Implement support for PCRE regular expressions for all criteria (for_window, commands, assignments)

common.mk

@@ -49,6 +49,7 @@ CFLAGS += $(call cflags_for_lib, xcursor)
 CFLAGS += $(call cflags_for_lib, x11)
 CFLAGS += $(call cflags_for_lib, yajl)
 CFLAGS += $(call cflags_for_lib, libev)
+CFLAGS += $(call cflags_for_lib, libpcre)
 CPPFLAGS += -DI3_VERSION=\"${GIT_VERSION}\"
 CPPFLAGS += -DSYSCONFDIR=\"${SYSCONFDIR}\"
 CPPFLAGS += -DTERM_EMU=\"$(TERM_EMU)\"
@@ -70,6 +71,7 @@ LIBS += $(call ldflags_for_lib, xcursor, Xcursor)
 LIBS += $(call ldflags_for_lib, x11, X11)
 LIBS += $(call ldflags_for_lib, yajl, yajl)
 LIBS += $(call ldflags_for_lib, libev, ev)
+LIBS += $(call ldflags_for_lib, libpcre, pcre)
 
 # Please test if -Wl,--as-needed works on your platform and send me a patch.
 # it is known not to work on Darwin (Mac OS X)

include/all.h

@@ -64,5 +64,6 @@
 #include "output.h"
 #include "ewmh.h"
 #include "assignments.h"
+#include "regex.h"
 
 #endif

include/data.h

@@ -10,6 +10,7 @@
 #include <xcb/randr.h>
 #include <xcb/xcb_atom.h>
 #include <stdbool.h>
+#include <pcre.h>
 
 #ifndef _DATA_H
 #define _DATA_H
@@ -137,6 +138,21 @@ struct Ignore_Event {
     SLIST_ENTRY(Ignore_Event) ignore_events;
 };
 
+/**
+ * Regular expression wrapper. It contains the pattern itself as a string (like
+ * ^foo[0-9]$) as well as a pointer to the compiled PCRE expression and the
+ * pcre_extra data returned by pcre_study().
+ *
+ * This makes it easier to have a useful logfile, including the matching or
+ * non-matching pattern.
+ *
+ */
+struct regex {
+    const char *pattern;
+    pcre *regex;
+    pcre_extra *extra;
+};
+
 /******************************************************************************
  * Major types
  *****************************************************************************/
@@ -277,12 +293,11 @@ struct Window {
 };
 
 struct Match {
-    char *title;
-    int title_len;
-    char *application;
-    char *class;
-    char *instance;
-    char *mark;
+    struct regex *title;
+    struct regex *application;
+    struct regex *class;
+    struct regex *instance;
+    struct regex *mark;
     enum {
         M_DONTCHECK = -1,
         M_NODOCK = 0,

include/regex.h

@@ -0,0 +1,28 @@
+/*
+ * vim:ts=4:sw=4:expandtab
+ *
+ */
+#ifndef _REGEX_H
+#define _REGEX_H
+
+/**
+ * Creates a new 'regex' struct containing the given pattern and a PCRE
+ * compiled regular expression. Also, calls pcre_study because this regex will
+ * most likely be used often (like for every new window and on every relevant
+ * property change of existing windows).
+ *
+ * Returns NULL if the pattern could not be compiled into a regular expression
+ * (and ELOGs an appropriate error message).
+ *
+ */
+struct regex *regex_new(const char *pattern);
+
+/**
+ * Checks if the given regular expression matches the given input and returns
+ * true if it does. In either case, it logs the outcome using LOG(), so it will
+ * be visible without any debug loglevel.
+ *
+ */
+bool regex_matches(struct regex *regex, const char *input);
+
+#endif

src/cfgparse.y

@@ -751,12 +751,14 @@ criterion:
     TOK_CLASS '=' STR
     {
         printf("criteria: class = %s\n", $3);
-        current_match.class = $3;
+        current_match.class = regex_new($3);
+        free($3);
     }
     | TOK_INSTANCE '=' STR
     {
         printf("criteria: instance = %s\n", $3);
-        current_match.instance = $3;
+        current_match.instance = regex_new($3);
+        free($3);
     }
     | TOK_CON_ID '=' STR
     {
@@ -791,12 +793,14 @@ criterion:
     | TOK_MARK '=' STR
     {
         printf("criteria: mark = %s\n", $3);
-        current_match.mark = $3;
+        current_match.mark = regex_new($3);
+        free($3);
     }
     | TOK_TITLE '=' STR
     {
         printf("criteria: title = %s\n", $3);
-        current_match.title = $3;
+        current_match.title = regex_new($3);
+        free($3);
     }
     ;
 
@@ -1054,6 +1058,8 @@ workspace_name:
 assign:
     TOKASSIGN window_class STR
     {
+        /* TODO: the assign command also needs some kind of new syntax where we
+         * just use criteria. Then deprecate the old form */
         printf("assignment of %s to *%s*\n", $2, $3);
         char *workspace = $3;
         char *criteria = $2;
@@ -1065,15 +1071,15 @@ assign:
         char *separator = NULL;
         if ((separator = strchr(criteria, '/')) != NULL) {
             *(separator++) = '\0';
-            match->title = sstrdup(separator);
+            match->title = regex_new(separator);
+            printf("  title = %s\n", separator);
+        }
+        if (*criteria != '\0') {
+            match->class = regex_new(criteria);
+            printf("  class = %s\n", criteria);
         }
-        if (*criteria != '\0')
-            match->class = sstrdup(criteria);
         free(criteria);
 
-        printf("  class = %s\n", match->class);
-        printf("  title = %s\n", match->title);
-
         /* Compatibility with older versions: If the assignment target starts
          * with ~, we create the equivalent of:
          *

src/cmdparse.y

@@ -267,10 +267,9 @@ matchend:
 
                 }
             } else if (current_match.mark != NULL && current->con->mark != NULL &&
-                    strcasecmp(current_match.mark, current->con->mark) == 0) {
+                       regex_matches(current_match.mark, current->con->mark)) {
                 printf("match by mark\n");
                 TAILQ_INSERT_TAIL(&owindows, current, owindows);
-
             } else {
                 if (current->con->window == NULL)
                     continue;
@@ -300,12 +299,14 @@ criterion:
     TOK_CLASS '=' STR
     {
         printf("criteria: class = %s\n", $3);
-        current_match.class = $3;
+        current_match.class = regex_new($3);
+        free($3);
     }
     | TOK_INSTANCE '=' STR
     {
         printf("criteria: instance = %s\n", $3);
-        current_match.instance = $3;
+        current_match.instance = regex_new($3);
+        free($3);
     }
     | TOK_CON_ID '=' STR
     {
@@ -340,12 +341,14 @@ criterion:
     | TOK_MARK '=' STR
     {
         printf("criteria: mark = %s\n", $3);
-        current_match.mark = $3;
+        current_match.mark = regex_new($3);
+        free($3);
     }
     | TOK_TITLE '=' STR
     {
         printf("criteria: title = %s\n", $3);
-        current_match.title = $3;
+        current_match.title = regex_new($3);
+        free($3);
     }
     ;
 

src/match.c

@@ -52,16 +52,19 @@ bool match_is_empty(Match *match) {
 void match_copy(Match *dest, Match *src) {
     memcpy(dest, src, sizeof(Match));
 
-#define STRDUP(field) do { \
+/* The DUPLICATE_REGEX macro creates a new regular expression from the
+ * ->pattern of the old one. It therefore does use a little more memory then
+ *  with a refcounting system, but it’s easier this way. */
+#define DUPLICATE_REGEX(field) do { \
     if (src->field != NULL) \
-        dest->field = sstrdup(src->field); \
+        dest->field = regex_new(src->field->pattern); \
 } while (0)
 
-    STRDUP(title);
-    STRDUP(mark);
-    STRDUP(application);
-    STRDUP(class);
-    STRDUP(instance);
+    DUPLICATE_REGEX(title);
+    DUPLICATE_REGEX(mark);
+    DUPLICATE_REGEX(application);
+    DUPLICATE_REGEX(class);
+    DUPLICATE_REGEX(instance);
 }
 
 /*
@@ -71,9 +74,9 @@ void match_copy(Match *dest, Match *src) {
 bool match_matches_window(Match *match, i3Window *window) {
     LOG("checking window %d (%s)\n", window->id, window->class_class);
 
-    /* TODO: pcre, full matching, … */
     if (match->class != NULL) {
-        if (window->class_class != NULL && strcasecmp(match->class, window->class_class) == 0) {
+        if (window->class_class != NULL &&
+            regex_matches(match->class, window->class_class)) {
             LOG("window class matches (%s)\n", window->class_class);
         } else {
             LOG("window class does not match\n");
@@ -82,7 +85,8 @@ bool match_matches_window(Match *match, i3Window *window) {
     }
 
     if (match->instance != NULL) {
-        if (window->class_instance != NULL && strcasecmp(match->instance, window->class_instance) == 0) {
+        if (window->class_instance != NULL &&
+            regex_matches(match->instance, window->class_instance)) {
             LOG("window instance matches (%s)\n", window->class_instance);
         } else {
             LOG("window instance does not match\n");
@@ -99,9 +103,9 @@ bool match_matches_window(Match *match, i3Window *window) {
         }
     }
 
-    /* TODO: pcre match */
     if (match->title != NULL) {
-        if (window->name_json != NULL && strcasecmp(match->title, window->name_json) == 0) {
+        if (window->name_json != NULL &&
+            regex_matches(match->title, window->name_json)) {
             LOG("title matches (%s)\n", window->name_json);
         } else {
             LOG("title does not match\n");

src/regex.c

@@ -0,0 +1,63 @@
+/*
+ * vim:ts=4:sw=4:expandtab
+ *
+ * i3 - an improved dynamic tiling window manager
+ * © 2009-2011 Michael Stapelberg and contributors (see also: LICENSE)
+ *
+ *
+ */
+
+#include "all.h"
+
+/*
+ * Creates a new 'regex' struct containing the given pattern and a PCRE
+ * compiled regular expression. Also, calls pcre_study because this regex will
+ * most likely be used often (like for every new window and on every relevant
+ * property change of existing windows).
+ *
+ * Returns NULL if the pattern could not be compiled into a regular expression
+ * (and ELOGs an appropriate error message).
+ *
+ */
+struct regex *regex_new(const char *pattern) {
+    const char *error;
+    int offset;
+
+    struct regex *re = scalloc(sizeof(struct regex));
+    re->pattern = sstrdup(pattern);
+    if (!(re->regex = pcre_compile(pattern, 0, &error, &offset, NULL))) {
+        ELOG("PCRE regular expression compilation failed at %d: %s",
+             offset, error);
+        return NULL;
+    }
+    re->extra = pcre_study(re->regex, 0, &error);
+    return re;
+}
+
+/*
+ * Checks if the given regular expression matches the given input and returns
+ * true if it does. In either case, it logs the outcome using LOG(), so it will
+ * be visible without any debug loglevel.
+ *
+ */
+bool regex_matches(struct regex *regex, const char *input) {
+    int rc;
+
+    /* TODO: is strlen(input) correct for UTF-8 matching? */
+    /* TODO: enable UTF-8 */
+    if ((rc = pcre_exec(regex->regex, regex->extra, input, strlen(input), 0, 0, NULL, 0)) == 0) {
+        LOG("Regular expression \"%s\" matches \"%s\"\n",
+            regex->pattern, input);
+        return true;
+    }
+
+    if (rc == PCRE_ERROR_NOMATCH) {
+        LOG("Regular expression \"%s\" does not match \"%s\"\n",
+            regex->pattern, input);
+        return false;
+    }
+
+    /* TODO: handle the other error codes */
+    LOG("PCRE error\n");
+    return false;
+}