add core.stdc.assert_

[WalterBright]

For those who would want to call the C assert function rather than the druntime one.

[rainers]

Adding -dip1008 seems unrelated. There are also a couple more unrelated changes to win64.mak that cause build failures.

[rainers]

There is no _assert in the MS runtime, they use _wassert with wide strings only:

void __cdecl _wassert(const wchar_t * _Message, const wchar_t *_File, unsigned _Line);

[WalterBright]

Actually, there is. It just isn't in the .h file.

[rainers]

I see. A comment would be nice for the next time I stumble over this.

Should we add the _wassert version, too, similar to the various declaration on the other platforms?

[WalterBright]

yes

[rainers]

Not a blocker, but the documentation only lists the glibc version declarations. This can be pretty confusing for other platforms. The ddox documentation doesn't list any function at all.

[PetarKirov]

pure?

[WalterBright]

I'm not sure that a function never returning means it is pure.

[PetarKirov]

As far as I understand, this provides an alternative to built-in assert. Since the built-in assert is usable from pure functions, I think this should also be.

I'm not sure that a function never returning means it is pure.

That should never happen in correct programs. But in general pure functions are allowed to terminate the program, according to the language spec. So I think the C assert functions fit the bill.

[rainers]

I agree with @ZombineDev, should be pure. The betterC purists will come back at you anyway without it.

[WalterBright]

Pure functions that return void can be elided by the compiler.

[rainers]

I think this would be more useful with a template c_assert or similar that translates the C macro assert for the different platforms.

[WalterBright]

I think this would be more useful with a template c_assert or similar that translates the C macro assert for the different platforms.

The purpose of these imports is not to fix the interface, but to simply declare it.

[rainers]

Pure functions that return void can be elided by the compiler.

If these functions are actually elided, then we'll have to go without pure for now. I guess we need a solution for this, though. There are already unintuitive hacks in druntime to work around this, e.g. onOutOfMemoryError.

[rainers]

The purpose of these imports is not to fix the interface, but to simply declare it.

The C interface is assert, though, not __assert_rtn. Other core.stdc files translate the macros, too. See https://github.com/dlang/druntime/blob/master/src/core/stdc/stdio.d#L1068 for an example.

[WalterBright]

I guess we need a solution for this, though.

The solution is a noreturn function attribute of some sort.

[WalterBright]

The C interface is assert, though, not __assert_rtn.

We also declare the internal functions in the core.stdc.* imports.

[rainers]

We also declare the internal functions in the core.stdc.* imports.

Sure, I never said these should be removed. But without an implementation of the C assert macro, everyone will have to implement their own version forwarding to the platform specific functions.

[rainers]

The solution is a noreturn function attribute of some sort.

This will also need a mightnotreturn attribute if the termination is conditional.

[WalterBright]

This will also need a mightnotreturn attribute if the termination is conditional.

Every function is mightnotreturn. It's OT anyway.

[rainers]

Every function is mightnotreturn. It's OT anyway.

If that's true you must not elide any pure function if you want to keep the side effect of noreturn pure calls.

[rainers]

Let's move forward with this. A C assert wrapper can be added later if the need pops up. We have lived without it for years.