POSIX _t Namespace Conflict

[wahern]

Not many people realize this, but POSIX reserves all identifiers with the suffix "_t". See http://pubs.opengroup.org/onlinepubs/9699919799/functions/V2_chap02.html

I suspect people tend to use the _t suffix precisely because it's so ubiquitous in C and POSIX identifiers, but it's ubiquitous in those standards for the reason that it's supposed to avoid conflicts with other namespaces.

[rossberg]

Hm, interesting, but the way I read it is that the restrictions stated there constrain implementations of POSIX and what implementation-specific symbols they can add to POSIX header files. AFAICS they do not constrain third-party libraries with independent header files.

For the Wasm API, all C names have consciously been given a wasm_ prefix, so I expect the potential for conflict to be minimal to non-existent.

[wahern]

The suffix is reserved for all POSIX and C headers. (Note the [CX], which means it's extending the C standard, which means its reserving for C headers as well.) If your code never includes a header like <stdlib.h>, then you're good. I've written C source file that didn't include any C or POSIX header. But if your code includes such a header and also defines identifiers (directly or indirectly) with the _t suffix, then there's potential for namespace clash, especially if you're defining them in headers to be used by other applications.

In practice few people know this, let alone obey it. But I figured standards should probably be neighborly.

[rossberg]

Sigh, modularity and name spacing are such a sad story in C.

The only unqualified type names we define are the byte_t, float32_t, and float64_t synonyms. I agree that we should probably avoid those. (It's unfortunate that they do not exist in any standard header.)

For everything else I don't know. We indeed include some C headers. However, as mentioned, all our names are sanitised with wasm_ prefixes. If POSIX makes such a sweeping land grab in the C name space that it even prevents harmless uses like that -- and generally is in conflict with large parts of dev reality -- then I don't know if third parties should feel obliged to obey to the letter. And I'm saying that as a person who usually very much prefers conforming to standards

[wahern]

FWIW, the use of _t in the API doesn't really add anything. For years, before I read and understood the standards, I thought _t was an idiom for scalar typedefs (integers, pointers) or more generally (in my rationalizations) POD types that could be copied by simple assignment, and that's how I always used it. IME most software that uses _t rarely uses it for aggregate objects like structs or unions. What originally clued me into the issue was how the pthreads API uses _t for aggregate objects, including ones that couldn't be copied, which seemed irregular until I learned that the whole point of _t was for namespacing, not readability.

That said, I only wanted to bring the issue to your attention. I don't care to bikeshed style. Feel free to dispose of this issue however you want ;)

[rossberg]

Actually, the main benefit of the convention from my perspective is that you don't get into syntactic issues where you want to name a variable or function after a type, because C syntax does not properly separate the name spaces.

Anyway, thanks for bringing this up. I at least want to reconsider the use of unqualified type names.