search, replace and backrefrences

[Helter]

I found a solution that works, but to gain more understanding (and I'm stubborn that way) I want to know why my first solution does not work.

I'm parsing a file, looking for a single line to replace 2 characters.
Line in question:

__msr2f6:       data8 0b00000000000000000000000000_1_0_000000000000001
+0101_00000000000000000

My initial search/replace was this:

$line =~ s/^(__msr2f6:\s+data8 0b[01]+_)[01]_[01](_[01]+_[01]+)/$10_0$
+2/;

This causes an error to be printed:

Use of uninitialized value in concatenation (.) at myscript.pl line 33
+, <FILE> line 915.

A co-worker suggested a change, but didn't know why it might work:

$line =~ s/^(__msr2f6:\s+data8 0b[01]+)_[01]_[01]_([01]+_[01]+)/$1_0_0
+_$2/;

So I moved the leading and trailing underscore out of the parens and put them in the replace string.

Here are some things I think I understand, but then again maybe not...

1. Backrefrences are 1-9, only a single digit.
2. Underscores are valid variable name components.

So what changes between these two examples? If the first variable is being treated as $10_0, then why is the second not being treated as $1_0_0_?

This solution also works:

$line =~ s/^(__msr2f6:\s+data8 0b[01]+_)[01]_[01](_[01]+_[01]+)/$1."0_
+0".$2/e;

But it is probably slower than the other working solution.

Thanks for the enlightenment!

[diotalevi]

I suspect perl is DWIMing your potential $1_0_0_ into $1. The actual fix is to just encose the '1' in curlies like s/ ... /${1}0_0$2/. You use curlies to disambiguate interpolated variables where needed.

[demerphq]

Well, im not real sure that $1_0_0_ isnt being treated as such.

But a useful trick in this situation and others is to use the form ${varname} to delimit the the identifier. Ie

my $foo="Groove";
print "$food";   # Error $food is not a declared indentifier.
print "${foo}d"; # same as $foo."d";

So the rhs of the first and second could be ${1}0_0$2 and ${1}_0_0_$2 respectively.

--- demerphq
my friends call me, usually because I'm late....

[rir]

Well, im not real sure that $1_0_0_ isnt being treated as such.

Variable names that start with a digit may only have
more digits in the name.

This seems a murky area. my $99; throws an error.

This post has little to do with the original question.

[diotalevi]

Even with the number formatting semantics where 1_000_000 == 1000000?

[demerphq]

Variable names that start with a digit may only have more digits in the name.

Cool. I didnt know that. What perldoc is it in?

This seems a murky area. my $99; throws an error.

Well, i agree its weird, but its not that murky. All of the ^\d+$ variables are global only and cannot be my()ed (and all of the scalars are read only).

--- demerphq
my friends call me, usually because I'm late....

[Helter]

After some more reading I found that my first assumption is only half true.
$1-$9 are ALWAYS backrefrences, where $10 and up are only backrefrences if more than 10 parens were used.
This mixes with the comments about a variable that starts with a digit, can only have digits.

So $10 is taken as the whole name of a variable (undefined)
$1_ gets treated like $1 because _ is not a digit.

Thanks for all the help, I'm slowly understanding the subtle nature of Perl :)