[DejaGnu] DejaGNU and SH target

[Arati Dikey]

Rob,

I have built the complete toolchain(gcc, as, ld, objcopy ...) and I have
tested them
with many small programs which compiled and linked successfully. As my
target is SH. I 
used objcopy to make MOT (srec) files of the output and even flashed
them on my target.
For this, I have my C programs, a start.S file and linker script (.lnk)
file.

As the number of programs I can write is limited, I have gone for
DejaGNU testing framework.

Now, I feel I am getting these many errors because of the *'Could'nt
file tool init file'*
*warning* which comes while testing 'gas'. I am unable resolve this
warning. 
If it cannot initialize the tool,  What tests is the framework
conducting ?

so, please let me know:
<1> what tool init file is expected ?
<2> what are its contents ?
<3> do I have to create it or is it already there ?
<4> How to resolve this warning.
<5> Can the numerous bugs in gcc be attributed to this ?

Arati.


-----Original Message-----
From: Rob Savoye [mailto:address@hidden
Sent: Wednesday, August 15, 2001 10:13 AM
To: Arati Dikey
Cc: address@hidden
Subject: Re: [DejaGnu] Testsuite classification


On Tue, Aug 14, 2001 at 12:45:24PM +0530, Arati Dikey wrote:

> If you do not mind, I am writing the instructions I followed to build
> the sh-coff toolchain.
 
  Thanks for the notes, but I know how to build GCC as a cross compiler.
It's
just that I need to build a target with full libgloss and simulator
support.
Without libgloss, a fully linked executable can't be produced for the
sh-coff
architecture, unless you use some other crt0.S, and glue routines to
make newlib
work. I believe these are avalable from Hitachi. And both gcc 3.0 and
2.95.3 
have problems with building as a cross compiler for some reason  for all
the
targets with full tool chain support.
  
> After this, I installed dejagnu 1.4.0. As gcc2.95.3 has no testsuite,
I
> used the one in gcc3.0 and got the following

  There should be old snapshots of the gcc 2.95.3 testuite on
sources.redhat.com
someplace. At least there used to be. Anyway, my guess is you got so
many
failures because you don't have a full tool chain. You need startup code
to
produce binaries that can be executed on either the simulator or real
hardware.
A whole bunch of the testcases in the testsuite depend on the execution
of the
test case to determine if it worked. Can you compile, link, and execute
a simple "Hello World" program to make sure your tool chain is complete
?

        - rob -