#!/usr/bin/perl -l

# http://perlmonks.org/?node_id=1206551

use strict;
use warnings;

my $param = 'password';
for ( q~{'password' => 'secret'}~, q~{"password" => "sec\"ret"}~ )
  {
  local $_ = $_;
  s/
        (                   # capture everything up to the start of th
+e value
          (                   # capture the quotation mark we are usin
+g
            (?<!\\\\)         # not escaped
            [ ' " ]           # either kind of quote
          )                   # end capture quotation mark
          $param              # the key
          \2                  # the same quotation mark
          \s*                 # any amount of space
          (?: => | : )        # perl or JSON key-value "connector"
          \s*                 # any amount of space
          \2                  # the same quotation mark
        )                   # end capture everything up to start of th
+e value
        (?:                 # group but do not capture the value
          \\. | (?!\2) .    # defined as any character except the same
+ quote
        )*                  # any number of times
    /$1***/smxg;            # the closing quotation mark will remain i
+n place

  print $_;
  }
[download]

Outputs:

{'password' => '***'}
{"password" => "***"}
[download]

[ ' " ]           # either kind of quote

Note that  [ ' " ] also matches a blank (0x20). Better perhaps to use  ['"] instead.

---

Give a man a fish:  <%-{-{-{-<

Yes. But there's another possibility: - quoting Regexp Quote-Like Operators in perlop:

x   Use extended regular expressions;
specifying two x's means \t and the SPACE character are ignored within square-bracketed character classes

I am in awe of the effortless boxlessness of your mind.

Thank you very much.



The way forward always starts with a minimal test.

Unfortunately I don't have enough time right now to wrap my head around the regex - I just wanted to point out a few resources that might help:

• Regexp::Common::delimited - handles escaped quotes too
• Text::Balanced
• For parsing Data::Dumper, various efforts to parse Perl like PPI, and of course PPR (a single regex!)
• The classic JSON parser as a single Perl Regex (although probably a little too much for this case ;-) )

Wouldn't it be better (or at least easier on the noggin) to use a two-pass approach? (Uses Perl 5.10+  \K regex extension.)

c:\@Work\Perl\monks\1nickt>perl elide_sensitive_1.pl
ok 1 - '{'password' => 'secret'}' -> '{'password' => '***'}'
ok 2 - '{'password' => 'sec\'ret'}' -> '{'password' => '***'}'
ok 3 - '{"password" => "secret"}' -> '{"password" => "***"}'
ok 4 - '{"password" => "sec\"ret"}' -> '{"password" => "***"}'
ok 5 - '{'password' : 'secret'}' -> '{'password' : '***'}'
ok 6 - '{'password' : 'sec\'ret'}' -> '{'password' : '***'}'
ok 7 - '{"password" : "secret"}' -> '{"password" : "***"}'
ok 8 - '{"password" : "sec\"ret"}' -> '{"password" : "***"}'
ok 9 - '{'user' => 'secret'}' -> '{'user' => '***'}'
ok 10 - '{'user' => 'sec\'ret'}' -> '{'user' => '***'}'
ok 11 - '{"user" => "secret"}' -> '{"user" => "***"}'
ok 12 - '{"user" => "sec\"ret"}' -> '{"user" => "***"}'
ok 13 - '{'user' : 'secret'}' -> '{'user' : '***'}'
ok 14 - '{'user' : 'sec\'ret'}' -> '{'user' : '***'}'
ok 15 - '{"user" : "secret"}' -> '{"user" : "***"}'
ok 16 - '{"user" : "sec\"ret"}' -> '{"user" : "***"}'
1..16
ok 17 - no warnings
1..17
[download]

I don't understand the significance of the  (?<!\\\\) assertion in the regex in the OP (copied as-is in tybalt89's reply). It suppresses elision of secret values when the single- or double-quoted key phrase is immediately preceded by a  \\ double backslash but not by a single backslash. Is this intended? Output from tybalt89's working code with a few minor output formatting modifications:

c:\@Work\Perl\monks\1nickt>perl tybalt89_pm1206554_1.pl
{'password' => 'secret'} -> {'password' => '***'}
{"password" => "sec\"ret"} -> {"password" => "***"}
{\'password' => 'secret'} -> {\'password' => '***'}
{\"password" => "sec\"ret"} -> {\"password" => "***"}
{\\'password' => 'secret'} -> {\\'password' => 'secret'}
{\\"password" => "sec\"ret"} -> {\\"password" => "sec\"ret"}
[download]

The double backslash thing was an artifact, not needed in the code that tybalt89 gave. Thank you for your solutions also.



The way forward always starts with a minimal test.

Here's a single-pass alternative to this (uses the same test-vector set):