Added support for JIT-compilation of patterns

CHANGES.md

@@ -1,3 +1,11 @@
+### 7.5.0 (2021-07-22)
+
+  * Added support for JIT-compilation of patterns, which can greatly improve
+    matching performance for most patterns.  Users need to explicitly add
+    the `jit_compile` flag to request JIT-compilation when creating regular
+    expressions.
+
+
 ### 7.4.6 (2020-08-04)
 
   * Removed mistakenly kept base library configuration dependency.

src/pcre.ml

@@ -242,7 +242,7 @@ type chtables
 external maketables : unit -> chtables = "pcre_maketables_stub"
 
 (*  Internal use only! *)
-external pcre_study : regexp -> unit = "pcre_study_stub"
+external pcre_study : regexp -> jit_compile : bool -> unit = "pcre_study_stub"
 
 external compile : (icflag  [@untagged]) -> chtables option -> string -> regexp
   = "pcre_compile_stub_bc" "pcre_compile_stub"
@@ -265,14 +265,16 @@ external set_imp_match_limit_recursion : regexp -> (int [@untagged]) -> regexp
   [@@noalloc]
 
 let regexp
-      ?(study = true) ?limit ?limit_recursion
+      ?(study = true)
+      ?(jit_compile = false)
+      ?limit ?limit_recursion
       ?(iflags = 0) ?flags ?chtables pat =
   let rex =
     match flags with
     | Some flag_list -> compile (cflags flag_list) chtables pat
     | _ -> compile iflags chtables pat
   in
-  if study then pcre_study rex;
+  if study then pcre_study ~jit_compile rex;
   let rex =
     match limit with
     | None -> rex
@@ -283,7 +285,8 @@ let regexp
   | Some lim -> set_imp_match_limit_recursion rex lim
 
 let regexp_or
-      ?study ?limit ?limit_recursion ?(iflags = 0) ?flags ?chtables pats =
+      ?study ?jit_compile ?limit ?limit_recursion ?(iflags = 0)
+      ?flags ?chtables pats =
   let check pat =
     try ignore (regexp ~study:false ~iflags ?flags ?chtables pat)
     with Error error -> raise (Regexp_or (pat, error))
@@ -293,7 +296,8 @@ let regexp_or
     let cnv pat = "(?:" ^ pat ^ ")" in
     String.concat "|" (List.rev (List.rev_map cnv pats))
   in
-  regexp ?study ?limit ?limit_recursion ~iflags ?flags ?chtables big_pat
+  regexp ?study ?jit_compile ?limit ?limit_recursion ~iflags ?flags ?chtables
+    big_pat
 
 let bytes_unsafe_blit_string str str_ofs bts bts_ofs len =
   let str_bts = Bytes.unsafe_of_string str in
@@ -324,7 +328,7 @@ let quote s =
 (* Matching of patterns and subpattern extraction *)
 
 (* Default regular expression when none is provided by the user *)
-let def_rex = regexp "\\s+"
+let def_rex = regexp ~jit_compile:true "\\s+"
 
 type substrings = string * int array
 

src/pcre.mli

@@ -231,21 +231,24 @@ val maketables : unit -> chtables
 
 val regexp :
   ?study : bool ->
+  ?jit_compile : bool ->
   ?limit : int ->
   ?limit_recursion : int ->
   ?iflags : icflag ->
   ?flags : cflag list ->
   ?chtables : chtables ->
   string -> regexp
-(** [regexp ?study ?limit ?limit_recursion ?iflags ?flags ?chtables pattern]
-    compiles [pattern] with [flags] when given, with [iflags] otherwise, and
-    with char tables [chtables]. If [study] is true, then the resulting regular
-    expression will be studied. If [limit] is specified, this sets a limit to
-    the amount of recursion and backtracking (only lower than the builtin
-    default!). If this limit is exceeded, [MatchLimit] will be raised during
-    matching.
+(** [regexp ?jit_compile ?study ?limit ?limit_recursion ?iflags ?flags
+    ?chtables pattern] compiles [pattern] with [flags] when given, with
+    [iflags] otherwise, and with char tables [chtables].  If [study] is true,
+    then the resulting regular expression will be studied.  If [jit_compile]
+    is true, studying will also perform JIT-compilation of the pattern.
+    [If [limit] is specified, this sets a limit to the amount of recursion
+    and backtracking (only lower than the builtin default!).  If this limit
+    is exceeded, [MatchLimit] will be raised during matching.
 
     @param study default = true
+    @param jit_compile default = false
     @param limit default = no extra limit other than default
     @param limit_recursion default = no extra limit_recursion other than default
     @param iflags default = no extra flags
@@ -261,6 +264,7 @@ val regexp :
 
 val regexp_or :
   ?study : bool ->
+  ?jit_compile : bool ->
   ?limit : int ->
   ?limit_recursion : int ->
   ?iflags : icflag ->
@@ -269,7 +273,8 @@ val regexp_or :
   string list -> regexp
 (** [regexp_or ?study ?limit ?limit_recursion ?iflags ?flags ?chtables patterns]
     like {!regexp}, but combines [patterns] as alternatives (or-patterns) into
-    one regular expression. *)
+    one regular expression.
+*)
 
 val quote : string -> string
 (** [quote str] @return the quoted string of [str]. *)

src/pcre_stubs.c

@@ -56,13 +56,18 @@
 
 /* Error codes as defined for pcre 7.9, undefined in pcre 4.5 */
 #ifndef PCRE_ERROR_PARTIAL
-#define PCRE_ERROR_PARTIAL        (-12)
+# define PCRE_ERROR_PARTIAL        (-12)
 #endif
 #ifndef PCRE_ERROR_BADPARTIAL
-#define PCRE_ERROR_BADPARTIAL     (-13)
+# define PCRE_ERROR_BADPARTIAL     (-13)
 #endif
 #ifndef PCRE_ERROR_RECURSIONLIMIT
-#define PCRE_ERROR_RECURSIONLIMIT (-21)
+# define PCRE_ERROR_RECURSIONLIMIT (-21)
+#endif
+
+/* Make sure to define JIT-compilation flag appropriately if unsupported */
+#ifndef PCRE_STUDY_JIT_COMPILE
+# define PCRE_STUDY_JIT_COMPILE 0
 #endif
 
 typedef const unsigned char *chartables;  /* Type of chartable sets */
@@ -351,12 +356,13 @@ CAMLprim value pcre_compile_stub_bc(value v_opt, value v_tables, value v_pat)
 
 
 /* Studies a regexp */
-CAMLprim value pcre_study_stub(value v_rex)
+CAMLprim value pcre_study_stub(value v_rex, value v_jit_compile)
 {
   /* If it has not yet been studied */
   if (! get_studied(v_rex)) {
     const char *error = NULL;
-    pcre_extra *extra = pcre_study(get_rex(v_rex), 0, &error);
+    int flags = Bool_val(v_jit_compile) ? PCRE_STUDY_JIT_COMPILE : 0;
+    pcre_extra *extra = pcre_study(get_rex(v_rex), flags, &error);
     if (error != NULL) caml_invalid_argument((char *) error);
     set_extra(v_rex, extra);
     set_studied(v_rex, 1);