For performance and sanity my routine will first do a test for file-existance and then try to glob it. It feels clumsy, though.

You need to check for file-existance afterwards, not beforehand. glob is not guaranteed to produce a list of existing files. For example, glob('file{a,b}') will return filea and fileb whether either, neither or both exists.

@list = bsd_glob('~gnat', GLOB_ERR);

die("Unable to glob: $!\n")
   if GLOB_ERROR;

# Keep only references to existing files.
@list = grep -f, @list;

die("glob returned no existing results\n")
   if not @list;
[download]

I am still confused about the way $! behaves in this case

This case is not special.

# XXX WRONG
$rv = print $fh (...);
if ($!) {
   die("Unable to write to file: $!\n")
}

# Ok
$rv = print $fh (...);
if ($rv) {
   die("Unable to write to file: $!\n")
}
[download]