Front page | perl.perl5.porters |
Postings from January 2012
On 27 January 2012 15:40, l.mai@web.de <perlbug-followup@perl.org> wrote:
> # New Ticket Created by l.mai@web.de
> # Please include the string: [perl #109206]
> # in the subject line of all future correspondence about this issue.
> # <URL: https://rt.perl.org:443/rt3/Ticket/Display.html?id=109206 >
>
>
>
> This is a bug report for perl from l.mai@web.de,
> generated with the help of perlbug 1.39 running under perl 5.14.2.
>
>
> -----------------------------------------------------------------
> [Please describe your issue here]
>
> % perl -wle '$_ = "\n"; print $+[0] while /[^\n]*/g'
> 0
> 1
>
> % perl -wle '$_ = "\n"; print $+[0] while /.*/g'
> 0
>
> I think this is a bug because in the absence of /s '.' should match any
> character except newline, i.e. be equivalent to '[^\n]'. The two programs
> should produce identical output.
>
> I also think the first result is correct because there are two zero-length
> matches in "\n", one at the beginning of the string and one at the end.
> In conclusion: it looks like /.*/g is broken.
This problem is caused by a broken optimisation. The ANCH_MBOL
optmisation. Notice it the principle difference in these two outputs:
$ perl -Mre=Debug,DUMP -wle '$_ = "\n"; print pos($_),":",$+[0] while /.*/g'
Compiling REx ".*"
Final program:
1: STAR (3)
2: REG_ANY (0)
3: END (0)
anchored(MBOL) implicit minlen 0
0:0
Freeing REx: ".*"
$ perl -Mre=Debug,DUMP -wle '$_ = "\n"; print pos($_),":",$+[0] while
/[^\n]*/g'
Compiling REx "[^\n]*"
Final program:
1: STAR (13)
2: ANYOF[\0-\11\13-\377][{unicode_all}] (0)
13: END (0)
minlen 0
0:0
1:1
Freeing REx: "[^\n]*"
It is enabled by this block of code in regcomp. Notice the comment:
/* turn .* into ^.* with an implied $*=1 */
I have to admit I have not checked to see what the heck $*=1 means.
else if ((!sawopen || !RExC_sawback) &&
(OP(first) == STAR &&
PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
!(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
{
/* turn .* into ^.* with an implied $*=1 */
const int type =
(OP(NEXTOPER(first)) == REG_ANY)
? RXf_ANCH_MBOL
: RXf_ANCH_SBOL;
r->extflags |= type;
r->intflags |= PREGf_IMPLICIT;
first = NEXTOPER(first);
goto again;
}
The following patch disables the optimization:
diff --git a/regcomp.c b/regcomp.c
index 668f8f7..12d0ac0 100644
--- a/regcomp.c
+++ b/regcomp.c
@@ -5235,7 +5235,7 @@ reStudy:
first = NEXTOPER(first);
goto again;
}
- else if ((!sawopen || !RExC_sawback) &&
+ else if (0 && (!sawopen || !RExC_sawback) &&
(OP(first) == STAR &&
PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
!(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
Producing this output:
$ ./perl -Ilib -Mre=Debug,DUMP -wle '$_ = "\n"; print
pos($_),":",$+[0] while /.*/g'
Compiling REx ".*"
Final program:
1: STAR (3)
2: REG_ANY (0)
3: END (0)
minlen 0
0:0
1:1
Freeing REx: ".*"
I have not committed this patch as I dont know what effects it might
have, however as it is a "conversion optimization" I would assume it
can be safely disabled until the underlying logic is fixed. However I
will note that fixing it might be tricky, the relevent code is spread
out over pp_hot.c and CALLREG_INTUIT_START(), and is particularly
hairy anyway. It always makes me kinda cringe when I look at pp_match.
Yves
--
perl -Mre=debug -e "/just|another|perl|hacker/"
Thread Previous
|
Thread Next