[haiku-development] Re: GCC2 and GCC4 on same System
- From: Oliver Tappe <zooey@xxxxxxxxxxxxxxx>
- To: haiku-development@xxxxxxxxxxxxx
- Date: Mon, 27 Aug 2007 18:23:59 +0200
Ingo Weinhold wrote:
> On 2007-08-27 at 07:12:34 [+0200], Michael Lotz <mmlr@xxxxxxxx> wrote:
>> On Mon, 27 Aug 2007 01:06:55 +0200, kamikazow wrote
>>> Hm... what about placing GCC 2.9x ELF binaries in /boot/beos and GCC
>>> 4 non-ELF universal binaries in /boot/haiku?
>> I don't see why we should have universal _binaries_ if all we need are
>> actually two versions of the libraries. The binary shouldn't really care,
>> the
>> loader just has to resolve the right symbols from the right libraries. So
>> we'd
>> rather go with universal _libraries_ here.
>> I don't know if it is technically possible, but would it be doable to
>> actually
>> link a library with objects from both GCC2 and GCC4? I mean the C functions
>> would collide, but they are compatible anyway. The C++ stuff wouldn't
>> collide
>> and both versions could exist side by side in a single library. The ELF
>> format
>> shouldn't care and the runtime_loader still just resolves the symbols
>> necessary, just that they are different for GCC2 and GCC4 apps. Crazy?
>> Probably it's too early in the morning for me to think...
>
> Doing that for C++ is definitely non-trivial. The C++ ABI changed
> significantly. The name mangling changed -- which is probably why you
> suggest there wouldn't be any collisions -- but I wouldn't rule out that
> individual symbol names could still be mangled the same way. Furthermore
> global stuff like initialization code and exception handling changed, too,
> so this would definitely need to be tinkered with. I've only relatively
> little knowledge about these things, but, I believe, enough to know that I
> wouldn't want to do that.
Yes, I am pretty sure that this scheme doesn't work, as not *all* symbols
are mangled differently, just *some* of them.
> At least, I think, it isn't really worth the hassle. When we build
> distribution supporting both compilers, it will very likely be heavily
> biased towards one compiler version, i.e. with only few exceptions
> everything will be compiled with the same one. I don't think we need to
> overly worry about replicants and mixing + matching add-ons.
I agree wholeheartedly, and without having thought about it much, I
currently favor using gcc4 as main compiler and provide a backwards
compatibility hack. After all, the aim is to phase out gcc-2.95.3 support
soon, right?
cheers,
Oliver
Other related posts:
