allow regexp interruption (e.g. with Ctrl-C in the REPL)

Author: Fabrice Bellard

libregexp.c

@@ -53,6 +53,9 @@ typedef enum {
 
 #define CAPTURE_COUNT_MAX 255
 #define STACK_SIZE_MAX 255
+/* must be large enough to have a negligible runtime cost and small
+   enough to call the interrupt callback often. */
+#define INTERRUPT_COUNTER_INIT 10000
 
 /* unicode code points */
 #define CP_LS   0x2028
@@ -1992,6 +1995,7 @@ typedef struct {
     bool multi_line;
     bool ignore_case;
     bool is_unicode;
+    int interrupt_counter;
     void *opaque; /* used for stack overflow check */
 
     size_t state_size;
@@ -2038,7 +2042,17 @@ static int push_state(REExecContext *s,
     return 0;
 }
 
-/* return 1 if match, 0 if not match or -1 if error. */
+static int lre_poll_timeout(REExecContext *s)
+{
+    if (unlikely(--s->interrupt_counter <= 0)) {
+        s->interrupt_counter = INTERRUPT_COUNTER_INIT;
+        if (lre_check_timeout(s->opaque))
+            return LRE_RET_TIMEOUT;
+    }
+    return 0;
+}
+
+/* return 1 if match, 0 if not match or < 0 if error. */
 static intptr_t lre_exec_backtrack(REExecContext *s, uint8_t **capture,
                                    StackInt *stack, int stack_len,
                                    const uint8_t *pc, const uint8_t *cptr,
@@ -2069,6 +2083,8 @@ static intptr_t lre_exec_backtrack(REExecContext *s, uint8_t **capture,
                 ret = 0;
             recurse:
                 for(;;) {
+                    if (lre_poll_timeout(s))
+                        return LRE_RET_TIMEOUT;
                     if (s->state_stack_len == 0)
                         return ret;
                     rs = (REExecState *)(s->state_stack +
@@ -2162,7 +2178,7 @@ static intptr_t lre_exec_backtrack(REExecContext *s, uint8_t **capture,
                 ret = push_state(s, capture, stack, stack_len,
                                  pc1, cptr, RE_EXEC_STATE_SPLIT, 0);
                 if (ret < 0)
-                    return -1;
+                    return LRE_RET_MEMORY_ERROR;
                 break;
             }
         case REOP_lookahead:
@@ -2174,12 +2190,14 @@ static intptr_t lre_exec_backtrack(REExecContext *s, uint8_t **capture,
                              RE_EXEC_STATE_LOOKAHEAD + opcode - REOP_lookahead,
                              0);
             if (ret < 0)
-                return -1;
+                return LRE_RET_MEMORY_ERROR;
             break;
 
         case REOP_goto:
             val = get_u32(pc);
             pc += 4 + (int)val;
+            if (lre_poll_timeout(s))
+                return LRE_RET_TIMEOUT;
             break;
         case REOP_line_start:
             if (cptr == s->cbuf)
@@ -2244,6 +2262,8 @@ static intptr_t lre_exec_backtrack(REExecContext *s, uint8_t **capture,
             pc += 4;
             if (--stack[stack_len - 1] != 0) {
                 pc += (int)val;
+                if (lre_poll_timeout(s))
+                    return LRE_RET_TIMEOUT;
             }
             break;
         case REOP_push_char_pos:
@@ -2418,9 +2438,12 @@ static intptr_t lre_exec_backtrack(REExecContext *s, uint8_t **capture,
 
                 q = 0;
                 for(;;) {
+                    if (lre_poll_timeout(s))
+                        return LRE_RET_TIMEOUT;
                     res = lre_exec_backtrack(s, capture, stack, stack_len,
                                              pc1, cptr, true);
-                    if (res == -1)
+                    if (res == LRE_RET_MEMORY_ERROR ||
+                        res == LRE_RET_TIMEOUT)
                         return res;
                     if (!res)
                         break;
@@ -2438,7 +2461,7 @@ static intptr_t lre_exec_backtrack(REExecContext *s, uint8_t **capture,
                                      RE_EXEC_STATE_GREEDY_QUANT,
                                      q - quant_min);
                     if (ret < 0)
-                        return -1;
+                        return LRE_RET_MEMORY_ERROR;
                 }
             }
             break;
@@ -2448,7 +2471,7 @@ static intptr_t lre_exec_backtrack(REExecContext *s, uint8_t **capture,
     }
 }
 
-/* Return 1 if match, 0 if not match or -1 if error. cindex is the
+/* Return 1 if match, 0 if not match or < 0 if error (see LRE_RET_x). cindex is the
    starting position of the match and must be such as 0 <= cindex <=
    clen. */
 int lre_exec(uint8_t **capture,
@@ -2470,6 +2493,7 @@ int lre_exec(uint8_t **capture,
     s->cbuf_type = cbuf_type;
     if (s->cbuf_type == 1 && s->is_unicode)
         s->cbuf_type = 2;
+    s->interrupt_counter = INTERRUPT_COUNTER_INIT;
     s->opaque = opaque;
 
     s->state_size = sizeof(REExecState) +

libregexp.h

@@ -43,6 +43,9 @@ extern "C" {
 #define LRE_FLAG_NAMED_GROUPS (1 << 7) /* named groups are present in the regexp */
 #define LRE_FLAG_UNICODE_SETS (1 << 8)
 
+#define LRE_RET_MEMORY_ERROR (-1)
+#define LRE_RET_TIMEOUT      (-2)
+
 uint8_t *lre_compile(int *plen, char *error_msg, int error_msg_size,
                      const char *buf, size_t buf_len, int re_flags,
                      void *opaque);
@@ -60,6 +63,8 @@ void lre_byte_swap(uint8_t *buf, size_t len, bool is_byte_swapped);
 
 /* must be provided by the user */
 bool lre_check_stack_overflow(void *opaque, size_t alloca_size);
+/* must be provided by the user, return non zero if time out */
+int lre_check_timeout(void *opaque);
 void *lre_realloc(void *opaque, void *ptr, size_t size);
 
 /* JS identifier test */

quickjs.c

@@ -7111,15 +7111,19 @@ static JSValue JS_ThrowTypeErrorInvalidClass(JSContext *ctx, int class_id)
     return JS_ThrowTypeErrorAtom(ctx, "%s object expected", name);
 }
 
+static void JS_ThrowInterrupted(JSContext *ctx)
+{
+    JS_ThrowInternalError(ctx, "interrupted");
+    JS_SetUncatchableError(ctx, ctx->rt->current_exception);
+}
+
 static no_inline __exception int __js_poll_interrupts(JSContext *ctx)
 {
     JSRuntime *rt = ctx->rt;
     ctx->interrupt_counter = JS_INTERRUPT_COUNTER_INIT;
     if (rt->interrupt_handler) {
         if (rt->interrupt_handler(rt, rt->interrupt_opaque)) {
-            /* XXX: should set a specific flag to avoid catching */
-            JS_ThrowInternalError(ctx, "interrupted");
-            js_set_uncatchable_error(ctx, ctx->rt->current_exception, true);
+            JS_ThrowInterrupted(ctx);
             return -1;
         }
     }
@@ -43878,6 +43882,14 @@ bool lre_check_stack_overflow(void *opaque, size_t alloca_size)
     return js_check_stack_overflow(ctx->rt, alloca_size);
 }
 
+int lre_check_timeout(void *opaque)
+{
+    JSContext *ctx = opaque;
+    JSRuntime *rt = ctx->rt;
+    return (rt->interrupt_handler &&
+            rt->interrupt_handler(rt, rt->interrupt_opaque));
+}
+
 void *lre_realloc(void *opaque, void *ptr, size_t size)
 {
     JSContext *ctx = opaque;
@@ -43992,7 +44004,11 @@ static JSValue js_regexp_exec(JSContext *ctx, JSValueConst this_val,
                     goto fail;
             }
         } else {
-            JS_ThrowInternalError(ctx, "out of memory in regexp execution");
+            if (rc == LRE_RET_TIMEOUT) {
+                JS_ThrowInterrupted(ctx);
+            } else {
+                JS_ThrowInternalError(ctx, "out of memory in regexp execution");
+            }
             goto fail;
         }
     } else {
@@ -44187,7 +44203,11 @@ static JSValue JS_RegExpDelete(JSContext *ctx, JSValueConst this_val, JSValue ar
                         goto fail;
                 }
             } else {
-                JS_ThrowInternalError(ctx, "out of memory in regexp execution");
+                if (ret == LRE_RET_TIMEOUT) {
+                    JS_ThrowInterrupted(ctx);
+                } else {
+                    JS_ThrowInternalError(ctx, "out of memory in regexp execution");
+                }
                 goto fail;
             }
             break;