keep in sync with criteria_strings[]
|
|
CRIT_APP_ID,
|
|
CRIT_CLASS,
|
|
CRIT_CON_ID,
|
|
CRIT_CON_MARK,
|
|
CRIT_FLOATING,
|
|
CRIT_ID,
|
|
CRIT_INSTANCE,
|
|
CRIT_TILING,
|
|
CRIT_TITLE,
|
|
CRIT_URGENT,
|
|
CRIT_WINDOW_ROLE,
|
|
CRIT_WINDOW_TYPE,
|
|
CRIT_WORKSPACE,
|
|
CRIT_LAST
|
|
};
|
|
|
|
static const char * const criteria_strings[CRIT_LAST] = {
|
|
[CRIT_APP_ID] = "app_id",
|
|
[CRIT_CLASS] = "class",
|
|
[CRIT_CON_ID] = "con_id",
|
|
[CRIT_CON_MARK] = "con_mark",
|
|
[CRIT_FLOATING] = "floating",
|
|
[CRIT_ID] = "id",
|
|
[CRIT_INSTANCE] = "instance",
|
|
[CRIT_TILING] = "tiling",
|
|
[CRIT_TITLE] = "title",
|
|
[CRIT_URGENT] = "urgent", // either "latest" or "oldest" ...
|
|
[CRIT_WINDOW_ROLE] = "window_role",
|
|
[CRIT_WINDOW_TYPE] = "window_type",
|
|
[CRIT_WORKSPACE] = "workspace"
|
|
};
|
|
|
|
/**
|
|
* A single criteria token (ie. value/regex pair),
|
|
* e.g. 'class="some class regex"'.
|
|
*/
|
|
struct crit_token {
|
|
enum criteria_type type;
|
|
pcre *regex;
|
|
char *raw;
|
|
};
|
|
|
|
static void free_crit_token(struct crit_token *crit) {
|
|
pcre_free(crit->regex);
|
|
free(crit->raw);
|
|
free(crit);
|
|
}
|
|
|
|
static void free_crit_tokens(list_t *crit_tokens) {
|
|
for (int i = 0; i < crit_tokens->length; i++) {
|
|
free_crit_token(crit_tokens->items[i]);
|
|
}
|
|
list_free(crit_tokens);
|
|
}
|
|
|
|
// Extracts criteria string from its brackets. Returns new (duplicate)
|
|
// substring.
|
|
static char *criteria_from(const char *arg) {
|
|
char *criteria = NULL;
|
|
if (*arg == '[') {
|
|
criteria = strdup(arg + 1);
|
|
} else {
|
|
criteria = strdup(arg);
|
|
}
|
|
|
|
int last = strlen(criteria) - 1;
|
|
if (criteria[last] == ']') {
|
|
criteria[last] = '\0';
|
|
}
|
|
return criteria;
|
|
}
|
|
|
|
// Return instances of c found in str.
|
|
static int countchr(char *str, char c) {
|
|
int found = 0;
|
|
for (int i = 0; str[i]; i++) {
|
|
if (str[i] == c) {
|
|
++found;
|
|
}
|
|
}
|
|
return found;
|
|
}
|
|
|
|
// criteria_str is e.g. '[class="some class regex" instance="instance name"]'.
|
|
//
|
|
// Will create array of pointers in buf, where first is duplicate of given
|
|
// string (must be freed) and the rest are pointers to names and values in the
|
|
// base string (every other, naturally). argc will be populated with the length
|
|
// of buf.
|
|
//
|
|
// Returns error string or NULL if successful.
|
|
static char *crit_tokens(int *argc, char ***buf,
|
|
const char * const criteria_str) {
|
|
wlr_log(L_DEBUG, "Parsing criteria: '%s'", criteria_str);
|
|
char *base = criteria_from(criteria_str);
|
|
char *head = base;
|
|
char *namep = head; // start of criteria name
|
|
char *valp = NULL; // start of value
|
|
|
|
// We're going to place EOS markers where we need to and fill up an array
|
|
// of pointers to the start of each token (either name or value).
|
|
int pairs = countchr(base, '=');
|
|
int max_tokens = pairs * 2 + 1; // this gives us at least enough slots
|
|
|
|
char **argv = *buf = calloc(max_tokens, sizeof(char*));
|
|
argv[0] = base; // this needs to be freed by caller
|
|
bool quoted = true;
|
|
|
|
*argc = 1; // uneven = name, even = value
|
|
while (*head && *argc < max_tokens) {
|
|
if (namep != head && *(head - 1) == '\\') {
|
|
// escaped character: don't try to parse this
|
|
} else if (*head == '=' && namep != head) {
|
|
if (*argc % 2 != 1) {
|
|
// we're not expecting a name
|
|
return strdup("Unable to parse criteria: "
|
|
"Found out of place equal sign");
|
|
} else {
|
|
// name ends here
|
|
char *end = head; // don't want to rewind the head
|
|
while (*(end - 1) == ' ') {
|
|
--end;
|
|
}
|
|
*end = '\0';
|
|
if (*(namep) == ' ') {
|
|
namep = strrchr(namep, ' ') + 1;
|
|
}
|
|
argv[*argc] = namep;
|
|
*argc += 1;
|
|
}
|
|
} else if (*head == '"') {
|
|
if (*argc % 2 != 0) {
|
|
// we're not expecting a value
|
|
return strdup("Unable to parse criteria: "
|
|
"Found quoted value where it was not expected");
|
|
} else if (!valp) { // value starts here
|
|
valp = head + 1;
|
|
quoted = true;
|
|
} else {
|
|
// value ends here
|
|
argv[*argc] = valp;
|
|
*argc += 1;
|
|
*head = '\0';
|
|
valp = NULL;
|
|
namep = head + 1;
|
|
}
|
|
} else if (*argc % 2 == 0 && *head != ' ') {
|
|
// parse unquoted values
|
|
if (!valp) {
|
|
quoted = false;
|
|
valp = head; // value starts here
|
|
}
|
|
} else if (valp && !quoted && *head == ' ') {
|
|
// value ends here
|
|
argv[*argc] = valp;
|
|
*argc += 1;
|
|
*head = '\0';
|
|
valp = NULL;
|
|
namep = head + 1;
|
|
}
|
|
head++;
|
|
}
|
|
|
|
// catch last unquoted value if needed
|
|
if (valp && !quoted && !*head) {
|
|
argv[*argc] = valp;
|
|
*argc += 1;
|
|
}
|
|
|
|
return NULL;
|
|
}
|
|
|
|
// Returns error string on failure or NULL otherwise.
|
|
static char *parse_criteria_name(enum criteria_type *type, char *name) {
|
|
*type = CRIT_LAST;
|
|
for (int i = 0; i < CRIT_LAST; i++) {
|
|
if (strcmp(criteria_strings[i], name) == 0) {
|
|
*type = (enum criteria_type) i;
|
|
break;
|
|
}
|
|
}
|
|
if (*type == CRIT_LAST) {
|
|
const char *fmt = "Criteria type '%s' is invalid or unsupported.";
|
|
int len = strlen(name) + strlen(fmt) - 1;
|
|
char *error = malloc(len);
|
|
snprintf(error, len, fmt, name);
|
|
return error;
|
|
} else if (*type == CRIT_URGENT || *type == CRIT_WINDOW_ROLE ||
|
|
*type == CRIT_WINDOW_TYPE) {
|
|
// (we're just being helpful here)
|
|
const char *fmt = "\"%s\" criteria currently unsupported, "
|
|
"no window will match this";
|
|
int len = strlen(fmt) + strlen(name) - 1;
|
|
char *error = malloc(len);
|
|
snprintf(error, len, fmt, name);
|
|
return error;
|
|
}
|
|
return NULL;
|
|
}
|
|
|
|
// Returns error string on failure or NULL otherwise.
|
|
static char *generate_regex(pcre **regex, char *value) {
|
|
const char *reg_err;
|
|
int offset;
|
|
|
|
*regex = pcre_compile(value, PCRE_UTF8 | PCRE_UCP, ®_err, &offset, NULL);
|
|
|
|
if (!*regex) {
|
|
const char *fmt = "Regex compilation (for '%s') failed: %s";
|
|
int len = strlen(fmt) + strlen(value) + strlen(reg_err) - 3;
|
|
char *error = malloc(len);
|
|
snprintf(error, len, fmt, value, reg_err);
|
|
return error;
|
|
}
|
|
return NULL;
|
|
}
|
|
|
|
// Test whether the criterion corresponds to the currently focused window
|
|
static bool crit_is_focused(const char *value) {
|
|
return !strcmp(value, "focused") || !strcmp(value, "__focused__");
|
|
}
|
|
|
|
// Populate list with crit_tokens extracted from criteria string, returns error
|
|
// string or NULL if successful.
|
|
char *extract_crit_tokens(list_t *tokens, const char * const criteria) {
|
|
int argc;
|
|
char **argv = NULL, *error = NULL;
|
|
if ((error = crit_tokens(&argc, &argv, criteria))) {
|
|
goto ect_cleanup;
|
|
}
|
|
for (int i = 1; i + 1 < argc; i += 2) {
|
|
char* name = argv[i], *value = argv[i + 1];
|
|
struct crit_token *token = calloc(1, sizeof(struct crit_token));
|
|
token->raw = strdup(value);
|
|
|
|
if ((error = parse_criteria_name(&token->type, name))) {
|
|
free_crit_token(token);
|
|
goto ect_cleanup;
|
|
} else if (token->type == CRIT_URGENT || crit_is_focused(value)) {
|
|
wlr_log(L_DEBUG, "%s -> \"%s\"", name, value);
|
|
list_add(tokens, token);
|
|
} else if((error = generate_regex(&token->regex, value))) {
|
|
free_crit_token(token);
|
|
goto ect_cleanup;
|
|
} else {
|
|
wlr_log(L_DEBUG, "%s -> /%s/", name, value);
|
|
list_add(tokens, token);
|
|
}
|
|
}
|
|
ect_cleanup:
|
|
free(argv[0]); // base string
|
|
free(argv);
|
|
return error;
|
|
}
|
|
|
|
static int regex_cmp(const char *item, const pcre *regex) {
|
|
return pcre_exec(regex, NULL, item, strlen(item), 0, 0, NULL, 0);
|
|
}
|
|
|
|
// test a single view if it matches list of criteria tokens (all of them).
|
|
static bool criteria_test(struct sway_container *cont, list_t *tokens) {
|
|
if (cont->type != C_VIEW) {
|
|
return false;
|
|
}
|
|
int matches = 0;
|
|
for (int i = 0; i < tokens->length; i++) {
|
|
struct crit_token *crit = tokens->items[i];
|
|
switch (crit->type) {
|
|
case CRIT_CLASS:
|
|
{
|
|
const char *class = view_get_class(cont->sway_view);
|
|
if (!class) {
|
|
break;
|
|
}
|
|
if (crit->regex && regex_cmp(class, crit->regex) == 0) {
|
|
matches++;
|
|
}
|
|
break;
|
|
}
|
|
case CRIT_CON_ID:
|
|
{
|
|
char *endptr;
|
|
size_t crit_id = strtoul(crit->raw, &endptr, 10);
|
|
|
|
if (*endptr == 0 && cont->id == crit_id) {
|
|
++matches;
|
|
}
|
|
break;
|
|
}
|
|
case CRIT_CON_MARK:
|
|
// TODO
|
|
break;
|
|
case CRIT_FLOATING:
|
|
// TODO
|
|
break;
|
|
case CRIT_ID:
|
|
// TODO
|
|
break;
|
|
case CRIT_APP_ID:
|
|
{
|
|
const char *app_id = view_get_app_id(cont->sway_view);
|
|
if (!app_id) {
|
|
break;
|
|
}
|
|
|
|
if (crit->regex && regex_cmp(app_id, crit->regex) == 0) {
|
|
matches++;
|
|
}
|
|
break;
|
|
}
|
|
case CRIT_INSTANCE:
|
|
{
|
|
const char *instance = view_get_instance(cont->sway_view);
|
|
if (!instance) {
|
|
break;
|
|
}
|
|
|
|
if (crit->regex && regex_cmp(instance, crit->regex) == 0) {
|
|
matches++;
|
|
}
|
|
break;
|
|
}
|
|
case CRIT_TILING:
|
|
// TODO
|
|
break;
|
|
case CRIT_TITLE:
|
|
{
|
|
const char *title = view_get_title(cont->sway_view);
|
|
if (!title) {
|
|
break;
|
|
}
|
|
|
|
if (crit->regex && regex_cmp(title, crit->regex) == 0) {
|
|
matches++;
|
|
}
|
|
break;
|
|
}
|
|
case CRIT_URGENT:
|
|
// TODO "latest" or "oldest"
|
|
break;
|
|
case CRIT_WINDOW_ROLE:
|
|
// TODO
|
|
break;
|
|
case CRIT_WINDOW_TYPE:
|
|
// TODO
|
|
break;
|
|
case CRIT_WORKSPACE:
|
|
// TODO
|
|
break;
|
|
default:
|
|
sway_abort("Invalid criteria type (%i)", crit->type);
|
|
break;
|
|
}
|
|
}
|
|
return matches == tokens->length;
|
|
}
|
|
|
|
int criteria_cmp(const void *a, const void *b) {
|
|
if (a == b) {
|
|
return 0;
|
|
} else if (!a) {
|
|
return -1;
|
|
} else if (!b) {
|
|
return 1;
|
|
}
|
|
const struct criteria *crit_a = a, *crit_b = b;
|
|
int cmp = lenient_strcmp(crit_a->cmdlist, crit_b->cmdlist);
|
|
if (cmp != 0) {
|
|
return cmp;
|
|
}
|
|
return lenient_strcmp(crit_a->crit_raw, crit_b->crit_raw);
|
|
}
|
|
|
|
void free_criteria(struct criteria *crit) {
|
|
if (crit->tokens) {
|
|
free_crit_tokens(crit->tokens);
|
|
}
|
|
if (crit->cmdlist) {
|
|
free(crit->cmdlist);
|
|
}
|
|
if (crit->crit_raw) {
|
|
free(crit->crit_raw);
|
|
}
|
|
free(crit);
|
|
}
|
|
|
|
bool criteria_any(struct sway_container *cont, list_t *criteria) {
|
|
for (int i = 0; i < criteria->length; i++) {
|
|
struct criteria *bc = criteria->items[i];
|
|
if (criteria_test(cont, bc->tokens)) {
|
|
return true;
|
|
}
|
|
}
|
|
return false;
|
|
}
|
|
|
|
list_t *criteria_for(struct sway_container *cont) {
|
|
list_t *criteria = config->criteria, *matches = create_list();
|
|
for (int i = 0; i < criteria->length; i++) {
|
|
struct criteria *bc = criteria->items[i];
|
|
if (criteria_test(cont, bc->tokens)) {
|
|
list_add(matches, bc);
|
|
}
|
|
}
|
|
return matches;
|
|
}
|
|
|
|
struct list_tokens {
|
|
list_t *list;
|
|
list_t *tokens;
|
|
};
|
|
|
|
static void container_match_add(struct sway_container *container,
|
|
struct list_tokens *list_tokens) {
|
|
if (criteria_test(container, list_tokens->tokens)) {
|
|
list_add(list_tokens->list, container);
|
|
}
|
|
}
|
|
|
|
list_t *container_for_crit_tokens(list_t *tokens) {
|
|
struct list_tokens list_tokens =
|
|
(struct list_tokens){create_list(), tokens};
|
|
|
|
container_for_each_descendant_dfs(&root_container,
|
|
(void (*)(struct sway_container *, void *))container_match_add,
|
|
&list_tokens);
|
|
|
|
// TODO look in the scratchpad
|
|
|
|
return list_tokens.list;
|
|
}
|