List entrypoints reverse parsed lists (Haskell/C)

[gapag]

BNFC release 2.8, and up to the latest commit.
The Java backend is sensitive to the order of the macros/productions.
For instance the following gives errors, both using CUP and ANTLR4:

separator Exp ",";
List . Something ::=  [Exp];
A . Exp ::= "a";
B . Exp ::= "b";

Instead

List . Something ::=  [Exp];
separator Exp ",";
A . Exp ::= "a";
B . Exp ::= "b";

seems to work fine.
I did not try all the other backends, but Haskell seems to work without problem with both formulations.

[gdetrez]

I think the problem is not so much the order of the rules but the entry points (wich, if unspecified, will be the first category). Many backends don't seem to support having a list as an entry point.

[andreasabel]

Maybe entry points could be handled in the general layer rather in the backends.

[andreasabel]

In the generated cup file I find:

start with [Exp];

ListExp ::= ...

It seems that the start non-terminals are not translated.

[andreasabel]

The botched entrypoint in the cup file is easily fixed. (Use identCat instead of show.)
For the OCaml backend, just the generated Test program was faulty.

The Haskell and C backends print a parsed top-level list backwards. (Not an issue with Java and CPP.) The reason is that e.g. Haskell generates left-recursive Happy rules for lists, generating snoc-lists. These are reversed when plugged into other AST nodes, but not at the top-level.
A more principled solution may be to have rules for snoc-lists and then rules for the corresponding cons-lists that just apply the reversal function.