To answer your specifc example..

$file =~ s%/[^/]+/\.\.$%%
[download]

... will eat the last directory name that's up'ed by the trailing ..
To give a more general/complete example that can handle multiple .'s and ..'s at any point in the path.

# colapse /./ and /.$ 
1 while ($file =~ s%/\.(/|$)%$1%g);

# colapse /dir/../ and /dir/..$ 
# ignore /../../ and /../..$
1 while ($file =~ s%/[^/]+/(?<!/\.\./)\.\.(/|$)%$1%g);

# remove ^dir/../ and ^dir/../
# ignore ^../../ and ^../..
1 while ($file =~ s%^[^/]+/(?<!^\.\./)\.\.(/|$)%%);

# colapse ^/../ to /
1 while ($file =~ s%^/\.\./%/%);

# expand null to .
$file =~ s%^$%.%;
[download]

We need the itterations of each loop as each substitution may reveal additional matchs.
We need multiple expressions to get the order of susbistion right. That is /../../ should colapse the two proceding directoires and not itself.
I'm sure it can be reduced more than that but it is as simple as I can see it can be and still catch all the special cases.
I use something very similar to this at the beginning of my scripts to get the conical path for $0 and not the called path. Although spliting the path into its components and working backwards is probably easier/clearer than using a regex.