RFC: FSF GCC, #import (and #pragma once)

[David Ayers]

Hello again...
hello Nicola,

(if I can't bark up the gdb tree, I'll try gcc ;-) ) no seriously...

I'm meaning to post this to the gcc list and just wanted to give people 
here (esp. Nicola) a chance to comment orvoice any reservations of it 
being counter productive.

Thanks,
Dave

-------- Original Message --------

Hello,

Please excuse me for not replying to the original thread, as I'm not on 
the list, just lurking through the archives.

I must admit that I'm not sure I understand the recent discussion about 
the semantic subtleties of the #import directive.  If I have understood 
the past development correctly, the FSF GCC Team (I'm not sure which 
whether this was a general consensus or a SC decision) had already 
identified #import as problematic and discouraged it's usage through 
deprecation with the intent for removal.  Reality has it, that there is 
a large amount of active (mostly ObjC) projects that rely on this 
directive, with the semantics of whatever the existing implementations 
actually implement.  These are largely projects that are expecting the 
semantics currently implemented by Apple's GCC fork, whatever they may be.

If Geoffery Keating is working on syncing the implementation of Apple's 
fork with the FSF GCC, then isn't what ever he implements "perfect" by 
definition?  Not because it catches all edge cases that anyone may think 
of and that some might interpret differently than others, but because 
the FSF GCC and Apple's fork will implement the *same* semantics, what 
ever they may be.

(I must admit though that I have no idea how many projects rely on 
#pragma once and expect certain edge conditions that might differ from 
the pending implementation.)

I think the bottom line isn't to improve a feature that has been 
identified as potentially broken by design, but it is much more in the 
interest of the FSF to provide a compiler that is *compatible* with 
Apple's fork on this issue, so that all those lines of code currently 
written for Apple's fork can become free software by being compiled with 
FSF GCC and GNUstep with as little modification as possible.

Note that I'm neither promoting nor discouraging the use of #import.  
And if the FSF deems it broken by design, then let the manual state 
that, and discourage the usage.  Add warnings all you like (I'm sure 
Apple's fork will suppress them, and it would be nice if FSF GCC could 
also suppress them as it would only annoy porters.) but please don't 
make it unnecessarily harder for all those Cocoa projects to be ported 
to GNUstep.

Removing #import or having radically different semantics than Apple's 
fork is a small annoyance for Apple as they merely have to maintain a 
patch for their fork, but it could very likely be a show-stopper for 
large ports to GNUstep.

Thank you for you attention,
David Ayers

PS: Oh, and the issue about allowing it only for ObjC and not for C++... 
Keep in mind that Apple's GCC supports ObjC++ (i.e. mixed languages in 
source files) and I believe that there was effort pending to include 
this feature in FSF GCC once the preprocessor issue was resolved.