Compilation fixes for iOS

[kud1ing]

No description provided.

[kud1ing]

_NSGetEnviron() is private and can not be called on iOS, crt_externs.h is not available.

I am not sure about .align 2. Apple's tools do not accept .align without an argument.
I think this document implies, that the default value should be 2: http://stuff.mit.edu/afs/athena/project/rhel-doc/3/rhel-as-en-3/arm-directives.html

[adrientetar]

I'd just drop the .align conditional then since 2 is the default argument.

Also, why not replace with .align 2 in morestack.S too? This way all arm asm files have the fix.

[kud1ing]

Regarding morestack.S: i've fixed the files where i encountered an error.
I have not met this file yet.

[alexcrichton]

Would you mind adding some comments in the assembly as to what the align directives are doing? Things like this are always good to have a reason as to why they exist in the first place.

Also, I've found that xcrun doesn't have a --show-sdk-path argument, so I'm always getting lots of make errors printed whenever I do anything. Do you know if this changed in a recent version of OSX? It would be great to stop causing these errors.

[kud1ing]

I've only tried with XCode 5 (which depends on OS X 10.9) and it's present there: https://developer.apple.com/library/mac/documentation/Darwin/Reference/ManPages/man1/xcrun.1.html

We could try to support older XCode/Mac OS X combinations, but since both are free nowadays i am not sure whether it's worth making things more complicated. What do you think?

[kud1ing]

@alexcrichton: Ah, sorry. You meant, you get the errors, even when not compiling for iOS/ARM?

Do xcrun --show-sdk-platform-path or xcode-select -p work for you?
Or how about the pragmatic xcrun --show-sdk-path 2>/dev/null?

[alexcrichton]

Looks like xcode-select -p prints something relevant, but I would also be ok with the 2> /dev/null idea. It looks like this is a relic of 4.6.3, so I'm ok just piping the error elsewhere and then requiring 5.0 (if that's necessary).

[kud1ing]

@alexcrichton: Ok, can you confirm that 2>/dev/null hides the error messages for you?

[alexcrichton]

It does indeed.

[kud1ing]

@adridu59: http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0489c/Babecdje.html says Use ALIGN to take advantage of caches on some ARM processors, but this does not seem to be supported by Apple tools. I think .align 2 makes sense for iOS.

The original .align was written by @kyeongwoon, maybe he can give some input.

[adrientetar]

My point is that since these docs seem to say that .align == .align 2 (2 is default), I'd suggest to use .align 2 for everyone, without ifdefs.

[kud1ing]

I think this PR should go in as is at this point in time:

• /usr/include should not be in the include path, the SDK has its correct version of /usr/include
• many projects solve the _NSGetEnviron() issue for iOS as we do above. It is in line with how we handle it for FreeBSD
• I've added 2>/dev/null to silence the error messages on systems with older XCode
• the alignment issue is not 100% clear to me. The current version allows us to go on and investigate compilation/linking problems further down. We should come back to this and I would be more than happy if Samsung (the original authors) could give some input here. The worst case is that incorrect alignment will get us runtime crashes when running Rust-compiled iOS executables.

@alexcrichton: r?

[kud1ing]

What a pitty, it needs to be defined(TARGET_OS_IPHONE) => !(TARGET_OS_IPHONE)in src/rt/rust_builtin.cpp.
Will fix it in 8 hours.

[kud1ing]

Fixed the TARGET_OS_IPHONE check. The include path additions in mk/rt.mk slipped in by accident, but it does not hurt. @alexcrichton r?

[alexcrichton]

Doesn't this also need 2>/dev/null?

[alexcrichton]

Hm, the number of commits is growing a bit too much, would you mind rebasing into one commit and adding some more description to the commit comments about what's going on? A commit message of "fix target" isn't really too useful when we're tracing back through the history.

[adrientetar]

@kud1ing Historically, the align directive is there because ARM SoCs used to work only on aligned mode. Nowadays they support unaligned code but it's still not something common.

[alexcrichton]

I think there's still one spot which needs 2>/dev/null

[kud1ing]

rebased