2 * vim:ts=4:sw=4:expandtab
4 * i3 - an improved dynamic tiling window manager
5 * © 2009-2012 Michael Stapelberg and contributors (see also: LICENSE)
7 * commands_parser.c: hand-written parser to parse commands (commands are what
8 * you bind on keys and what you can send to i3 using the IPC interface, like
9 * 'move left' or 'workspace 4').
11 * We use a hand-written parser instead of lex/yacc because our commands are
12 * easy for humans, not for computers. Thus, it’s quite hard to specify a
13 * context-free grammar for the commands. A PEG grammar would be easier, but
14 * there’s downsides to every PEG parser generator I have come accross so far.
16 * This parser is basically a state machine which looks for literals or strings
17 * and can push either on a stack. After identifying a literal or string, it
18 * will either transition to the current state, to a different state, or call a
19 * function (like cmd_move()).
21 * Special care has been taken that error messages are useful and the code is
22 * well testable (when compiled with -DTEST_PARSER it will output to stdout
23 * instead of actually calling any function).
35 // Macros to make the YAJL API a bit easier to use.
36 #define y(x, ...) yajl_gen_ ## x (command_output.json_gen, ##__VA_ARGS__)
37 #define ystr(str) yajl_gen_string(command_output.json_gen, (unsigned char*)str, strlen(str))
39 /*******************************************************************************
40 * The data structures used for parsing. Essentially the current state and a
41 * list of tokens for that state.
43 * The GENERATED_* files are generated by generate-commands-parser.pl with the
44 * input parser-specs/commands.spec.
45 ******************************************************************************/
47 #include "GENERATED_enums.h"
49 typedef struct token {
52 /* This might be __CALL */
53 cmdp_state next_state;
55 uint16_t call_identifier;
59 typedef struct tokenptr {
64 #include "GENERATED_tokens.h"
66 /*******************************************************************************
67 * The (small) stack where identified literals are stored during the parsing
68 * of a single command (like $workspace).
69 ******************************************************************************/
72 /* Just a pointer, not dynamically allocated. */
73 const char *identifier;
77 /* 10 entries should be enough for everybody. */
78 static struct stack_entry stack[10];
81 * Pushes a string (identified by 'identifier') on the stack. We simply use a
82 * single array, since the number of entries we have to store is very small.
85 static void push_string(const char *identifier, char *str) {
86 for (int c = 0; c < 10; c++) {
87 if (stack[c].identifier != NULL)
89 /* Found a free slot, let’s store it here. */
90 stack[c].identifier = identifier;
95 /* When we arrive here, the stack is full. This should not happen and
96 * means there’s either a bug in this parser or the specification
97 * contains a command with more than 10 identified tokens. */
98 fprintf(stderr, "BUG: commands_parser stack full. This means either a bug "
99 "in the code, or a new command which contains more than "
100 "10 identified tokens.\n");
104 // XXX: ideally, this would be const char. need to check if that works with all
106 static char *get_string(const char *identifier) {
107 DLOG("Getting string %s from stack...\n", identifier);
108 for (int c = 0; c < 10; c++) {
109 if (stack[c].identifier == NULL)
111 if (strcmp(identifier, stack[c].identifier) == 0)
117 static void clear_stack(void) {
118 DLOG("clearing stack.\n");
119 for (int c = 0; c < 10; c++) {
120 if (stack[c].str != NULL)
122 stack[c].identifier = NULL;
127 // TODO: remove this if it turns out we don’t need it for testing.
129 /*******************************************************************************
130 * A dynamically growing linked list which holds the criteria for the current
132 ******************************************************************************/
134 typedef struct criterion {
138 TAILQ_ENTRY(criterion) criteria;
141 static TAILQ_HEAD(criteria_head, criterion) criteria =
142 TAILQ_HEAD_INITIALIZER(criteria);
145 * Stores the given type/value in the list of criteria.
146 * Accepts a pointer as first argument, since it is 'call'ed by the parser.
149 static void push_criterion(void *unused_criteria, const char *type,
151 struct criterion *criterion = malloc(sizeof(struct criterion));
152 criterion->type = strdup(type);
153 criterion->value = strdup(value);
154 TAILQ_INSERT_TAIL(&criteria, criterion, criteria);
158 * Clears the criteria linked list.
159 * Accepts a pointer as first argument, since it is 'call'ed by the parser.
162 static void clear_criteria(void *unused_criteria) {
163 struct criterion *criterion;
164 while (!TAILQ_EMPTY(&criteria)) {
165 criterion = TAILQ_FIRST(&criteria);
166 free(criterion->type);
167 free(criterion->value);
168 TAILQ_REMOVE(&criteria, criterion, criteria);
174 /*******************************************************************************
176 ******************************************************************************/
178 static cmdp_state state;
180 static Match current_match;
182 static struct CommandResult subcommand_output;
183 static struct CommandResult command_output;
185 #include "GENERATED_call.h"
188 static void next_state(const cmdp_token *token) {
189 if (token->next_state == __CALL) {
190 DLOG("should call stuff, yay. call_id = %d\n",
191 token->extra.call_identifier);
192 subcommand_output.json_gen = command_output.json_gen;
193 subcommand_output.needs_tree_render = false;
194 GENERATED_call(token->extra.call_identifier, &subcommand_output);
195 /* If any subcommand requires a tree_render(), we need to make the
196 * whole parser result request a tree_render(). */
197 if (subcommand_output.needs_tree_render)
198 command_output.needs_tree_render = true;
203 state = token->next_state;
204 if (state == INITIAL) {
209 /* TODO: Return parsing errors via JSON. */
210 struct CommandResult *parse_command(const char *input) {
211 DLOG("new parser handling: %s\n", input);
214 /* A YAJL JSON generator used for formatting replies. */
216 command_output.json_gen = yajl_gen_alloc(NULL);
218 command_output.json_gen = yajl_gen_alloc(NULL, NULL);
222 command_output.needs_tree_render = false;
224 const char *walk = input;
225 const size_t len = strlen(input);
227 const cmdp_token *token;
230 // TODO: make this testable
232 cmd_criteria_init(¤t_match, &subcommand_output);
235 /* The "<=" operator is intentional: We also handle the terminating 0-byte
236 * explicitly by looking for an 'end' token. */
237 while ((walk - input) <= len) {
238 /* skip whitespace and newlines before every token */
239 while ((*walk == ' ' || *walk == '\t' ||
240 *walk == '\r' || *walk == '\n') && *walk != '\0')
243 DLOG("remaining input = %s\n", walk);
245 cmdp_token_ptr *ptr = &(tokens[state]);
246 token_handled = false;
247 for (c = 0; c < ptr->n; c++) {
248 token = &(ptr->array[c]);
249 DLOG("trying token %d = %s\n", c, token->name);
252 if (token->name[0] == '\'') {
254 if (strncasecmp(walk, token->name + 1, strlen(token->name) - 1) == 0) {
255 DLOG("found literal, moving to next state\n");
256 if (token->identifier != NULL)
257 push_string(token->identifier, sstrdup(token->name + 1));
258 walk += strlen(token->name) - 1;
260 token_handled = true;
266 if (strcmp(token->name, "string") == 0 ||
267 strcmp(token->name, "word") == 0) {
268 DLOG("parsing this as a string\n");
269 const char *beginning = walk;
270 /* Handle quoted strings (or words). */
274 while (*walk != '\0' && (*walk != '"' || *(walk-1) == '\\'))
277 if (token->name[0] == 's') {
278 /* For a string (starting with 's'), the delimiters are
279 * comma (,) and semicolon (;) which introduce a new
280 * operation or command, respectively. Also, newlines
282 while (*walk != ';' && *walk != ',' &&
283 *walk != '\0' && *walk != '\r' &&
287 /* For a word, the delimiters are white space (' ' or
288 * '\t'), closing square bracket (]), comma (,) and
290 while (*walk != ' ' && *walk != '\t' &&
291 *walk != ']' && *walk != ',' &&
292 *walk != ';' && *walk != '\r' &&
293 *walk != '\n' && *walk != '\0')
297 if (walk != beginning) {
298 char *str = scalloc(walk-beginning + 1);
299 /* We copy manually to handle escaping of characters. */
301 for (inpos = 0, outpos = 0;
302 inpos < (walk-beginning);
304 /* We only handle escaped double quotes to not break
305 * backwards compatibility with people using \w in
306 * regular expressions etc. */
307 if (beginning[inpos] == '\\' && beginning[inpos+1] == '"')
309 str[outpos] = beginning[inpos];
311 if (token->identifier)
312 push_string(token->identifier, str);
313 DLOG("str is \"%s\"\n", str);
314 /* If we are at the end of a quoted string, skip the ending
319 token_handled = true;
324 if (strcmp(token->name, "end") == 0) {
325 DLOG("checking for the end token.\n");
326 if (*walk == '\0' || *walk == ',' || *walk == ';') {
327 DLOG("yes, indeed. end\n");
329 token_handled = true;
330 /* To make sure we start with an appropriate matching
331 * datastructure for commands which do *not* specify any
332 * criteria, we re-initialize the criteria system after
334 // TODO: make this testable
336 if (*walk == '\0' || *walk == ';')
337 cmd_criteria_init(¤t_match, &subcommand_output);
345 if (!token_handled) {
346 /* Figure out how much memory we will need to fill in the names of
347 * all tokens afterwards. */
349 for (c = 0; c < ptr->n; c++)
350 tokenlen += strlen(ptr->array[c].name) + strlen("'', ");
352 /* Build up a decent error message. We include the problem, the
353 * full input, and underline the position where the parser
356 char *possible_tokens = smalloc(tokenlen + 1);
357 char *tokenwalk = possible_tokens;
358 for (c = 0; c < ptr->n; c++) {
359 token = &(ptr->array[c]);
360 if (token->name[0] == '\'') {
361 /* A literal is copied to the error message enclosed with
364 strcpy(tokenwalk, token->name + 1);
365 tokenwalk += strlen(token->name + 1);
368 /* Any other token is copied to the error message enclosed
369 * with angle brackets. */
371 strcpy(tokenwalk, token->name);
372 tokenwalk += strlen(token->name);
375 if (c < (ptr->n - 1)) {
381 sasprintf(&errormessage, "Expected one of these tokens: %s",
383 free(possible_tokens);
385 /* Contains the same amount of characters as 'input' has, but with
386 * the unparseable part highlighted using ^ characters. */
387 char *position = smalloc(len + 1);
388 for (const char *copywalk = input; *copywalk != '\0'; copywalk++)
389 position[(copywalk - input)] = (copywalk >= walk ? '^' : ' ');
390 position[len] = '\0';
392 printf("%s\n", errormessage);
393 printf("Your command: %s\n", input);
394 printf(" %s\n", position);
396 /* Format this error message as a JSON reply. */
404 ystr("errorposition");
417 DLOG("command_output.needs_tree_render = %d\n", command_output.needs_tree_render);
418 return &command_output;
421 /*******************************************************************************
422 * Code for building the stand-alone binary test.commands_parser which is used
423 * by t/187-commands-parser.t.
424 ******************************************************************************/
429 * Logs the given message to stdout while prefixing the current time to it,
430 * but only if the corresponding debug loglevel was activated.
431 * This is to be called by DLOG() which includes filename/linenumber
434 void debuglog(uint64_t lev, char *fmt, ...) {
438 fprintf(stderr, "# ");
439 vfprintf(stderr, fmt, args);
443 int main(int argc, char *argv[]) {
445 fprintf(stderr, "Syntax: %s <command>\n", argv[0]);
448 parse_command(argv[1]);