I see three problems:

1)The [^>] in (?=[^>]*\n) can match newlines.

2) ">" isn't allowed anywhere on the ">>>" line (after the ">>>"). Is that intentional?

3) The second negative lookahead is currently anchored to the character right after the ">>>", not to the start of the next line.

Fixes:

s/
   (?<!^>>[^>])
   ^>>>
   (?=
      [^\n]*\n      # <-- removed < and added first \n
      (?!^>>[^>])   # <-- moved into (?=...)
   )
/PYTHON_PROMPT/xmg;
[download]

Updated

Also, the above (like your original) won't work if anything appears after ">>" on the previous line (including spaces). For example,

>> text
>>> text
[download]

gets changed to

>> text
PYTHON_PROMPT text
[download]

Is that how it should be? Fixing this isn't easy, and it might be impossible without $1

Update: Well, the following doesn't use $1. However, I don't know about effeciency (since I don't have another solution with which to compare). It works, though, which is more than I can say for my attempts that did use $1.

Read more... (658 Bytes)

{
   foreach ($_) {
      if (/\G (?= >>> )/xgc) {
         # Substitution supressed by following ">>".
         redo if /\G >>> [^\n]* \n (?= >> (?!>) ) /xcg;

         # Do substitution.
         s/\G >>>/PYTHON_PROMPT/xcg;

         # Go process next line.
         /\G [^\n]* \n /xcg;
         redo;
      }

      # Substitution prevented by preceeding ">>"
      /\G >> (?!>) [^\n]* \n >>> [^\n]* \n /xcg && redo;
           # ^^^^^ optional due to statement order.

      # Go process next line.
      m/\G [^\n]* \n /xcg && redo;
   }

   print;
   print("==========\n");
}
[download]

#!/usr/bin/perl

use strict;

my ( $prev, $this ) = ( undef(), undef() );
while( <> ) {
    $this = $_;
    /^>>[^>]/ or mySubst( \$prev );
    print $prev;
    $prev = $this;
}

# finally, process the last line of input which got left over
mySubst( \$prev );
print $prev;

sub mySubst {
    my $sref = shift;
    $$sref =~ s/^>>>/PYTHON_PROMPT/;
}
[download]

my $end = @test;
my $cnt = 0;
until ($cnt == $end)
    
    if (($test[$cnt] =~ /^>>>\s/) && ($test[$cnt+1] !~ /^>>\s/)){
        $tst[$cnt] =~ s/^>>>/PYTHON_PROMPT/;
    }
    $cnt++;
}
[download]

s/^>>>(?!>)(?![^\n]*\n>>(?!>))/PYTHON_PROMPT/gm;
[download]

Also, I'd prefer to avoid $1 if possible, for performance reasons.

Clarification:

>>> etc
>>> etc
>>> etc

should be converted to
PYTHON_PROMPT etc
PYTHON_PROMPT etc
PYTHON_PROMPT etc

while
>>> etc
>>> etc
>> etc
should not.

So, those read aheads / arrays / etc won't work, will they?
[download]

Also, re not using $1 for performance - it's not so much that I mind for this regex, it's that I'm under the impression that once you use $1 once, perl needs to track it for every regex in the entire program... which I don't want to slow down. Is there a way to tell Perl to only turn on $1 here, but no where else...

it's that I'm under the impression that once you use $1 once, perl needs to track it for every regex in the entire program...

You're thinking of $&, $` and $'. They are always set by all regexps when they are used anywhere in the program.

$1 .. $9 are always set by every regexps that succeeds. When a given regexp has no corresponding capture, they are set to undefined. No attention is payed to whether $1 .. $9 are used anywhere.

Speaking of test, here's some code I wrote to do basic testing

use Test::Simple tests => 7;

sub cpi
{
# the converter goes here
}

$a =
'>> no snake here
>>> or here

I agree';

$b =
'>> and neither should this
>>> be understood snakily
>>> because it is a nested quote
>> yes I know it is not really named after a snake
';

$c =
'>>
>>>no snake
>>>no snake
';

$d=
'>>> still no snake seen
>> right!
';

$snakea =
'>>> i = j
';

$snakeb =
'>>> i = j + 1
>>> q

is good code';

$snakec = 
'>> does this work right?
lets see:
>>> i = test(func)
';

ok (cpi($a) eq $a);
ok (cpi($b) eq $b);
ok (cpi($c) eq $c);
ok (cpi($d) eq $d);

ok (cpi($snakea) eq 
'PYTHON_PROMPT_PROTECTED i = j
');
ok (cpi($snakeb) eq
'PYTHON_PROMPT_PROTECTED i = j + 1
PYTHON_PROMPT_PROTECTED q

is good code');
ok(cpi($snakec) eq
'>> does this work right?
lets see:
PYTHON_PROMPT_PROTECTED i = test(func)
');
[download]

Please see my test cases below - your code failed 6 out of 7!