re-eval lexvar/global mixup

[GoogleCodeExporter]

$>perl -e 'my $code = q[{$blah = 45}]; our $blah = 12; eval "/(?$code)/"; print 
"$blah\n"'
45

$>perlcc -O3 -r -e 'my $code = q[{$blah = 45}]; our $blah = 12; eval 
"/(?$code)/"; print "$blah\n"'
12

Original issue reported on code.google.com by todd.e.rinaldo on 9 May 2014 at 9:06

[GoogleCodeExporter]

We can't test most of the re test suite without fixing this one.

Original comment by todd.e.rinaldo on 9 May 2014 at 9:13

• Added labels: Priority-High
• Removed labels: Priority-Medium

[GoogleCodeExporter]

same kind of issue when trying to call a sub

> perlcc -r -e 'sub func { 42 } $_ = "x"; /x(?{func "in multiline regexp"})/m'
String found where operator expected at (re_eval 1) line 1, near "func "in 
multiline regexp""
    (Do you need to predeclare func?)
syntax error at (re_eval 1) line 1, near "func "in multiline regexp""
Compilation failed in regexp.

> perl -e 'sub func { 42 } $_ = "x"; /x(?{func "in multiline regexp"})/m'

Original comment by nicolas....@gmail.com on 9 May 2014 at 10:25

[GoogleCodeExporter]

$>perlcc -O3 -r -e 'my $code = q[{$blah = 45}]; our $blah = 12; eval 
"/(?$code)/"; print "$@\n"; print "$blah\n"'
Modification of a read-only value attempted at (re_eval 2) line 1.

12

Original comment by todd.e.rinaldo on 9 May 2014 at 10:41

[GoogleCodeExporter]

So looks like it's modifying the variable in this case that's going wrong, not 
the execution in a regex:

$>perlcc -O3 -r -e 'my $code = q[{print "hello"}]; our $blah = 12; eval 
"/(?$code)/"; print "$@\n"; '               
hello

Original comment by todd.e.rinaldo on 9 May 2014 at 10:42

[GoogleCodeExporter]

Converting the our to my fixes things

$>perlcc -O3 -r -e 'my $code = q[{$blah = 45}]; my $blah = 12; eval 
"/(?$code)/"; print "$@\n"; print "$blah\n"'      

45

Original comment by todd.e.rinaldo on 9 May 2014 at 10:47

[GoogleCodeExporter]

Original comment by reini.urban on 9 May 2014 at 11:09

• Changed state: Duplicate

[GoogleCodeExporter]

See also issue 135 which errors with:

Eval-group not allowed at runtime, use re 'eval' in regex

Original comment by reini.urban on 12 May 2014 at 2:05

• Changed state: Started

[GoogleCodeExporter]

328 run via -Dtr:

(ccode328.pl:2) entereval
Compiling REx "(?{$blah = 45})"
Final program:
   1: EVAL (3)
   3: END (0)
minlen 0 with eval 
((eval 1):0)    nextstate
((eval 1):1)    match
Matching REx "(?{$blah = 45})" against ""
   0 <> <>                   |  1:EVAL(3)
((eval 1):1)    nextstate
((re_eval 2):1) const(IV(45))

Difference: compiled
((re_eval 2):1) padsv($blah)
((re_eval 2):1) sassign

uncompiled:
((re_eval 2):1) gvsv(main::blah)
((re_eval 2):1) sassign

gvsv finds the blah symbol and assigns to it, padsv not.

Original comment by reini.urban on 2 Jun 2014 at 9:22

[GoogleCodeExporter]

The problem in 5.14 is caused by Perl_sv_compile_2op_is_broken() at 
regcomp.c:6819 which miscompiles "$blah = 45" to a padsv instead of gvsv, just 
because there also exists a lexical with the same name in the pad.

pp_ctl.c:3072
/* Don't use this. It will go away without warning once the regexp engine is
   refactored not to use it.  */

Looks like an upstream bug, EWONTFIX'able in the compiler for now.

Original comment by reini.urban on 3 Jun 2014 at 3:22

• Changed title: re-eval lexvar/global mixup
• Changed state: WontFix

[GoogleCodeExporter]

Original comment by todd.e.rinaldo on 3 Jun 2014 at 3:29

• Added labels: re-eval-5.20