gri3-wm/i3-config-wizard/cfgparse.y

162 lines
4.1 KiB
Plaintext
Raw Normal View History

%{
/*
* vim:ts=4:sw=4:expandtab
*
*/
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <fcntl.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <X11/Xlib.h>
extern Display *dpy;
struct context {
int line_number;
char *line_copy;
char *compact_error;
/* These are the same as in YYLTYPE */
int first_column;
int last_column;
char *result;
};
typedef struct yy_buffer_state *YY_BUFFER_STATE;
extern int yylex(struct context *context);
extern int yyparse(void);
extern FILE *yyin;
YY_BUFFER_STATE yy_scan_string(const char *);
static struct context *context;
/* We dont need yydebug for now, as we got decent error messages using
* yyerror(). Should you ever want to extend the parser, it might be handy
* to just comment it in again, so it stays here. */
//int yydebug = 1;
void yyerror(const char *error_message) {
fprintf(stderr, "\n");
fprintf(stderr, "CONFIG: %s\n", error_message);
fprintf(stderr, "CONFIG: line %d:\n",
context->line_number);
fprintf(stderr, "CONFIG: %s\n", context->line_copy);
fprintf(stderr, "CONFIG: ");
for (int c = 1; c <= context->last_column; c++)
if (c >= context->first_column)
fprintf(stderr, "^");
else fprintf(stderr, " ");
fprintf(stderr, "\n");
fprintf(stderr, "\n");
}
int yywrap() {
return 1;
}
char *rewrite_binding(const char *bindingline) {
char *result = NULL;
context = calloc(sizeof(struct context), 1);
yy_scan_string(bindingline);
if (yyparse() != 0) {
fprintf(stderr, "Could not parse configfile\n");
exit(1);
}
result = context->result;
if (context->line_copy)
free(context->line_copy);
free(context);
return result;
}
/* XXX: does not work for combinations of modifiers yet */
static char *modifier_to_string(int modifiers) {
//printf("should convert %d to string\n", modifiers);
if (modifiers == (1 << 3))
return strdup("$mod+");
else if (modifiers == ((1 << 3) | (1 << 0)))
return strdup("$mod+Shift+");
else if (modifiers == (1 << 9))
return strdup("$mod+");
else if (modifiers == ((1 << 9) | (1 << 0)))
return strdup("$mod+Shift+");
else if (modifiers == (1 << 0))
return strdup("Shift+");
else return strdup("");
}
%}
%error-verbose
%lex-param { struct context *context }
%union {
int number;
char *string;
}
%token <number>NUMBER "<number>"
%token <string>STR "<string>"
%token TOKBINDCODE
%token TOKMODVAR "$mod"
%token MODIFIER "<modifier>"
%token TOKCONTROL "control"
%token TOKSHIFT "shift"
%token WHITESPACE "<whitespace>"
%%
lines: /* empty */
| lines WHITESPACE bindcode
| lines error
| lines bindcode
;
bindcode:
TOKBINDCODE WHITESPACE binding_modifiers NUMBER WHITESPACE STR
{
//printf("\tFound keycode binding mod%d with key %d and command %s\n", $<number>3, $4, $<string>6);
int level = 0;
if (($<number>3 & (1 << 0))) {
/* When shift is included, we really need to use the second-level
* symbol (upper-case). The lower-case symbol could be on a
* different key than the upper-case one (unlikely for letters, but
* more likely for special characters). */
level = 1;
}
KeySym sym = XKeycodeToKeysym(dpy, $4, level);
char *str = XKeysymToString(sym);
char *modifiers = modifier_to_string($<number>3);
// TODO: modifier to string
asprintf(&(context->result), "bindsym %s%s %s\n", modifiers, str, $<string>6);
free(modifiers);
}
;
binding_modifiers:
/* NULL */ { $<number>$ = 0; }
| binding_modifier
| binding_modifiers '+' binding_modifier { $<number>$ = $<number>1 | $<number>3; }
| binding_modifiers '+' { $<number>$ = $<number>1; }
;
binding_modifier:
MODIFIER { $<number>$ = $<number>1; }
| TOKMODVAR { $<number>$ = $<number>1; }
| TOKCONTROL { $<number>$ = (1 << 2); }
| TOKSHIFT { $<number>$ = (1 << 0); }
;