"unique" filename?

[le]

Hi folks,

I'm looking for a module or some other algorithm, that can generate unique filenames (and maybe returns the filehandle).

I'm currently working on a web app, that writes a file into a directory. The problem is, that the filenames should be unique, so if the script attempts to create two files at the same time the files don't clobber.

I hope you get what I mean, something like "I need a new file, and I think the filename 99887766 is ok, but if there's already 99887766, give me 99887767."

[quidity]

The module File::MkTemp seems to be what you want:

use File::MkTemp;

$dir = '.'; # directory to create file in
$string = mktemp('tempXXXXXX',$dir);
open(F,$string);

Will return a unique file name, you can also have it return a unique file handle so that there is no chance of opening an identical file in two instances of your script:

use File::MkTemp;

$dir = '.'; # directory to create file in
$fh = mkstemp('tempXXXXXX',$dir);
print $fh "stuff";
$fh->close;

You can add more 'X's, but you need at least six.

[Fastolfe]

For generating, say, unique temporary files, use IO::File::new_tmpfile. In addition there's the POSIX::tmpnam() function which would get you a unique filename (subject to a race condition, however, since there's a period between when it's generated and when it's used).

In a pinch, use something like this:

$num = 12345;
$dir = "/some/dir/";

$num++ while -e "$dir/$num";

open(F, ">$dir/$num") or die ...

But of course, there's still a race condition. A more secure solution would be to take advantage of sysopen and O_EXCL to iterate through filenames:

use Fcntl;
use FileHandle;

$num++ while !sysopen(F, "$dir/$num", O_EXCL|O_CREAT, 0777) && $! eq "
+File exists";

die "$dir/$num: $!" if $!;

Once you start adding all of the bells and whistles, though, you're just re-implementing IO::File new_tmpfile.

[2501]

how about the following:


$filename = $$.time;

This is a combination of process ID and time. This would prevent duplicate file names from being a possibility.

[myocom]

Actually, that wouldn't prevent duplicate filenames at all, depending on how fast you're generating filenames. For example, look what the following does:


while (1) {
    print $$.time."\n";
}


Also, on Win32, there's a non-zero chance of a PID being reused.

[2501]

When you log into a site, you have two options.
You can start running a script from a new connection or you can use an existing connection.
Since only one person can be using a PID at a time, the combination of time and PID is fine.

[the_slycer]

I use the following, which seems to work well (though results in some horribly long file names).

my $outfile = localtime;
print "$outfile\n";
$outfile=~s/\s|://g;
$outfile="$outfile" . rand($outfile);
$outfile=~s/\.//g; #for win32 - doesn't like multiple dots
print $outfile;

[tune]

nobody answered the original question. (that is what i learned from revote hearings CNN Live...:)))

do {
$newfilename = $$.time();
} until (!-e $newfilename);
print $newfilename;

# and then do your stuff

-- tune

[merlyn]

Since $$ doesn't change, you've effectively created a busy-wait sleep. Why not just stick a sleep 1 in there instead?

-- Randal L. Schwartz, Perl hacker

[tune]

but the existing file can be removed by another thread in less time than 1 second. I understand why I have got downvoted, but the point is not on the method of creating the unique filename it is on how to avoid overwrinting the existing file.

-- tune