Compiler: mark LLVM return type as Nil when program ends with no typed ASTNode

[maiha]

Compiler crashes when the crystal code ends with no typed ASTNode such as @[Foo].

test.cr

@[Foo]

run

% crystal -v
Crystal 0.19.0 [dcfb2b6] (2016-09-02)

% crystal test.cr
(...snip...)
@[Foo]
` at  has no type
[8917383] *CallStack::unwind:Array(Pointer(Void)) +87
[8917274] *CallStack#initialize:Array(Pointer(Void)) +10
[8917226] *CallStack::new:CallStack +42
[8756568] *raise<Exception>:NoReturn +24
[8756542] ???
[13486612] *Crystal::ASTNode+ +148
[21386979] *Crystal::CodeGenVisitor#initialize<Crystal::Program, Crystal::AS
TNode+, (Array(String) | Bool | Nil), Bool, LLVM::Module, Bool>:Bool +339
[21386606] *Crystal::CodeGenVisitor::new:single_module:debug:llvm_mod:expose
_crystal_main<Crystal::Program, Crystal::ASTNode+, (Array(String) | Bool | N
il), Bool, LLVM::Module, Bool>:Crystal::CodeGenVisitor +222
...

I think this will occur in all crystal versions because I got same errors in 0.18.7.

PR

This PR uses nil for LLVM return type when typeless node appears in the last sentence.

% ./bin/crystal -v
Using compiled compiler at .build/crystal
Crystal 0.19.0+24 [4e00257] (2016-09-08)

% ./bin/crystal test.cr
Using compiled compiler at .build/crystal
# (→ no errors )

Although it's nice that the attributes should be typed correctly,
we also need this fix because compiler must not crash in any cases we don't predict.

make clean crystal spec has been passed.

Thanks.

[asterite]

Please use node.type? || @program.nil_type

[asterite]

Thank you, looks good! I just made a small comment. Maybe a spec would be nice, but I think it breaks in the spec code because it assumes the node will have a type, so that would need to change too...

[maiha]

Hi asterite!

Thank you for your advice. I fixed about it and pushed again.

I think it breaks in the spec code because it assumes the node will have a type

Well, could you tell me more about this spec file?
Although I've already passed make clean crystal spec, it would be I missed some steps
for all tests because it's first time to test compiler for me.

I'd like to try to fix the spec too.

Thanks!

[asterite]

@maiha Don't worry, the change is OK. We could maybe need a spec like codegen("@[Foo]") and see that it doesn't break. In any case these programs are not very useful, but it's better if the compiler doesn't break.

Thank you!