Each engine is supposed to provide access to a constant structure of the following format:
typedef struct regexp_engine {
REGEXP* (*comp) (pTHX_ const SV * const pattern, const U32 flags);
I32 (*exec) (pTHX_ REGEXP * const rx, char* stringarg, char* strend,
char* strbeg, I32 minend, SV* screamer,
void* data, U32 flags);
char* (*intuit) (pTHX_ REGEXP * const rx, SV *sv, char *strpos,
char *strend, U32 flags,
struct re_scream_pos_data_s *data);
SV* (*checkstr) (pTHX_ REGEXP * const rx);
void (*free) (pTHX_ REGEXP * const rx);
void (*numbered_buff_FETCH) (pTHX_ REGEXP * const rx, const I32 paren,
SV * const sv);
void (*numbered_buff_STORE) (pTHX_ REGEXP * const rx, const I32 paren,
SV const * const value);
I32 (*numbered_buff_LENGTH) (pTHX_ REGEXP * const rx, const SV * const sv,
const I32 paren);
SV* (*named_buff) (pTHX_ REGEXP * const rx, SV * const key,
SV * const value, U32 flags);
SV* (*named_buff_iter) (pTHX_ REGEXP * const rx, const SV * const lastkey,
const U32 flags);
SV* (*qr_package)(pTHX_ REGEXP * const rx);
#ifdef USE_ITHREADS
void* (*dupe) (pTHX_ REGEXP * const rx, CLONE_PARAMS *param);
#endif
When a regexp is compiled, its "engine" field is then set to point at the appropriate structure, so that when it needs to be used Perl can find the right routines to do so.
In order to install a new regexp handler, $^H{regcomp} is set to an integer which (when casted appropriately) resolves to one of these structures. When compiling, the "comp" method is executed, and the resulting regexp structure's engine field is expected to point back at the same structure.
The pTHX_ symbol in the definition is a macro used by perl under threading to provide an extra argument to the routine holding a pointer back to the interpreter that is executing the regexp. So under threading all routines get an extra argument.
REGEXP* comp(pTHX_ const SV * const pattern, const U32 flags);
Compile the pattern stored in "pattern" using the given "flags" and return a pointer to a prepared "REGEXP" structure that can perform the match. See ``The REGEXP structure'' below for an explanation of the individual fields in the REGEXP struct.
The "pattern" parameter is the scalar that was used as the pattern. previous versions of perl would pass two "char*" indicating the start and end of the stringified pattern, the following snippet can be used to get the old parameters:
STRLEN plen;
char* exp = SvPV(pattern, plen);
char* xend = exp + plen;
Since any scalar can be passed as a pattern it's possible to implement an engine that does something with an array (""ook" =~ [ qw/ eek hlagh / ]") or with the non-stringified form of a compiled regular expression (""ook" =~ qr/eek/"). perl's own engine will always stringify everything using the snippet above but that doesn't mean other engines have to.
The "flags" parameter is a bitfield which indicates which of the "msixp" flags the regex was compiled with. It also contains additional info such as whether "use locale" is in effect.
The "eogc" flags are stripped out before being passed to the comp routine. The regex engine does not need to know whether any of these are set as those flags should only affect what perl does with the pattern and its match variables, not how it gets compiled and executed.
By the time the comp callback is called, some of these flags have already had effect (noted below where applicable). However most of their effect occurs after the comp callback has run in routines that read the "rx->extflags" field which it populates.
In general the flags should be preserved in "rx->extflags" after compilation, although the regex engine might want to add or delete some of them to invoke or disable some special behavior in perl. The flags along with any special behavior they cause are documented below:
The pattern modifiers:
TODO: Document those cases.
Additional flags:
A regex engine may want to set or disable this flag during compilation. The perl engine for instance may upgrade non-UTF-8 strings to UTF-8 if the pattern includes constructs such as "\x{...}" that can only match Unicode values.
if (flags & RXf_SPLIT && r->prelen == 1 && r->precomp[0] == ' ')
r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
These flags can be set during compilation to enable optimizations in the "split" operator.
If RXf_WHITE is set in addition to this flag "split" will behave like "split " "" under the perl engine.
Perl's engine sets this if the pattern is "/^/" ("plen == 1 && *exp == '^'"), even under "/^/s", see split. Of course a different regex engine might want to use the same optimizations with a different syntax.
Perl's engine sets this flag if the pattern is "\s+".
Perl's engine sets this flag on empty patterns, this optimization makes "split //" much faster than it would otherwise be. It's even faster than "unpack".
I32 exec(pTHX_ REGEXP * const rx,
char *stringarg, char* strend, char* strbeg,
I32 minend, SV* screamer,
void* data, U32 flags);
char* intuit(pTHX_ REGEXP * const rx,
SV *sv, char *strpos, char *strend,
const U32 flags, struct re_scream_pos_data_s *data);
Find the start position where a regex match should be attempted, or possibly whether the regex engine should not be run because the pattern can't match. This is called as appropriate by the core depending on the values of the extflags member of the regexp structure.
SV* checkstr(pTHX_ REGEXP * const rx);
Return a SV containing a string that must appear in the pattern. Used by "split" for optimising matches.
void free(pTHX_ REGEXP * const rx);
Called by perl when it is freeing a regexp pattern so that the engine can release any resources pointed to by the "pprivate" member of the regexp structure. This is only responsible for freeing private data; perl will handle releasing anything else contained in the regexp structure.
The "paren" parameter will be "-2" for "$`", "-1" for "$'", 0 for $&, 1 for $1 and so forth.
The names have been chosen by analogy with Tie::Scalar methods names with an additional LENGTH callback for efficiency. However named capture variables are currently not tied internally but implemented via magic.
void numbered_buff_FETCH(pTHX_ REGEXP * const rx, const I32 paren,
SV * const sv);
Fetch a specified numbered capture. "sv" should be set to the scalar to return, the scalar is passed as an argument rather than being returned from the function because when it's called perl already has a scalar to store the value, creating another one would be redundant. The scalar can be set with "sv_setsv", "sv_setpvn" and friends, see perlapi.
This callback is where perl untaints its own capture variables under taint mode (see perlsec). See the "Perl_reg_numbered_buff_fetch" function in regcomp.c for how to untaint capture variables if that's something you'd like your engine to do as well.
void (*numbered_buff_STORE) (pTHX_ REGEXP * const rx, const I32 paren,
SV const * const value);
Set the value of a numbered capture variable. "value" is the scalar that is to be used as the new value. It's up to the engine to make sure this is used as the new value (or reject it).
Example:
if ("ook" =~ /(o*)/) {
# `paren' will be `1' and `value' will be `ee'
$1 =~ tr/o/e/;
}
Perl's own engine will croak on any attempt to modify the capture variables, to do this in another engine use the following callback (copied from "Perl_reg_numbered_buff_store"):
void
Example_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
SV const * const value)
{
PERL_UNUSED_ARG(rx);
PERL_UNUSED_ARG(paren);
PERL_UNUSED_ARG(value);
if (!PL_localizing)
Perl_croak(aTHX_ PL_no_modify);
}
Actually perl will not always croak in a statement that looks like it would modify a numbered capture variable. This is because the STORE callback will not be called if perl can determine that it doesn't have to modify the value. This is exactly how tied variables behave in the same situation:
package CaptureVar;
use base 'Tie::Scalar';
sub TIESCALAR { bless [] }
sub FETCH { undef }
sub STORE { die "This doesn't get called" }
package main;
tie my $sv => "CatptureVar";
$sv =~ y/a/b/;
Because $sv is "undef" when the "y///" operator is applied to it the transliteration won't actually execute and the program won't "die". This is different to how 5.8 and earlier versions behaved since the capture variables were READONLY variables then, now they'll just die when assigned to in the default engine.
I32 numbered_buff_LENGTH (pTHX_ REGEXP * const rx, const SV * const sv,
const I32 paren);
Get the "length" of a capture variable. There's a special callback for this so that perl doesn't have to do a FETCH and run "length" on the result, since the length is (in perl's case) known from an offset stored in "<rx-"offs> this is much more efficient:
I32 s1 = rx->offs[paren].start;
I32 s2 = rx->offs[paren].end;
I32 len = t1 - s1;
This is a little bit more complex in the case of UTF-8, see what "Perl_reg_numbered_buff_length" does with is_utf8_string_loclen.
There are two callbacks, "named_buff" is called in all the cases the FETCH, STORE, DELETE, CLEAR, EXISTS and SCALAR Tie::Hash callbacks would be on changes to "%+" and "%-" and "named_buff_iter" in the same cases as FIRSTKEY and NEXTKEY.
The "flags" parameter can be used to determine which of these operations the callbacks should respond to, the following flags are currently defined:
Which Tie::Hash operation is being performed from the Perl level on "%+" or "%+", if any:
RXapif_FETCH
RXapif_STORE
RXapif_DELETE
RXapif_CLEAR
RXapif_EXISTS
RXapif_SCALAR
RXapif_FIRSTKEY
RXapif_NEXTKEY
Whether "%+" or "%-" is being operated on, if any.
RXapif_ONE /* %+ */
RXapif_ALL /* %- */
Whether this is being called as "re::regname", "re::regnames" or "re::regnames_count", if any. The first two will be combined with "RXapif_ONE" or "RXapif_ALL".
RXapif_REGNAME
RXapif_REGNAMES
RXapif_REGNAMES_COUNT
Internally "%+" and "%-" are implemented with a real tied interface via Tie::Hash::NamedCapture. The methods in that package will call back into these functions. However the usage of Tie::Hash::NamedCapture for this purpose might change in future releases. For instance this might be implemented by magic instead (would need an extension to mgvtbl).
SV* (*named_buff) (pTHX_ REGEXP * const rx, SV * const key,
SV * const value, U32 flags);
SV* (*named_buff_iter) (pTHX_ REGEXP * const rx, const SV * const lastkey,
const U32 flags);
SV* qr_package(pTHX_ REGEXP * const rx);
The package the qr// magic object is blessed into (as seen by "ref qr//"). It is recommended that engines change this to their package name for identification regardless of whether they implement methods on the object.
The package this method returns should also have the internal "Regexp" package in its @ISA. "qr//-"isa(``Regexp'')> should always be true regardless of what engine is being used.
Example implementation might be:
SV*
Example_qr_package(pTHX_ REGEXP * const rx)
{
PERL_UNUSED_ARG(rx);
return newSVpvs("re::engine::Example");
}
Any method calls on an object created with "qr//" will be dispatched to the package as a normal object.
use re::engine::Example;
my $re = qr//;
$re->meth; # dispatched to re::engine::Example::meth()
To retrieve the "REGEXP" object from the scalar in an XS function use the "SvRX" macro, see ``REGEXP Functions'' in perlapi.
void meth(SV * rv)
PPCODE:
REGEXP * re = SvRX(sv);
void* dupe(pTHX_ REGEXP * const rx, CLONE_PARAMS *param);
On threaded builds a regexp may need to be duplicated so that the pattern can be used by multiple threads. This routine is expected to handle the duplication of any private data pointed to by the "pprivate" member of the regexp structure. It will be called with the preconstructed new regexp structure as an argument, the "pprivate" member will point at the old private structure, and it is this routine's responsibility to construct a copy and return a pointer to it (which perl will then use to overwrite the field as passed to this routine.)
This allows the engine to dupe its private data but also if necessary modify the final structure if it really must.
On unthreaded builds this field doesn't exist.
The REGEXP structure contains all the data that perl needs to be aware of to properly work with the regular expression. It includes data about optimisations that perl can use to determine if the regex engine should really be used, and various other control info that is needed to properly execute patterns in various contexts such as is the pattern anchored in some way, or what flags were used during the compile, or whether the program contains special constructs that perl needs to be aware of.
In addition it contains two fields that are intended for the private use of the regex engine that compiled the pattern. These are the "intflags" and "pprivate" members. "pprivate" is a void pointer to an arbitrary structure whose use and management is the responsibility of the compiling engine. perl will never modify either of these values.
typedef struct regexp {
/* what engine created this regexp? */
const struct regexp_engine* engine;
/* what re is this a lightweight copy of? */
struct regexp* mother_re;
/* Information about the match that the perl core uses to manage things */
U32 extflags; /* Flags used both externally and internally */
I32 minlen; /* mininum possible length of string to match */
I32 minlenret; /* mininum possible length of $& */
U32 gofs; /* chars left of pos that we search from */
/* substring data about strings that must appear
in the final match, used for optimisations */
struct reg_substr_data *substrs;
U32 nparens; /* number of capture buffers */
/* private engine specific data */
U32 intflags; /* Engine Specific Internal flags */
void *pprivate; /* Data private to the regex engine which
created this object. */
/* Data about the last/current match. These are modified during matching*/
U32 lastparen; /* last open paren matched */
U32 lastcloseparen; /* last close paren matched */
regexp_paren_pair *swap; /* Swap copy of *offs */
regexp_paren_pair *offs; /* Array of offsets for (@-) and (@+) */
char *subbeg; /* saved or original string so \digit works forever. */
SV_SAVED_COPY /* If non-NULL, SV which is COW from original */
I32 sublen; /* Length of string pointed by subbeg */
/* Information about the match that isn't often used */
I32 prelen; /* length of precomp */
const char *precomp; /* pre-compilation regular expression */
char *wrapped; /* wrapped version of the pattern */
I32 wraplen; /* length of wrapped */
I32 seen_evals; /* number of eval groups in the pattern - for security checks */
HV *paren_names; /* Optional hash of paren names */
/* Refcount of this regexp */
I32 refcnt; /* Refcount of this regexp */
} regexp;
The fields are discussed in more detail below:
Internally this is set to "NULL" unless a custom engine is specified in $^H{regcomp}, perl's own set of callbacks can be accessed in the struct pointed to by "RE_ENGINE_PTR".
"minlenret" is the minimum length of the string that would be found in $& after a match.
The difference between "minlen" and "minlenret" can be seen in the following pattern:
/ns(?=\d)/
where the "minlen" would be 3 but "minlenret" would only be 2 as the \d is required to match but is not actually included in the matched content. This distinction is particularly important as the substitution logic uses the "minlenret" to tell whether it can do in-place substitution which can result in considerable speedup.
typedef struct regexp_paren_pair {
I32 start;
I32 end;
} regexp_paren_pair;
If "->offs[num].start" or "->offs[num].end" is "-1" then that capture buffer did not match. "->offs[0].start/end" represents $& (or "${^MATCH" under "//p") and "->offs[paren].end" matches $$paren where $paren = 1>.
The relevant snippet from "Perl_pp_regcomp":
if (!re || !re->precomp || re->prelen != (I32)len ||
memNE(re->precomp, t, len))
/* Compile a new pattern */
When using a custom engine that doesn't support the "(?:)" construct for inline modifiers, it's probably best to have "qr//" stringify to the supplied pattern, note that this will create undesired patterns in cases such as:
my $x = qr/a|b/; # "a|b"
my $y = qr/c/i; # "c"
my $z = qr/$x$y/; # "a|bc"
There's no solution for this problem other than making the custom engine understand a construct like "(?:)".
This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.