1 (* Bitstring syntax extension.
2 * Copyright (C) 2008 Red Hat Inc., Richard W.M. Jones
4 * This library is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2 of the License, or (at your option) any later version,
8 * with the OCaml linking exception described in COPYING.LIB.
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
29 module P = Bitstring_persistent
31 (* If this is true then we emit some debugging code which can
32 * be useful to tell what is happening during matches. You
33 * also need to do 'Bitstring.debug := true' in your main program.
35 * If this is false then no extra debugging code is emitted.
39 (* Hashtable storing named persistent patterns. *)
40 let pattern_hash : (string, P.pattern) Hashtbl.t = Hashtbl.create 13
42 let locfail _loc msg = Loc.raise _loc (Failure msg)
44 (* Work out if an expression is an integer constant.
46 * Returns [Some i] if so (where i is the integer value), else [None].
48 * Fairly simplistic algorithm: we can only detect simple constant
49 * expressions such as [k], [k+c], [k-c] etc.
51 let rec expr_is_constant = function
52 | <:expr< $int:i$ >> -> (* Literal integer constant. *)
53 Some (int_of_string i)
54 | <:expr< $lid:op$ $a$ $b$ >> ->
55 (match expr_is_constant a, expr_is_constant b with
56 | Some a, Some b -> (* Integer binary operations. *)
57 let ops = ["+", (+); "-", (-); "*", ( * ); "/", (/);
58 (* NB: explicit fun .. -> is necessary here to work
59 * around a camlp4 bug in OCaml 3.10.0.
61 "land", (fun a b -> a land b);
62 "lor", (fun a b -> a lor b);
63 "lxor", (fun a b -> a lxor b);
64 "lsl", (fun a b -> a lsl b);
65 "lsr", (fun a b -> a lsr b);
66 "asr", (fun a b -> a asr b);
67 "mod", (fun a b -> a mod b)] in
68 (try Some ((List.assoc op ops) a b) with Not_found -> None)
72 (* Generate a fresh, unique symbol each time called. *)
77 sprintf "__pabitstring_%s_%d" name i
79 (* Used to keep track of which qualifiers we've seen in parse_field. *)
81 endian_set : bool; signed_set : bool; type_set : bool;
82 offset_set : bool; check_set : bool; bind_set : bool;
83 save_offset_to_set : bool;
86 endian_set = false; signed_set = false; type_set = false;
87 offset_set = false; check_set = false; bind_set = false;
88 save_offset_to_set = false
91 (* Deal with the qualifiers which appear for a field of both types. *)
92 let parse_field _loc field qs =
93 let fail = locfail _loc in
97 | None -> noneset, field
99 let check already_set msg = if already_set then fail msg in
100 let apply_qualifier (whatset, field) =
102 | "endian", Some expr ->
103 check whatset.endian_set "an endian flag has been set already";
104 let field = P.set_endian_expr field expr in
105 { whatset with endian_set = true }, field
107 fail "qualifier 'endian' should be followed by an expression"
108 | "offset", Some expr ->
109 check whatset.offset_set "an offset has been set already";
110 let field = P.set_offset field expr in
111 { whatset with offset_set = true }, field
113 fail "qualifier 'offset' should be followed by an expression"
114 | "check", Some expr ->
115 check whatset.check_set "a check-qualifier has been set already";
116 let field = P.set_check field expr in
117 { whatset with check_set = true }, field
119 fail "qualifier 'check' should be followed by an expression"
120 | "bind", Some expr ->
121 check whatset.bind_set "a bind expression has been set already";
122 let field = P.set_bind field expr in
123 { whatset with bind_set = true }, field
125 fail "qualifier 'bind' should be followed by an expression"
126 | "save_offset_to", Some expr (* XXX should be a pattern *) ->
127 check whatset.save_offset_to_set
128 "a save_offset_to-qualifier has been set already";
131 | <:expr< $lid:id$ >> -> id
133 failwith "pa_bitstring: internal error: save_offset_to only supports simple identifiers at the moment. In future we should support full patterns." in
134 let field = P.set_save_offset_to_lident field id in
135 { whatset with save_offset_to_set = true }, field
136 | "save_offset_to", None ->
137 fail "qualifier 'save_offset_to' should be followed by a binding expression"
139 fail (s ^ ": unknown qualifier, or qualifier should not be followed by an expression")
141 let endian_quals = ["bigendian", BigEndian;
142 "littleendian", LittleEndian;
143 "nativeendian", NativeEndian] in
144 let sign_quals = ["signed", true; "unsigned", false] in
145 let type_quals = ["int", P.set_type_int;
146 "string", P.set_type_string;
147 "bitstring", P.set_type_bitstring] in
148 if List.mem_assoc qual endian_quals then (
149 check whatset.endian_set "an endian flag has been set already";
150 let field = P.set_endian field (List.assoc qual endian_quals) in
151 { whatset with endian_set = true }, field
152 ) else if List.mem_assoc qual sign_quals then (
153 check whatset.signed_set "a signed flag has been set already";
154 let field = P.set_signed field (List.assoc qual sign_quals) in
155 { whatset with signed_set = true }, field
156 ) else if List.mem_assoc qual type_quals then (
157 check whatset.type_set "a type flag has been set already";
158 let field = (List.assoc qual type_quals) field in
159 { whatset with type_set = true }, field
161 fail (qual ^ ": unknown qualifier, or qualifier should be followed by an expression") in
162 List.fold_left apply_qualifier (noneset, field) qs in
164 (* If type is set to string or bitstring then endianness and
165 * signedness qualifiers are meaningless and must not be set.
168 let t = P.get_type field in
169 if (t = P.Bitstring || t = P.String) &&
170 (whatset.endian_set || whatset.signed_set) then
171 fail "string types and endian or signed qualifiers cannot be mixed" in
173 (* Default endianness, signedness, type if not set already. *)
175 if whatset.endian_set then field else P.set_endian field BigEndian in
177 if whatset.signed_set then field else P.set_signed field false in
179 if whatset.type_set then field else P.set_type_int field in
183 (* Choose the right constructor function. *)
184 let build_bitstring_call _loc funcname length endian signed =
185 match length, endian, signed with
186 (* XXX The meaning of signed/unsigned breaks down at
187 * 31, 32, 63 and 64 bits.
189 | (Some 1, _, _) -> <:expr< Bitstring.$lid:funcname ^ "_bit"$ >>
190 | (Some (2|3|4|5|6|7|8), _, sign) ->
191 let call = Printf.sprintf "%s_char_%s"
192 funcname (if sign then "signed" else "unsigned") in
193 <:expr< Bitstring.$lid:call$ >>
194 | (len, endian, signed) ->
195 let t = match len with
196 | Some i when i <= 31 -> "int"
199 let sign = if signed then "signed" else "unsigned" in
201 | P.ConstantEndian constant ->
202 let endianness = match constant with
204 | LittleEndian -> "le"
205 | NativeEndian -> "ne" in
206 let call = Printf.sprintf "%s_%s_%s_%s"
207 funcname t endianness sign in
208 <:expr< Bitstring.$lid:call$ >>
209 | P.EndianExpr expr ->
210 let call = Printf.sprintf "%s_%s_%s_%s"
211 funcname t "ee" sign in
212 <:expr< Bitstring.$lid:call$ $expr$ >>
214 (* Generate the code for a constructor, ie. 'BITSTRING ...'. *)
215 let output_constructor _loc fields =
216 (* This function makes code to raise a Bitstring.Construct_failure exception
217 * containing a message and the current _loc context.
218 * (Thanks to Bluestorm for suggesting this).
220 let construct_failure _loc msg =
222 Bitstring.Construct_failure
224 $`str:Loc.file_name _loc$,
225 $`int:Loc.start_line _loc$,
226 $`int:Loc.start_off _loc - Loc.start_bol _loc$)
229 let raise_construct_failure _loc msg =
230 <:expr< raise $construct_failure _loc msg$ >>
233 (* Bitstrings are created like the 'Buffer' module (in fact, using
234 * the Buffer module), by appending snippets to a growing buffer.
235 * This is reasonably efficient and avoids a lot of garbage.
237 let buffer = gensym "buffer" in
239 (* General exception which is raised inside the constructor functions
240 * when an int expression is out of range at runtime.
242 let exn = gensym "exn" in
243 let exn_used = ref false in
245 (* Convert each field to a simple bitstring-generating expression. *)
246 let fields = List.map (
248 let fexpr = P.get_expr field in
249 let flen = P.get_length field in
250 let endian = P.get_endian field in
251 let signed = P.get_signed field in
252 let t = P.get_type field in
253 let _loc = P.get_location field in
255 let fail = locfail _loc in
257 (* offset(), check(), bind(), save_offset_to() not supported in
260 * Implementation of forward-only offsets is fairly
261 * straightforward: we would need to just calculate the length of
262 * padding here and add it to what has been constructed. For
263 * general offsets, including going backwards, that would require
264 * a rethink in how we construct bitstrings.
266 if P.get_offset field <> None then
267 fail "offset expressions are not supported in BITSTRING constructors";
268 if P.get_check field <> None then
269 fail "check expressions are not supported in BITSTRING constructors";
270 if P.get_bind field <> None then
271 fail "bind expressions are not supported in BITSTRING constructors";
272 if P.get_save_offset_to field <> None then
273 fail "save_offset_to is not supported in BITSTRING constructors";
275 (* Is flen an integer constant? If so, what is it? This
276 * is very simple-minded and only detects simple constants.
278 let flen_is_const = expr_is_constant flen in
280 let int_construct_const (i, endian, signed) =
281 build_bitstring_call _loc "construct" (Some i) endian signed in
282 let int_construct (endian, signed) =
283 build_bitstring_call _loc "construct" None endian signed in
286 match t, flen_is_const with
287 (* Common case: int field, constant flen.
289 * Range checks are done inside the construction function
290 * because that's a lot simpler w.r.t. types. It might
291 * be better to move them here. XXX
293 | P.Int, Some i when i > 0 && i <= 64 ->
294 let construct_fn = int_construct_const (i,endian,signed) in
298 $construct_fn$ $lid:buffer$ $fexpr$ $`int:i$ $lid:exn$
302 fail "length of int field must be [1..64]"
304 (* Int field, non-constant length. We need to perform a runtime
305 * test to ensure the length is [1..64].
307 * Range checks are done inside the construction function
308 * because that's a lot simpler w.r.t. types. It might
309 * be better to move them here. XXX
312 let construct_fn = int_construct (endian,signed) in
316 if $flen$ >= 1 && $flen$ <= 64 then
317 $construct_fn$ $lid:buffer$ $fexpr$ $flen$ $lid:exn$
319 $raise_construct_failure _loc "length of int field must be [1..64]"$
322 (* String, constant length > 0, must be a multiple of 8. *)
323 | P.String, Some i when i > 0 && i land 7 = 0 ->
324 let bs = gensym "bs" in
327 let $lid:bs$ = $fexpr$ in
328 if String.length $lid:bs$ = $`int:j$ then
329 Bitstring.construct_string $lid:buffer$ $lid:bs$
331 $raise_construct_failure _loc "length of string does not match declaration"$
334 (* String, constant length -1, means variable length string
337 | P.String, Some (-1) ->
338 <:expr< Bitstring.construct_string $lid:buffer$ $fexpr$ >>
340 (* String, constant length = 0 is probably an error, and so is
343 | P.String, Some _ ->
344 fail "length of string must be > 0 and a multiple of 8, or the special value -1"
346 (* String, non-constant length.
347 * We check at runtime that the length is > 0, a multiple of 8,
348 * and matches the declared length.
351 let bslen = gensym "bslen" in
352 let bs = gensym "bs" in
354 let $lid:bslen$ = $flen$ in
355 if $lid:bslen$ > 0 then (
356 if $lid:bslen$ land 7 = 0 then (
357 let $lid:bs$ = $fexpr$ in
358 if String.length $lid:bs$ = ($lid:bslen$ lsr 3) then
359 Bitstring.construct_string $lid:buffer$ $lid:bs$
361 $raise_construct_failure _loc "length of string does not match declaration"$
363 $raise_construct_failure _loc "length of string must be a multiple of 8"$
365 $raise_construct_failure _loc "length of string must be > 0"$
368 (* Bitstring, constant length >= 0. *)
369 | P.Bitstring, Some i when i >= 0 ->
370 let bs = gensym "bs" in
372 let $lid:bs$ = $fexpr$ in
373 if Bitstring.bitstring_length $lid:bs$ = $`int:i$ then
374 Bitstring.construct_bitstring $lid:buffer$ $lid:bs$
376 $raise_construct_failure _loc "length of bitstring does not match declaration"$
379 (* Bitstring, constant length -1, means variable length bitstring
382 | P.Bitstring, Some (-1) ->
383 <:expr< Bitstring.construct_bitstring $lid:buffer$ $fexpr$ >>
385 (* Bitstring, constant length < -1 is an error. *)
386 | P.Bitstring, Some _ ->
387 fail "length of bitstring must be >= 0 or the special value -1"
389 (* Bitstring, non-constant length.
390 * We check at runtime that the length is >= 0 and matches
391 * the declared length.
393 | P.Bitstring, None ->
394 let bslen = gensym "bslen" in
395 let bs = gensym "bs" in
397 let $lid:bslen$ = $flen$ in
398 if $lid:bslen$ >= 0 then (
399 let $lid:bs$ = $fexpr$ in
400 if Bitstring.bitstring_length $lid:bs$ = $lid:bslen$ then
401 Bitstring.construct_bitstring $lid:buffer$ $lid:bs$
403 $raise_construct_failure _loc "length of bitstring does not match declaration"$
405 $raise_construct_failure _loc "length of bitstring must be > 0"$
410 (* Create the final bitstring. Start by creating an empty buffer
411 * and then evaluate each expression above in turn which will
412 * append some more to the bitstring buffer. Finally extract
415 * XXX We almost have enough information to be able to guess
416 * a good initial size for the buffer.
420 | [] -> <:expr< [] >>
421 | h::t -> List.fold_left (fun h t -> <:expr< $h$; $t$ >>) h t in
425 let $lid:buffer$ = Bitstring.Buffer.create () in
427 Bitstring.Buffer.contents $lid:buffer$
432 let $lid:exn$ = $construct_failure _loc "value out of range"$ in
438 (* Generate the code for a bitmatch statement. '_loc' is the
439 * location, 'bs' is the bitstring parameter, 'cases' are
440 * the list of cases to test against.
442 let output_bitmatch _loc bs cases =
443 (* These symbols are used through the generated code to record our
444 * current position within the bitstring:
446 * data - original bitstring data (string, never changes)
448 * off - current offset within data (int, increments as we move through
450 * len - current remaining length within data (int, decrements as
451 * we move through the bitstring)
453 * original_off - saved offset at the start of the match (never changes)
454 * original_len - saved length at the start of the match (never changes)
456 let data = gensym "data"
457 and off = gensym "off"
458 and len = gensym "len"
459 and original_off = gensym "original_off"
460 and original_len = gensym "original_len"
461 (* This is where the result will be stored (a reference). *)
462 and result = gensym "result" in
464 (* This generates the field extraction code for each
465 * field in a single case. There must be enough remaining data
466 * in the bitstring to satisfy the field.
468 * As we go through the fields, symbols 'data', 'off' and 'len'
469 * track our position and remaining length in the bitstring.
471 * The whole thing is a lot of nested 'if'/'match' statements.
472 * Code is generated from the inner-most (last) field outwards.
474 let rec output_field_extraction inner = function
477 let fpatt = P.get_patt field in
478 let flen = P.get_length field in
479 let endian = P.get_endian field in
480 let signed = P.get_signed field in
481 let t = P.get_type field in
482 let _loc = P.get_location field in
484 let fail = locfail _loc in
486 (* Is flen (field len) an integer constant? If so, what is it?
487 * This will be [Some i] if it's a constant or [None] if it's
488 * non-constant or we couldn't determine.
490 let flen_is_const = expr_is_constant flen in
492 (* Surround the inner expression by check and bind clauses, so:
496 * where the check and bind are switched on only if they are
497 * present in the field. (In the common case when neither
498 * clause is present, expr = inner). Note the order of the
499 * check & bind is visible to the user and defined in the
500 * documentation, so it must not change.
504 match P.get_bind field with
507 <:expr< let $fpatt$ = $bind_expr$ in $expr$ >> in
509 match P.get_check field with
512 <:expr< if $check_expr$ then $expr$ >> in
514 (* Now build the code which matches a field. *)
515 let int_extract_const (i, endian, signed) =
516 build_bitstring_call _loc "extract" (Some i) endian signed in
517 let int_extract (endian, signed) =
518 build_bitstring_call _loc "extract" None endian signed in
521 match t, flen_is_const with
522 (* Common case: int field, constant flen *)
523 | P.Int, Some i when i > 0 && i <= 64 ->
524 let extract_fn = int_extract_const (i,endian,signed) in
525 let v = gensym "val" in
527 if $lid:len$ >= $`int:i$ then (
528 let $lid:v$, $lid:off$, $lid:len$ =
529 $extract_fn$ $lid:data$ $lid:off$ $lid:len$ $`int:i$ in
530 match $lid:v$ with $fpatt$ when true -> $expr$ | _ -> ()
535 fail "length of int field must be [1..64]"
537 (* Int field, non-const flen. We have to test the range of
538 * the field at runtime. If outside the range it's a no-match
542 let extract_fn = int_extract (endian,signed) in
543 let v = gensym "val" in
545 if $flen$ >= 1 && $flen$ <= 64 && $flen$ <= $lid:len$ then (
546 let $lid:v$, $lid:off$, $lid:len$ =
547 $extract_fn$ $lid:data$ $lid:off$ $lid:len$ $flen$ in
548 match $lid:v$ with $fpatt$ when true -> $expr$ | _ -> ()
552 (* String, constant flen > 0. *)
553 | P.String, Some i when i > 0 && i land 7 = 0 ->
554 let bs = gensym "bs" in
556 if $lid:len$ >= $`int:i$ then (
557 let $lid:bs$, $lid:off$, $lid:len$ =
558 Bitstring.extract_bitstring $lid:data$ $lid:off$ $lid:len$
560 match Bitstring.string_of_bitstring $lid:bs$ with
561 | $fpatt$ when true -> $expr$
566 (* String, constant flen = -1, means consume all the
569 | P.String, Some i when i = -1 ->
570 let bs = gensym "bs" in
572 let $lid:bs$, $lid:off$, $lid:len$ =
573 Bitstring.extract_remainder $lid:data$ $lid:off$ $lid:len$ in
574 match Bitstring.string_of_bitstring $lid:bs$ with
575 | $fpatt$ when true -> $expr$
579 | P.String, Some _ ->
580 fail "length of string must be > 0 and a multiple of 8, or the special value -1"
582 (* String field, non-const flen. We check the flen is > 0
583 * and a multiple of 8 (-1 is not allowed here), at runtime.
586 let bs = gensym "bs" in
588 if $flen$ >= 0 && $flen$ <= $lid:len$
589 && $flen$ land 7 = 0 then (
590 let $lid:bs$, $lid:off$, $lid:len$ =
591 Bitstring.extract_bitstring
592 $lid:data$ $lid:off$ $lid:len$ $flen$ in
593 match Bitstring.string_of_bitstring $lid:bs$ with
594 | $fpatt$ when true -> $expr$
599 (* Bitstring, constant flen >= 0.
600 * At the moment all we can do is assign the bitstring to an
603 | P.Bitstring, Some i when i >= 0 ->
606 | <:patt< $lid:ident$ >> -> ident
607 | <:patt< _ >> -> "_"
609 fail "cannot compare a bitstring to a constant" in
611 if $lid:len$ >= $`int:i$ then (
612 let $lid:ident$, $lid:off$, $lid:len$ =
613 Bitstring.extract_bitstring $lid:data$ $lid:off$ $lid:len$
619 (* Bitstring, constant flen = -1, means consume all the
622 | P.Bitstring, Some i when i = -1 ->
625 | <:patt< $lid:ident$ >> -> ident
626 | <:patt< _ >> -> "_"
628 fail "cannot compare a bitstring to a constant" in
630 let $lid:ident$, $lid:off$, $lid:len$ =
631 Bitstring.extract_remainder $lid:data$ $lid:off$ $lid:len$ in
635 | P.Bitstring, Some _ ->
636 fail "length of bitstring must be >= 0 or the special value -1"
638 (* Bitstring field, non-const flen. We check the flen is >= 0
639 * (-1 is not allowed here) at runtime.
641 | P.Bitstring, None ->
644 | <:patt< $lid:ident$ >> -> ident
645 | <:patt< _ >> -> "_"
647 fail "cannot compare a bitstring to a constant" in
649 if $flen$ >= 0 && $flen$ <= $lid:len$ then (
650 let $lid:ident$, $lid:off$, $lid:len$ =
651 Bitstring.extract_bitstring $lid:data$ $lid:off$ $lid:len$
658 (* Computed offset: only offsets forward are supported.
660 * We try hard to optimize this based on what we know. Are
661 * we at a predictable offset now? (Look at the outer 'fields'
662 * list and see if they all have constant field length starting
663 * at some constant offset). Is this offset constant?
665 * Based on this we can do a lot of the computation at
666 * compile time, or defer it to runtime only if necessary.
668 * In all cases, the off and len fields get updated.
671 match P.get_offset field with
672 | None -> expr (* common case: there was no offset expression *)
673 | Some offset_expr ->
674 (* This will be [Some i] if offset is a constant expression
675 * or [None] if it's a non-constant.
677 let requested_offset = expr_is_constant offset_expr in
679 (* This will be [Some i] if our current offset is known
680 * at compile time, or [None] if we can't determine it.
683 let has_constant_offset field =
684 match P.get_offset field with
687 match expr_is_constant expr with
691 let get_constant_offset field =
692 match P.get_offset field with
693 | None -> assert false
695 match expr_is_constant expr with
696 | None -> assert false
700 let has_constant_len field =
701 match expr_is_constant (P.get_length field) with
703 | Some i when i > 0 -> true
706 let get_constant_len field =
707 match expr_is_constant (P.get_length field) with
708 | None -> assert false
709 | Some i when i > 0 -> i
710 | Some _ -> assert false
713 let rec loop = function
714 (* first field has constant offset 0 *)
716 (* field with constant offset & length *)
718 when has_constant_offset field &&
719 has_constant_len field ->
720 Some (get_constant_offset field + get_constant_len field)
721 (* field with no offset & constant length *)
723 when P.get_offset field = None &&
724 has_constant_len field ->
725 (match loop fields with
727 | Some offset -> Some (offset + get_constant_len field))
728 (* else, can't work out the offset *)
733 (* Look at the current offset and requested offset cases and
734 * determine what code to generate.
736 match current_offset, requested_offset with
737 (* This is the good case: both the current offset and
738 * the requested offset are constant, so we can remove
739 * almost all the runtime checks.
741 | Some current_offset, Some requested_offset ->
742 let move = requested_offset - current_offset in
744 fail (sprintf "requested offset is less than the current offset (%d < %d)" requested_offset current_offset);
745 (* Add some code to move the offset and length by a
746 * constant amount, and a runtime test that len >= 0
747 * (XXX possibly the runtime test is unnecessary?)
750 let $lid:off$ = $lid:off$ + $`int:move$ in
751 let $lid:len$ = $lid:len$ - $`int:move$ in
752 if $lid:len$ >= 0 then $expr$
754 (* In any other case, we need to use runtime checks.
756 * XXX It's not clear if a backwards move detected at runtime
757 * is merely a match failure, or a runtime error. At the
758 * moment it's just a match failure since bitmatch generally
759 * doesn't raise runtime errors.
762 let move = gensym "move" in
765 $offset_expr$ - ($lid:off$ - $lid:original_off$) in
766 if $lid:move$ >= 0 then (
767 let $lid:off$ = $lid:off$ + $lid:move$ in
768 let $lid:len$ = $lid:len$ - $lid:move$ in
769 if $lid:len$ >= 0 then $expr$
771 >> in (* end of computed offset code *)
773 (* save_offset_to(patt) saves the current offset into a variable. *)
775 match P.get_save_offset_to field with
776 | None -> expr (* no save_offset_to *)
779 let $patt$ = $lid:off$ - $lid:original_off$ in
783 (* Emit extra debugging code. *)
785 if not debug then expr else (
786 let field = P.string_of_pattern_field field in
789 if !Bitstring.debug then (
790 Printf.eprintf "PA_BITSTRING: TEST:\n";
791 Printf.eprintf " %s\n" $str:field$;
792 Printf.eprintf " off %d len %d\n%!" $lid:off$ $lid:len$;
793 (*Bitstring.hexdump_bitstring stderr
794 ($lid:data$,$lid:off$,$lid:len$);*)
800 output_field_extraction expr fields
803 (* Convert each case in the match. *)
804 let cases = List.map (
805 fun (fields, bind, whenclause, code) ->
806 let inner = <:expr< $lid:result$ := Some ($code$); raise Exit >> in
808 match whenclause with
810 <:expr< if $whenclause$ then $inner$ >>
816 let $lid:name$ = ($lid:data$, $lid:off$, $lid:len$) in
820 output_field_extraction inner (List.rev fields)
823 (* Join them into a single expression.
825 * Don't do it with a normal fold_right because that leaves
826 * 'raise Exit; ()' at the end which causes a compiler warning.
827 * Hence a bit of complexity here.
829 * Note that the number of cases is always >= 1 so List.hd is safe.
831 let cases = List.rev cases in
833 List.fold_left (fun base case -> <:expr< $case$ ; $base$ >>)
834 (List.hd cases) (List.tl cases) in
836 (* The final code just wraps the list of cases in a
837 * try/with construct so that each case is tried in
838 * turn until one case matches (that case sets 'result'
839 * and raises 'Exit' to leave the whole statement).
840 * If result isn't set by the end then we will raise
841 * Match_failure with the location of the bitmatch
842 * statement in the original code.
844 let loc_fname = Loc.file_name _loc in
845 let loc_line = string_of_int (Loc.start_line _loc) in
846 let loc_char = string_of_int (Loc.start_off _loc - Loc.start_bol _loc) in
849 (* Note we save the original offset/length at the start of the match
850 * in 'original_off'/'original_len' symbols. 'data' never changes.
852 let ($lid:data$, $lid:original_off$, $lid:original_len$) = $bs$ in
853 let $lid:off$ = $lid:original_off$ and $lid:len$ = $lid:original_len$ in
854 let $lid:result$ = ref None in
858 match ! $lid:result$ with
860 | None -> raise (Match_failure ($str:loc_fname$,
861 $int:loc_line$, $int:loc_char$))
864 (* Add a named pattern. *)
865 let add_named_pattern _loc name pattern =
866 Hashtbl.add pattern_hash name pattern
868 (* Expand a named pattern from the pattern_hash. *)
869 let expand_named_pattern _loc name =
870 try Hashtbl.find pattern_hash name
872 locfail _loc (sprintf "named pattern not found: %s" name)
874 (* Add named patterns from a file. See the documentation on the
875 * directory search path in bitstring_persistent.mli
877 let load_patterns_from_file _loc filename =
879 if Filename.is_relative filename && Filename.is_implicit filename then (
880 (* Try current directory. *)
883 (* Try OCaml library directory. *)
884 try open_in (Filename.concat Bitstring_config.ocamllibdir filename)
885 with exn -> Loc.raise _loc exn
888 with exn -> Loc.raise _loc exn
890 let names = ref [] in
893 let name = P.named_from_channel chan in
894 names := name :: !names
897 with End_of_file -> ()
900 let names = List.rev !names in
903 | name, P.Pattern patt ->
905 locfail _loc (sprintf "pattern %s: no fields" name);
906 add_named_pattern _loc name patt
907 | _, P.Constructor _ -> () (* just ignore these for now *)
911 GLOBAL: expr str_item;
913 (* Qualifiers are a list of identifiers ("string", "bigendian", etc.)
914 * followed by an optional expression (used in certain cases). Note
915 * that we are careful not to declare any explicit reserved words.
920 e = OPT [ "("; e = expr; ")" -> e ] -> (q, e) ]
924 (* Field used in the bitmatch operator (a pattern). This can actually
925 * return multiple fields, in the case where the 'field' is a named
929 [ fpatt = patt; ":"; len = expr LEVEL "top";
930 qs = OPT [ ":"; qs = qualifiers -> qs ] ->
931 let field = P.create_pattern_field _loc in
932 let field = P.set_patt field fpatt in
933 let field = P.set_length field len in
934 [parse_field _loc field qs] (* Normal, single field. *)
935 | ":"; name = LIDENT ->
936 expand_named_pattern _loc name (* Named -> list of fields. *)
940 (* Case inside bitmatch operator. *)
943 fields = LIST0 patt_field SEP ";";
950 [ fields = patt_fields;
951 bind = OPT [ "as"; name = LIDENT -> name ];
952 whenclause = OPT [ "when"; e = expr -> e ]; "->";
954 (fields, bind, whenclause, code)
958 (* Field used in the BITSTRING constructor (an expression). *)
960 [ fexpr = expr LEVEL "top"; ":"; len = expr LEVEL "top";
961 qs = OPT [ ":"; qs = qualifiers -> qs ] ->
962 let field = P.create_constructor_field _loc in
963 let field = P.set_expr field fexpr in
964 let field = P.set_length field len in
965 parse_field _loc field qs
971 fields = LIST0 constr_field SEP ";";
977 (* 'bitmatch' expressions. *)
980 bs = expr; "with"; OPT "|";
981 cases = LIST1 patt_case SEP "|" ->
982 output_bitmatch _loc bs cases
987 fields = constr_fields ->
988 output_constructor _loc fields
992 (* Named persistent patterns.
994 * NB: Currently only allowed at the top level. We can probably lift
995 * this restriction later if necessary. We only deal with patterns
996 * at the moment, not constructors, but the infrastructure to do
997 * constructors is in place.
999 str_item: LEVEL "top" [
1000 [ "let"; "bitmatch";
1001 name = LIDENT; "="; fields = patt_fields ->
1002 add_named_pattern _loc name fields;
1003 (* The statement disappears, but we still need a str_item so ... *)
1005 | "open"; "bitmatch"; filename = STRING ->
1006 load_patterns_from_file _loc filename;