RFC: Remove Delegate

[Theodus]

Remove the implementation and tests for delegates from the compiler and the explanation and examples from the tutorial.

https://github.com/ponylang/rfcs/blob/master/text/0031-remove-delegate.md

[Theodus]

I have a working implementation of this RFC, but I have a strange problem in the parser macros here. For reference here is the code in the master branch of ponyc.

[Praetonus]

What kind of problems is it causing?

[Theodus]

When running make test config=debug, if that line is commented out the following error occurs:

libponyc.tests: src/libponyc/ast/ast.c:465: token_id ast_id(ast_t *): Assertion `ast != NULL' failed.
Makefile:648: recipe for target 'test' failed
make: *** [test] Aborted (core dumped)

[Theodus]

Without config=debug, the test for BadPonyTest.EmbedNestedTuple causes a segfault

[Praetonus]

I think there is a function somewhere that still expects the old AST format for the rule you modified. The fastest way to find it probably is to run the asserting code in a debugger and to print the backtrace.

[Theodus]

I wasn't able to figure out the problem with a debugger. I did find that in flatten.c L227 provides is null and the AST of provider is (embed (id foo) (nominal (id $1) (id Foo) x ref x x) x) when it is expecting (embed (id foo) (nominal (id $1) (id Foo) x ref x x) x x). I'm not entirely sure what the extra x is, but it's the difference between the the tests passing and failing.

[Praetonus]

An x in the textual representation is a TK_NONE child in the actual AST data structure.

I don't know if you're familiar with the relationship between the parser rules and the resulting ASTs so I'll walk you through it. I'll take the rule causing you trouble as an example.

DEF(field);
  // The first TOKEN of a rule defines the node ID of the AST
  TOKEN(NULL, TK_VAR, TK_LET, TK_EMBED);
  MAP_ID(TK_VAR, TK_FVAR);
  MAP_ID(TK_LET, TK_FLET);
  // Subsequent TOKENs or RULEs create child nodes
  TOKEN("field name", TK_ID);
  // SKIP checks that the expected token is present but doesn't create a child node
  SKIP("mandatory type declaration on field", TK_COLON);
  RULE("field type", type);
  // An IF rule is optional. If the specified token isn't present, a default TK_NONE child node is created
  IF(TK_DELEGATE, RULE("delegated type", provides));
  IF(TK_ASSIGN, RULE("field value", infix));
  REORDER(0, 1, 3, 2);
DONE();

Because an IF rule always creates a child node, the resulting AST will have one child less if you remove one of these rules. The flatten_provides_list will then try to get that nonexistent child and will get NULL back.

With that change, fields won't be able to provide types anymore. To fix the problem in flatten.c, you can simply remove the calls to flatten_provides_list for fields, i.e. here and here.