Zig: make the C pointer type always a compile error unless the file declares that it is generated code

[andrewrk]

Unsurprisingly, the relaxed rules about C pointer types have resulted in people using them where other, more restricted types, would be preferred. Perhaps it's time for the language to get more involved in policing this.

[daurnimator]

Do you have a link to examples of people using it as such?

[andrewrk]

How this will work:

• New keyword: thisfileisautotranslatedfromc. Nobody wants to type that!
• The keyword may be the first token in a file, followed by a semicolon. It is illegal otherwise.
• If the keyword is found, then a file is allowed to contain [*c] pointers. Otherwise, the compile error is error: [*c] pointers are only allowed in auto-translated C code, note: * and [*] pointers may be used where a [*c] pointer is expected.

[marler8997]

Another use case for using C pointers is for testing, i.e.

https://github.com/ziglang/zig/pull/3111/files#diff-9bb6309b18efdcc04260fad6ad9f55edR83

[marler8997]

So would we add the thisfileisautotranslatedfromc token to this test file...even though it isn't translated from C?

https://github.com/ziglang/zig/pull/3111/files#diff-9bb6309b18efdcc04260fad6ad9f55edR91

I could be missing something, but there seems to be more valid use cases for C pointers besides translating from C. If you're interfacing with a C library, maybe you need to write some utility functions that help bridge the GAP between Zig and C? It may make sense to use C pointers to add special handling for those specific pointer types?

[marler8997]

One thought came to mind. You could also always allow C pointers in test blocks. That would solve the problem of using C pointers in test code. But still not sure whether or not there are other valid use cases for C pointers...

[daurnimator]

I had a thought today re: C pointers: all [*c]T are currently nullable. What if we instead translated C pointers as ?[*c]T, which would be a "optional C pointer". (you could get a plain [*c]T if the C source had __attribute__((nonnull)))
That was [*c] only means "unknown if single pointer or pointer to many"?

[andrewrk]

Another example today:

<riba> i need to provide a callback to C code that takes a ** argument
<riba> i tried it like this: [*c][*c]
<riba> now i want to check both pointers to null
<riba> and "if (stanza == null || stanza.* == null)" does not work

[emekoi]

perhaps, we could add a note suggesting to use a different pointer type, rather than adding a new keyword. e.g. note: * and [*] pointers may be used where a [*c] pointer is expected.

[hryx]

Random thought: rename [*c]T to [*ambiguous]T or even [*autotranslated]T

[Rocknest]

@hryx [*ambiguous]T seems descriptive plus awful to type which probably is good for C pointers :+1: