Make Ulf Norell's define example work in all backends

[andreasabel]

The original example for define by Ulf Norell is currently only accepted by the Haskell backends.

bnfc/examples/DefinedRules.cf

Lines 22 to 23 in 3da1b0e

	define if e s = If e s (Block []) ;
	define for i c s b = Block [i, While c (Block [b, s])] ;

• Agda: wrong list constructors
• C: list constructors rendered wrongly as make_(:) and make_[].
• CPP: missing sanitization of if and for (keyword clash, regression), wrong list constructors.
• Ocaml: missing sanitization (regression), list cons (:) (a,b) instead of (a :: b).
• Java: missing sanitization (regression), complaint that if clashes with If. Works in 2.9.1 with --force.

Missing sanitization is a regression wrt. 2.9.1, introduced by #287.
The list constructors probably never worked.

[shlevy]

@andreasabel We just hit:

C: list constructors rendered wrongly as make_(:) and make_[]

Any pointers to fix?

[andreasabel]

I am now working to extend the expression type by typing information, so that the printers can pick the right list constructors.

[andreasabel]

Remaining problem (C): non-linear definitions, like:

bnfc/examples/define/test.cf

Line 25 in 228e2ef

define inc x = Assign x (EOp (EVar x) Plus (EInt 1)) ;

This creates syntax trees that not trees but DAGs.

  make_Assign (x, make_EOp (make_EVar(x), make_Plus(), make_EInt(1)))

E.g. here, x is shared. This causes problems for the C deallocator which assumes that it is dealing with a tree.

[shlevy]

the C deallocator

A bit off topic, but is there a deallocation function provided with the generated files? I haven't seen one, I've just been manually freeing recursively down the generated structs.

[andreasabel]

@shlevy : The deallocation is a new, yet unreleased functionality added in a9d360d (#348).

[andreasabel]

Remaining problem (C): non-linear definitions, like:

This has been fixed by introduction of clone_* and the cloning of duplicate components to prevent pointer aliasing and to preserve the tree-property.