This is the mail archive of the crossgcc@cygnus.com mailing list for the crossgcc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: possible success with h8300 cross egcs


Kai Ruottu wrote:
>
>  I didn't find egcs-1.1 for libc-6 either... I know RedHat 5.1 is using
> egcs-1.0.2, don't know about S.u.S.E 5.3, haven't followed the Linux
> groups hunting clues about possible bugs in egcs-1.1 with Linux... So,
> I really don't have any clues if egcs-1.1 can be trusted under Linux.
> The fact that it can compile itself doesn't prove very much. There are
> quite a lot different code for the h8300 target to be compiled. And
> your problem isn't in the C compiler at all -- it is your h8300-target
> linker crashing...
>
I'm going to upgrade my linux this weekend to redhat 5.2,
I'll be sure to install the ready made binaries for egcs-1.02 and
libc-5xxx
and try this to build the h8300 tools

 
>  Your problem can be a good way to remove bugs from egcs-1.1... I have
> the egcs-1.1 for Linux too, which compiled itself with flying colors,
> configured to use libc-5 headers and binaries. So building binutils-2.9.1
> with it and seeing if my 'h8300-hms-ld' behaves just the same way is to
> be done...
> 

I'll be interested to see if your linker has the same trouble.  I'll
try compiling h8300 binutils 2.9.1 with egcs-1.02 this weekend and
compare
it to the one I build with egcs-1.1

> You can be proud if you found a bug in egcs-1.1 !
> 

whoopeee :)
Well, at least I don't feel like a victim. :)


> > Also, doesn't the compiler carry it's own library for it's own needs?
> > So it doesn't matter whether I have libc-5xxxx or not?
> 
>  It matters, you build your compiler to use either the old 'libc-5', using the
> target 'i586-linux-gnulibc1', or the new multithreaded 'libc-6' (glibc-2.0.x),
> using the target 'i586-linux'. If you build for the latter, but use it with
> the old libs & headers -- expect all your programs to crash...
> 
>  Funny, I read your message, but after writing all the stuff, I find that you
> could have made this error... This is perhaps one evidence about the
> "everybody knows this, so I don't write it here"-syndrome we see every day...
> 
Yes I can see where that would be a problem.  I built the native
egcs-1.1
letting config.guess figure out the host/target  would be
"i586-pc-linux-gnulibc1".  The libc on my system is: libc.so.5.3.12
I didn't upgrade to a newer libc after I built egcs-1.1, so presumably,
my native egcs-1.1 compiler will use libc.so.5.3.12.

> 
> > This wasn't an exercise in vanity or to satisfy some NIH pomposity,
> > the only binaries of h8 tools my net searches turned up were old gcc
> > 2.6.x dos and solaris hosted versions.  None for linux.
> 
>  I apologize my words... I meaned that you should be VERY CAREFUL with your
> 'production compiler', the native compiler, which you use to build things.
> It's nice to have something that you know you can trust on...  Of course I
> know there probably isn't a ready-made Linux-hosted h8300-hms target compiler
> available. The reason is that I haven't seen anybody asking this...
> 
Ok yeah, no offence taken, I think I misunderstood you, you're right,
it would be wise to use a pre-built/tested binary release of linux
native egcs
(1.02 or 1.1 if available) and use that to build my h8300 tools rather
than
build both the native and h8cross (thus two unknowns to look for bugs
in).
 
>  I understand that a student wants to learn to do things, but that those doing
> it at work, are allowed to do it, when there could be a possibility to get
> up-to-date binaries for almost every possible target, is really amazing.
>
When compiling h8300 cross egcs-1.1 directly (using gcc 2.7.2) didn't
work out,
I supposed the most expedient path available to me at the moment was to
first just compile egcs-1.1 myself as a native and use that to make the
cross.
That may not have been the best thing to do, the cause of my bugs could
either be caused by my egcs-1.1 native's compilation of the egcs-1.1
cross,
or in the egcs-1.1 cross itself.

>  You have probably (= should have for kernel build) the old 2.7.2.x for Linux,
> so using it to build the h8300-hms target binutils could tell something. I have
> used quite many GCCs to build binutils, but never got a crashing ld like yours,
> so my thought that there must be something badly wrong with the egcs-1.1 is the
> first thought which comes into my mind. It can be you configuring egcs for
> '--host=i586-linux --target=i586-linux' and using 'libc-5' with it or something
> else...

I think I was also doing my build wrongly when I first tried directly
using
gcc 2.7.2 to build egcs-1.1 h8300 cross. (I have learned MUCH over the
course
of this exercise!)  I'll go back and try that again too.
 
> Ok, I apologize again my talent to insult people when trying to help them... I
> have seen the same capability in many others, like Robert Lipe, who built the
> gcc for SCO OpenServer with big trouble, it is delivered freely in a Skunk-CDROM
> for every SCO licensee, and still people ask help when trying to build it 
> themselves --- just thinking it is 'a piece of cake'...  Robert saying that 
> "it isn't easy and you should use the available binaries", seems to be an big 
> insult to many...

My fault, I misunderstood your meaning, and your help is HUGELY useful
and
appreciated.  I also have to agree, if you don't have to build the
compiler
yourself, DON'T.  Every time I get a build wrong, and there's plenty of
room
for that, I'm looking at HOURS of re-compilation in order to try
something
else.  And that's on a machine which doesn't exactly suck....Pentium 166
w/64M EDO, 512K pipeline cache, and fast 3.2G hard drives.

> 
>  Regards, Kai
cheers, randall
-- 
|\/|
|/\| randall@elgar.com
|\/| rsl@zanshinsys.com http://www.zanshinsys.com
________________________________________________
To get help for the crossgcc list, send mail to
crossgcc-request@cygnus.com with the text 'help'
(without the quotes) in the body of the message.