[ILUG] Can anyone see any reason why a customer would need to compile a new compiler?

[Declan Moriarty]

On Sun, 2006-06-25 at 23:55 +0100, paul at clubi.ie wrote:
> On Sun, 25 Jun 2006, Declan Moriarty wrote:
> 
> > Yes, actually you are right.
> 
> Well, "paul" and "right" are two words often used in the same 
> sentence...
> 
> > But unless you know the system intimately at a code level, you 
> > won't have a great idea of what those bits of the system are, will 
> > you? Hence the all or nothing approach is best. Or, you could grok 
> > all the code...
> 
> Well, no. Again:
> 
[snip]
> 
> IIRC: at least one distro only /selectively/ compiled network-enabled
>      applications with SSP, precisely due to the above reasoning ;).
> 
Well, I'm not going to argue that stuff with you Paul. The guy wanted to
know why a customer would want gcc-4.1, and I told him. Please mention
what distro selectively installs ssp and I'll avoid it. I would prefer
to have it throughout if going to the pain (and it is pain) of compiling
a hardened system. largely because I would have to go on the word of
others about where I needed it. No soldier of old would go into battle
saying "Most blows are to my head", and leave off the rest of his
armour :-). It is very easy to put together a half hardened system, but
more difficult to predict if some future hack will find a way of
stepping through the non hardened bits.

> > And the biggest word in there was the "IF". I think it extremely 
> > unlikely that a vanilla gcc-4.1 will compile on any sort of an old 
> > system.
> 
> I havn't dug around gcc sources, but I have to say that would 
> surprise me given that:
> 
> a) gcc (regardless of fact that it is a compiler) is meant to be
>     highly portable.
> 
> b) I would /suspect/ that compilers would tend to be quite
>     free-standing bodies of code, with the most minimal of external
>     dependencies of any code you could find.. (just a suspicion..)
> 
> > They seem to be using very recent kernel headers and binutils 
> > versions to get it going.
>  Binutils maybe. Kernel headers??? You thinking of glibc maybe?

>From a developer on the HLFS list:

GCC 4.1 changes.txt says Binutils 2.16.91 is needed now, for the
IA-32/x86-64 platforms. I didn't find the 2.16.91 tarball, but the .92
is dated from April, and .93 is dated from May (2005). 2.16.1 is dated
June (2005).. so it looks like 2.16.1 is an appropriate binutils version
for gcc-4.1. Does anyone happen to know a reason gcc-4.1.1, or
glibc-2.4, would need a newer (hjl) binutils version?

Some time later, he came back with the compile using glibc 'in the bag'
and this:

# - Kernel headers (I used linux-2.6.16.18):
# I dont know if this will be the way kernel headers are installed, but
# it works  for now. It's a good idea to be running this kernel version
# on the host system,  for Glibc.  Unpack your kernel source.

Now I'm not arguing with him, and I'm not arguing with you. Goodbye :-)
-- 
Declan Moriarty <junk_mail at iol.ie>