fix -c -inline ddmd/mars.d

[MartinNowak]

• fails when compiling because ddmd/mars.d imports ddmd.objc and the
  modulename != filename hack doesn't work for single file
  compilation, so ddmd/objc.d (instead of ddmd/objc_stub.d) gets
  imported and fails on a OSX-only symbol
• fixed by always defining objc_initSymbols and adding it to the stubs

[MartinNowak]

While you're at it, may you have a look at this as well?

[CyberShadow]

fails when compiling b/c ddmd/mars.d

Please avoid abbreviations in commit messages - in that context especially it can be confused by a path or identifiers.

Aside from that, this (DMD code and especially Objective C stuff) is really not up my alley... CC @jacob-carlborg

[MartinNowak]

Blocking proper release building (docs).

[CyberShadow]

I'll enable auto-merge since this is a blocker.

[jacob-carlborg]

LGTM 👌

[jacob-carlborg]

Out of curiosity, why did this start to fail now?

[jacob-carlborg]

BTW, I'm working on a better way to select if the Objective-C integration is enabled or not [1], although it seems difficult to come to some form of agreement.

[1] http://forum.dlang.org/post/2D143DD8-A978-4C60-9C78-488E19802B88@me.com

[CyberShadow]

Strange, why is the commit CI status red? Travis seems to have failed but I don't see any failures on https://travis-ci.org/dlang/dmd/builds/216000505 ? Why did @dlang-bot merge it despite it being red?

[jacob-carlborg]

The macOS builds where canceled?

[CyberShadow]

Yeah, why?

[wilzbach]

Quite simple, because those CIs aren't set as required yet and the bot only enforces the status of the required CIs. I have been pushing for setting all CIs except for CodeCov to required, but this has always been blocked with "our CIs have too many random failures and are unreliable"...

[CyberShadow]

our CIs have too many random failures and are unreliable

Well, are they?

[CyberShadow]

I think @dlang-bot should still wait for non-required CIs too, and not auto-merge if any CIs, required or not, have failures. If the non-required CIs have failures, a human can look at the failures, decide whether there's a problem, and merge manually.

[wilzbach]

Well, are they

Hehe, I have done a couple of these statistics in the past. In short our infrastructure isn't optimal, so we do get some random connection timeouts and there are still random failures hidden in Phobos, but my point has always been that if we can't fix these issues how do we expect normal D projects to work? Moreover, by enforcement we can no longer ignore the fact that we have to eat our dog food from time to time.

I think @dlang-bot should still wait for non-required CIs too

I agree: dlang/dlang-bot#69

[MartinNowak]

Yes, this was a slight design mistake we did when implementing the auto-merge feature. At that point there were a lot more failing/WIP CIs.

Out of curiosity, why did this start to fail now?

Think it's b/c we started to build ddmd docs (or maybe build them differently or so).