Date: Wednesday, August 12, 2020 2:37 PM
Lassi gave us a great example of why 'code is subpar compared to 'number:
After searching around more, it seems to be true for every facility there is either an integer code, a symbolic name, or both. I was trying to find situations where there is a non-numeric code (something that can't be interpreted by a human) as well as a name. But that doesn't seem to be a pattern actually in use for errors.
Can't recall any myself. But there's also the simple library example like libsodium where you get 0 or -1 returns, and that's the limit of it's organization, no code, no per error type number, no name distinct from the C function that returned -1.
I think rather than have a distinct 'libsodium 'facility, I'll coin one like 'standard-unix-lib. Because its very few conventions are going to be the same as many other libraries and perhaps other systems.
So I can live with 'number and 'name.
For errno and errno define; good.
Maybe "for a (possibly localized or localizable)" since we may require the user or IDE/whatever to go to some effort, plus the English version must always be accessible for searching?
Done. However, as I keep pointing out, in a client/server situation you may not be able to get text that is not localized after the fact.
[ Other minor changes made. ]
All these changes are pushed to johnwcowan now.
I'll push a change to the Chibi Scheme sample implementation for set-file-owner soon.