"-w and strict are all i know."

Time to add another weapon to your armory:

use Diagnostics re which you'll do well to see perldoc perldoc Diagnostics (with careful attention to the discussion of splain).

Re "test case" and "fluke"-- relying on a single test case is better than non-testing, but if you have any reason to suspect something odd about your "small files" then you'll do well to expand your testing with another data set.

OTOH, given your decision to open all three files involved and your description of the (production?) files as "(v)ery large," you might want to check memory use on the production machines (<= WAG; not supported by data other than code view).