Added structure-valued expressions

[mihaibudiu]

This should be merged after #717

[mihaibudiu]

Fixes #772

[vgurevich]

Can we also agree that if a field is not mentioned in kvlist, it will be automatically assigned a 0 value and also agree that kvlist can be empty, meaning that:

mystruct_t mystruct = {};

will do the right thing?

[mihaibudiu]

@vgurevich : There is a discussion on this topic #717
I have added this as a topic to discuss in the LDWG.

[liujed]

This creates the awkward situation where all of the following are valid syntax:

• { a=0, b=1 }
• S { a=0, b=1 }
• (S) { a=0, b=1 }
• (S1) S2 { a=0, b=1 }

One way around this awkwardness would be to only allow the '{' kvList '}' production and remove the typeName '{' kvList '}' production, but then typing of struct-valued expressions can become ambiguous when the program has defined multiple types that match.

Another alternative would be demote '{' kvList '}' from being a first-class expression, and only allow it as a struct initializer: S s = { a=0, b=1 };. Then, struct-valued expressions would always use the S { a=0, b=1 } form. My impression is that this would be compatible with what we released in 1.2.0, but I haven't confirmed. (This is what I was previously advocating for, but I guess that wasn't clear.)

[mihaibudiu]

the typechecker will reject today casts with struct types.

[mihaibudiu]

maybe that's a good rule: if a cast is to a Struct or header type then it is only allowed for struct-valued expressions.

[mihaibudiu]

We should try the syntax with a cast instead of a type.

[jafingerhut]

Would it make sense that if/when this is merged in, it should replace the text about structure initializers? It seems that this proposal does everything that does, and more.

Also, is it fair to say that while this proposal is not identical to "Compound literals" in C/C++, it is pretty close?

• https://gcc.gnu.org/onlinedocs/gcc/Compound-Literals.html
• https://www.drdobbs.com/the-new-c-compound-literals/184401404

[jafingerhut]

LGTM