compile Phobos with -dip1000

[WalterBright]

Here to gauge progress in getting Phobos to work with -dip1000

[andralex]

Auto-merge toggled on

[JackStouffer]

Auto-merge toggled off

[JackStouffer]

No need to prioritize this in the auto tester just yet.

[dnadlinger]

Is there a chance for -dip1000 to cause ABI mismatch with non-dip1000 client code?

[nordlow]

@klickverbot https://github.com/dlang/DIPs/blob/master/DIPs/DIP1000.md#mangling

[dnadlinger]

@nordlow: Does it or doesn't it? The question is about throwing the switch for the same source code.

[nordlow]

You mean if the same D source code compiled into two versions of objects file, one with -dip1000 and one without, are replaceable? I'm guessing the M-qualifier is added to the mangling of parameters that have their scope inferred in the -dip1000-version but not in the other. @WalterBright and @andralex ? I guess a test is in place to be sure.

[dnadlinger]

You mean if the same D source code compiled into two versions of objects file, one with -dip1000 and one without, are replaceable?

Yes – in particular, whether a -dip1000-compiled Phobos is compatible with non--dip1000 client code.

[MartinNowak]

Yes, they are ABI compatible.

We recently fixed a related issue caused by return inference (Issue 17123 – [REG 2.073] Issues with return @safe inference).
The solution to the ABI change is straightforward, always infer return, but only enforce it when using -dip1000.

[MetaLang]

@WalterBright what's the status of this?

[JackStouffer]

See the discussion here on why this isn't viable at this point: #5915