Implement the betterC switch like DMD does.

[LemonBoy]

Namely, don't emit the ModuleInfo.
Additionally, for the following code we emit less symbols than dmd (28 vs 38) and the resulting executable actually runs as we don't emit any dso section (which explains the former) and no code that fondles with _d_dso_registry (which explains the latter).

module main;

extern(C) __gshared int __dmd_personality_v0;
extern(C) __gshared void* _d_dso_registry;

extern (C) int printf(in char*, ...);

extern (C) int main()
{
    printf("asd\n");
    return 0;
}

[JohanEngelen]

No tests!

Can you try to include a test that shows that the resulting object file can be linked without druntime?

[9il]

Other issues:

1. switch errors
2. asserts
3. cast between arrays
4. Type info
5. Check that DRuntime is 100% does not required by generated code.

[LemonBoy]

IR + Linking test has been attached.
@9il This PR implements betterC a'la dmd, it's not an effort to have a better betterC :) See here for more details.

[9il]

@9il This PR implements betterC a'la dmd, it's not an effort to have a better betterC :) See here for more details.

OK )

[kinke]

@9il: Is it just that you'd like -betterC to imply -release, getting rid of all assertions, bounds checks, contracts etc.?

[dnadlinger]

Should be okay as a DMD compatibility thing. We might want to sit down and properly discuss our "minimal runtime" to come up with a sensible and flexible design rather than Walter's one-off hack.

[dnadlinger]

Why would we require __dmd_personality_v0 in the first place?

[LemonBoy]

That's a leftover from the testing against dmd, can be safely removed.
Appveyor shows that _mainCRTStartup can't be found though, don't know why since I expect the MSVC crt to provide that anyway.

[kinke]

The MSVC libs are pulled in exclusively via the C parts in druntime lib. :/

[LemonBoy]

That's very bare metal we're running on then heh
Adding a stub declaration in a version(CRuntime_Microsoft) would do the trick, no ?

[kinke]

Yep, good idea.

[JohanEngelen]

Why is this needed? Doesn't that mean that more code emission needs to be disabled with -betterC?

[9il]

@9il: Is it just that you'd like -betterC to imply -release, getting rid of all assertions, bounds checks, contracts etc.?

betterC (in my vision) should verify that compiled code does not depend on Druntime. A code can still use some generic code from Phobos. assertions in non release mode should be replaced with inlined checks that use libc to exit from the program.

[kinke]

Please document the new switch in LDMD's usage output too.

[dnadlinger]

Could you add the description from DMD to -help, please?

-betterC omit generating some runtime information and helper functions

[kinke]

I amended the LDMD addition and pushed it directly to master (62a2252), so that it can be part of beta 4.