[Ecls-list] Re: ECL ignoring command line parameters?

[Juan Jose Garcia-Ripoll]

On Wednesday 03 December 2003 19:10, Julian St. wrote:
> On Wed, 3 Dec 2003 15:08:26 +0100
> New kind of error:
> Configuring without parameters gives me:
> (gdb) r
> Starting program: /usr/local/bin/ecl
>
> Unrecoverable error:
> Lisp initialization error.
>
> Program received signal SIGABRT, Aborted.
> 0x281fd36f in kill () from /lib/libc.so.5
> (gdb) backtrace
> #0  0x281fd36f in kill () from /lib/libc.so.5
> #1  0x281f2068 in raise () from /lib/libc.so.5
> #2  0x28267e93 in abort () from /lib/libc.so.5
> #3  0x280a91c1 in error () at /usr/home/blitz/src/ecls/src/c/error.d:43

Ok, first about this error. I downloaded the release ECL-0.9c, applied the
patches (http://ecls.sf.net/patch-current-exp.gz) on a FreeBSD 4.7 box
(I should thank HP for their wonderful Testdrive program), and run
"./configure; make; make install".

A small problem which I have solved today was that "mkinstalldir" did not
work with the C-shell and thus the files did not get installed. Make sure
whether this happened in your box.

You should also check that you do not rely on CVS and have the right patches.
For a while I have been uploading patches under the name
	http://ecls.sf.net/patch-current-mp.gz
instead of
	http://ecls.sf.net/patch-current-exp.gz
because I was not sure whether the multithreading part would break something.
Today I have made the first file just a link to the second.

Other than this, the program compiled and installed just fine, and running
DISASSEMBLE, COMPILE, COMPILE-FILE is no problem.

> Configuring with --enable-local-gmp gives a more complete backtrace
> (why??):

This is because the new configuration process retrieves the optimal
compilation flags from the GMP library (which knows a lot about different
compilers, processors, ABIs, etc). One of the flags that GMP likes very
very much is -fomit-frame-pointer. It leads to smaller binaries, at the
cost of preventing debugging (by breaking backtrace).

I am not sure whether this search for optimal flags is good. I only know
that it is required for Solaris, because the GMP library typically decides
to compile itself using the 64-bit ABI when it is available, and ECL needs
to then use the same flags as GMP.

For the moment, if you want to ECL to rely on the bignum library to guess
the compilation flags, you have two options: use --enable-local-gmp, or
use --enable-simple-conf.

Juanjo