What does exactly perl -pi -e do?

Question

I would like to know what is the equivalent code that Perl runs when executed with the options perl -pi -e?

On some SO question I can read this:

while (<>) {
    ...     # your script goes here
} continue {
    print;
}

But this example does not show the part where the file is saved.

How does Perl determine the EOL? Does it touch the file when no changes occured? For example if I have a old MAC file (\r only). How does it deal with s/^foo/bar/gm?

I tried to use the Perl debugger but it doesn't really help. So I am just trying to guess:

#!/usr/bin/env perl

my $pattern = shift;
map &process, @ARGV;
# perl -pi -e PATTERN <files>...
sub process {
    next unless -f;
    open my $fh, '<', $_;
    my $extract;
    read $fh, $extract, 1024;
    seek &fh, 0, 0;
    if ($extract =~ /\r\n/) {
        $/ = "\r\n";
    } elsif ($extract =~ /\r[^\n]/) {
        $/ = "\r";
    } else {
        $/ = "\n";
    }

    my $out = '';
    while(<&fh>) {
        my $__ = $_;

        eval $pattern;

        my $changes = 1 if $_ ne $__;
        $out .= $_;
    }

    if($changes)
    {
        open my $fh, '>', $_;
        print $fh $out;
    }
    close &fh;
}

Answer 1

You can inspect the code actually used by Perl with the core module B::Deparse. This compiler backend module is activated with the option -MO=Deparse.

$ perl -MO=Deparse -p -i -e 's/X/U/' ./*.txt
BEGIN { $^I = ""; }
LINE: while (defined($_ = <ARGV>)) {
    s/X/U/;
}
continue {
    die "-p destination: $!\n" unless print $_;
}
-e syntax OK

Thus perl is looping over the lines in the given files, executes the code with $_ set to the line and prints the resulting $_.

The magic variabe $^I is set to an empty string. This turns on in place editing. In place editing is explained in perldoc perlrun. There is no check whether the file is unchanged. Thus the modified time of the edited file is always updated. Apparently the modified time of the backup file is the same as the modified time of the original file.

Using the -0 flag you can set the input record separator for using "\r" for your Mac files.

$ perl -e "print qq{aa\raa\raa}" > t.txt
$perl -015 -p -i.ori -e 's/a/b/' t.txt
$cat t.txt
ba
$ perl -MO=Deparse -015 -p -i.ori -e 's/a/b/'.txt
BEGIN { $^I = ".ori"; }
BEGIN { $/ = "\r"; $\ = undef; }
LINE: while (defined($_ = <ARGV>)) {
    s/a/b/;
}
continue {
    die "-p destination: $!\n" unless print $_;
}
-e syntax OK