Re: LuaJIT ignores conflicting typedefs

[Florian Weimer <fw@...>]

* Mike Pall:

> Florian Weimer wrote:
>> I wonder what the correct semantics would be.  Error out on any
>> redeclaration?  Or only if the redeclaration differs in a significant
>> way?  (Which could boil down to graph isormorphism.)
>
> C99 has strict rules for that. But most compilers use relaxed
> rules.

For duplicate typedefs?  Those seem to be allowed in C++.  But even in
C++, redeclaration of POD structs is not permitted.

The only extension in this area I'm aware of stems from the lack of
whole-program optimization: You can get away with declaring types
differently in different translation units.  But this has never been a
deliberate extension as such.

> For the FFI one would want to use different rules in some places.
> E.g. some users may (wrongly) redeclare intptr_t because they just
> copied the CPP output for the wrong architecture.

My main concern is the lack of an include guard mechanism.  For
instance, I may want to declare ssize_t because a C API I want to use
uses that type.  But another library tries to declare ssize_t, too.
What should happen in this case?

Responsibility for resolving this conflict could be pushed to library
authors, either through coordination or name mangling.  Both
approaches are a bit at odds with the FFI's appeal of simplicity
("just paste the declaration from the manpage into your sources").