Type erl_anno:anno() should not be opaque

[josevalim]

The docs
refer to the annotation simply as LINE, however, if we were
to add the lines manually and give it to erl_eval:expr/2,
Dialyzer will complain as erl_anno:anno() is listed as opaque.

This makes erl_anno:anno() as non-opaque, with the
location() as public type but the annotation() as private.

[kostis]

What exactly do you mean by

if we were to add the lines manually

Write just 42?

Why can't you write erl_anno:new(42) instead as e.g. PropEr does all over the place? (For example, search for ?anno in proper_typeserver.

Dialyzer is rarely an excuse to break programming abstractions.

[josevalim]

I am aware. If LINE is no longer accepted, then the docs have to be updated and it needs to be listed as a breaking change. At the moment, the docs and specs are inconsistent.

[kostis]

Yes. IMO it's the docs that need to be updated -- they should have been when the erl_anno module was introduced (18.x ?)

[uabboli]

I agree with Kostis. I'll write a ticket on updating the docs.

If you cannot find the cause of Dialyzer warnings and suspect erl_anno i involved, you can compile erl_anno with DEBUG. erl_anno will then only work with annotations that are lists.

[josevalim]

Thanks everyone! I will close to this then and make sure we are using erl_anno across the board.