Yes, you're definitely right. I did automatize the process of retrieving the entities from the webpages but I parsed it from the .html files actually. :-) And I use bash, wget and perl for it, which I find much more comfortable in this case than pure perl.

I surrounded the things in subs so it won't get evaluated if someone is only interested in one set and not in others. I should definitely cache the functions' return values. And are you sure it's better to add the semicolons by a map? I didn't really benchmark it, but I think this should be a bit faster and the code is simpler and more transparent, if twice as big. Dunno... it seems to me this is the better way but I am certainly open to counterarguments.

Update: Looking at the .ent files, I see now that half of my effort was needless. :-) Oh well...