Haskell: errors as record types about bnfc

[dpwiz]

--errors=string|text|structured

I'm confused. Having string|text selector in multiple places (--text-token, maybe there are others) would give spurious combinations like Text tokens / String errors, String tokens / Text errors.

And if the token name type is driven by the existing option, this option would look like --errors=plain|structured which is basically a --structured-errors flag.

[andreasabel]

Hello Anton @vkpgwt, thanks for your suggestion and your interest in BNFC!
I think it is a very good proposal.

I suppose this is for the Haskell backend only? (The other backends also do not have structured error reporting.) You might be aware that there is a common testsuite for all backends, to ensure that the generated parsers and printers behave uniformly. (However, this does not stop individual backends to have special features.)

To keep code duplication minimal, I suggest to refactor the Haskell backend to always produce a structured error in the parser, an only convert to string errors in the top-level parser functions (unless --errors=structured is given).

I suggest --errors=string|text|structured as option name to make it less Haskell-specific (and open the avenue to have this in the other backends as well).