libf2c problem

libf2c problem

[H.J. Lu]

> 
> In message <m0y1kRV-0004ecC@ocean.lucon.org>, H.J. Lu writes:
> 
> >BTW, I can provide you a patch to egcs 1.0.1 to fix the g77 problem.
> >It will be hard to make one for gcc 2.7.2.3. egcs 1.0.1 is also
> >better than gcc 2.8.0 in the current form.
> 
> gcc 2.7.2.3/g77 0.5.21 would be preferred, but egcs 980205 would be okay too.

egcs 980205 has that fix.

> The latest binutils makes it fail all of the g77 tests:
> 
>                 === g77 Summary ===
> 
>  # of expected passes            40
>  # of unexpected failures        46
>  # of untested testcases         46
> /usr/src/egcs-980205/gcc/g77 version egcs-2.91.07 980205 (gcc-2.8.0 release)
> 

The latest binutils is fine. It is dejagnu that is broken.

Someone forgot to update dejagnu when libf2c was moved out of the
gcc directory. Is there a patch for dejagnu to find where libf2c.a
lives?

Thanks.


H.J.

Re: libf2c problem

[Joe Buck]

> The latest binutils is fine. It is dejagnu that is broken.
> 
> Someone forgot to update dejagnu when libf2c was moved out of the
> gcc directory. Is there a patch for dejagnu to find where libf2c.a
> lives?

It seems that dejagnu is too tightly coupled to the Cygnus tree structure
and has too much builtin knowledge of How Things Are Supposed To Be,
so it is very brittle.  Is there any way that this can be addressed?
It seems that paths to libraries should be part of the egcs distribution
and not dejagnu.

Re: libf2c problem

[John Gotts]

In message < 199802100014.QAA19278@atrus.synopsys.com >, Joe Buck writes:

>> The latest binutils is fine. It is dejagnu that is broken.

>> Someone forgot to update dejagnu when libf2c was moved out of the
>> gcc directory. Is there a patch for dejagnu to find where libf2c.a
>> lives?

>It seems that dejagnu is too tightly coupled to the Cygnus tree structure
>and has too much builtin knowledge of How Things Are Supposed To Be,
>so it is very brittle.  Is there any way that this can be addressed?
>It seems that paths to libraries should be part of the egcs distribution
>and not dejagnu.

Admittedly, I was using a recent snapshot of dejagnu (971222 to be specific),
but I forsee dejagnu as becoming a generally useful tool like autoconf and
libtool.  Octave already successfully implements it.  Perhaps after a bit of
work on dejagnu it will be ready for general use and a berth on ftp.gnu.org.

--
John GOTTS <jgotts@engin.umich.edu> http://www-personal.engin.umich.edu/~jgotts

Re: libf2c problem

[Jeffrey A Law]

  In message < 199802100014.QAA19278@atrus.synopsys.com >you write:
  > It seems that dejagnu is too tightly coupled to the Cygnus tree structure
  > and has too much builtin knowledge of How Things Are Supposed To Be,
  > so it is very brittle.  Is there any way that this can be addressed?
  > It seems that paths to libraries should be part of the egcs distribution
  > and not dejagnu.
Hard to argue with that statement...  The dejagnu situation is far
from ideal.

I don't have any good technical ideas on how to implement this
though.

jeff

Re: libf2c problem

[Jeffrey A Law]

  In message < 199802100219.VAA04703@pa.engin.umich.edu >you write:
  > Admittedly, I was using a recent snapshot of dejagnu (971222 to be specific),
  > but I forsee dejagnu as becoming a generally useful tool like autoconf and
  > libtool.  Octave already successfully implements it.  Perhaps after a bit of
  > work on dejagnu it will be ready for general use and a berth on ftp.gnu.org
Actually, dejagnu has been on prep.ai.mit.edu:/pub/gnu for years.

Unfortunately, the version on prep is quite out of date and not suitable
for use with egcs.

dejagnu is generally useful (I've used it to build testsuites for gcc, gas,
gdb, ld, libstdc++, libio, libg++, a research linker project, and a few
other things I can't remember).

The problem is it needs a fair amount of information to do things like
build and execute programs.  It has to know how to find your compiler,
header files, libraries, linker scripts, startup code, object file converter
(if you are talking to a target board for example), load the program, etc etc.

Putting that information into a generally maintainable form is a nontrivial
project.

On might consider trying to encapsulate some of the compiler, header files,
and library information into some structures similar to the target_board
stuff.  I don't know how well it will work in practice though...

jeff

Re: libf2c problem

[Mark Mitchell]

>>>>> "Jeffrey" == Jeffrey A Law <law@cygnus.com> writes:

    Jeffrey> dejagnu is generally useful (I've used it to build
    Jeffrey> testsuites for gcc, gas, gdb, ld, libstdc++, libio,
    Jeffrey> libg++, a research linker project, and a few other things
    Jeffrey> I can't remember).

In fact, a company I worked for used deajgnu as a general-purpose
test-harness to test many things that were not at all related to
compilers, or development tools.

-- 
Mark Mitchell		mmitchell@usa.net
Stanford University	http://www.stanford.edu

Re: libf2c problem

[Mike Stump]

> Date: Mon, 09 Feb 1998 21:19:25 -0500
> From: John Gotts <jgotts@engin.umich.edu>

> In message < 199802100014.QAA19278@atrus.synopsys.com >, Joe Buck writes:

> Admittedly, I was using a recent snapshot of dejagnu (971222 to be
> specific), but I forsee dejagnu as becoming a generally useful tool
> like autoconf and libtool.  Octave already successfully implements
> it.  Perhaps after a bit of work on dejagnu it will be ready for
> general use and a berth on ftp.gnu.org.

You mean like this:

kankakee bash[510] ftp ftp.gnu.org
Connected to ftp.gnu.org.
220 aeneas FTP server (Version wu-2.4(3) Mon Jun 5 16:40:22 EDT 1995) ready.
ftp> ls deja*
200 PORT command successful.
150 Opening ASCII mode data connection for file list.
dejagnu-1.3.tar.gz
226 Transfer complete.

?  1.3 is fairly old now...