Re: flexfax: Signal 11 on Solaris 2.7

[Nico Kadel-Garcia <raoul@cirl.meei.harvard.edu>]

-----BEGIN PGP SIGNED MESSAGE-----

On Mon, 25 Oct 1999, Rudolf Wegmann wrote:

> I'am new to this list and to hylafax.
> I'v been trying to compile and install hylafax on our server during the last
> days.
> Unfortunately, i encountered a few problems compiling this stuff. Finally, i
> get it compiled, even with help of the mailing list archive of hylafax, but
> the first test to connect to the hylafax daemon by using faxstat or telnet
> results in a Signal 11 crash of the daemon.

Gack. Let's see what's going on.

> Following my configuration:
> SUN E450, Solaris 2.7, NIS+, gcc2.95-1, tiff-v3.4, ghostscript5.5.,
> HylaFax-v.0pl2

OK, here is part of your issue: v4.0pl2 has a stack of patches
needed for later gcc versions. We're *TRYING* to get the beta
out the door, but you probably want to grab the latest CVS code.

Also, a note for the unwarey: NIS+ has a very poor reputation....

> The main problems building hylafax were as follows:
> - Because of low space on /usr i had to choose /opt as install destination.

This seems semi-standard for Sun OS's. I used /usr/local myself for SunOS
for similar reasons.

> - Configure failed because of test for libtiff was not successful.
> LD_LIBRARY_PATH was not set to correct destination. I know this is mentioned
> in the doc's therefore my fault.

Ohh, this is a fun part. Had it set to "/usr/openwin/lib", didn't you?

> - First compile run failed because of wrong makedepend path to assert.h. I
> was able to fix this also with help of the mailing list.

The patch for this is in the latest CVS sources at www.hylafax.org:
basically, there is a problem in the "sed" scripting of "mkdepend"
that should turn something like this:

	/usr/local/lib/gcc-lib/sparc-sun-sunos4.1.4/2.95.1/include/../../../../../sparc-sun-sunos4.1.4/include/assert.h

into something like this:

	/usr/local/sparc-sun-sunos4.1.4/include/assert.h

If anyone can figure out the problem in the "sed" script, it would be good
to correct that. It would help with some GNU "configure" scripts as well...

> - Next compile run failed because gcc2.95 is more restrictive. According to
> hints on the mailing list, i should apply patch gcc-2.8.x.patch. This failed
> also with message:

Oh, yeah. Especially with g++, it's quite anal.

> So, i found another hint on the list using compiler flag
> LC++OPT -fpermissive. Afterwards the next compile run was successful but
> with a lot of warning messages. Hmm.. I thougt this can be a problem at
> runtime.

Potentially: grab the CVS source and try that: Solaris 2.7 is the bleeding
edge of freeware development right now. I've also just updated
my patches at http://cirl.meei.harvard.edu/hylafax/patches/ for
the latest CVS source (as of 5 minutes ago).

The "-fpermissive" was due to g++ being quite picky about implicit
declarations, etc. The handling of this *changed* between version 2.8
of g++ and the 2.95.1 release. There are several declarations of
"bzero" and "munmap" in the patches I mentioned to help correct this,
and they may already be integrated into the CVS source (or not! Robert
has a day job too!)

Getting HylaFAX onto new operating systems relies on people like you,
Rudolf, who have gotten the new OS and are thrashing out the problems.
Thank you for the contributions: it helps the improvements get folded
back into the code quicker when we find out they're not just helpful
but *necessary*. (That was a royal "we"....)

			Nico Kadel-Garcia
			Senior Engineer, CIRL 
			Mass. Eye and Ear Infirmary
			raoul@cirl.meei.harvard.edu

-----BEGIN PGP SIGNATURE-----
Version: 2.6.2

iQCVAwUBOBRbJT/+ItycgIJRAQGnHAP+JD8yP4WhVeiZp43lIDRxyB11sY5l5Uhz
Ef2pzR8CQFvauFz/m4oxulbXinIelMBEMoH2Oe/GMlhoWYzFD6lDn5Dsu5A1t00J
SqIaEOj/C/wUAjrXagkBAl6ghMWmPmqDu1yMtlDTlhEKUzxWKsDrwb8vu82toMog
QaT4+JqlbBQ=
=Lsbz
-----END PGP SIGNATURE-----