Re: [klibc 07/31] i386 support for klibc

["H. Peter Anvin" <hpa@zytor.com>]

Roman Zippel wrote:

>> The way libgcc is handled inside gcc is, indeed, completely screwed up; even
>> the gcc people admit that.  They pretty much don't have a way to handle the
>> effects of compiler options on libgcc, especially the ones that affect binary
>> compatibility.
> 
> Nobody said it's perfect. Especially the last point speaks against 
> multiple versions of the same library, as it makes it hard to mix 
> binaries/libraries. With a single kinit binary it's not really a problem 
> yet, but will it stay this way?

What on earth are you talking about?

a. The semantics of these functions are well-defined, stable, and 
documented in the gcc documentation.  It's not like they have 
compiler-version-specific definitions that could change.

b. For static binaries, this is no issue.  klibc is shared, not dynamic 
(thus eliminating the need for a space-consuming dynamic linker), but it 
also means that there is no cross-version calling; each build of the 
shared klibc library has a hashed filename, thus allowing multiple 
versions of klibc to coexist if absolutely necessary.

Either way, this is a red herring.

>>> The standard libgcc may not be as small as you like, but it still should be
>>> the first choice. If there is a problem with it, the gcc people do accept
>>> patches.
>> That's just an asinine statement.  Under that logic we should just forget
>> about the kernel and go hack the gcc bugs du jour; we certainly have enough
>> workarounds for gcc bugs in the kernel.
> 
> Sorry, but I can't follow this logic.

I'm not entirely surprised.

	-hpa