UNIX SIG by Chris Fearnley

A Miscellany

I don't have a good topic for this article, so I will sketch out my
thoughts about various things this month. First, for those of you looking
to purchase a Linux CD, my current recommendation nods toward Red Hat.
I still think that Debian has several fundamental advantages over Red Hat.
But Debian is still "a work in progress" with a few "rough edges".
In a few more months, Debian will have migrated to ELF and several of
the basic installation problems will be resolved (I hope!). Red Hat
is talking with several Debian developers about integrating the two
distributions (at least to the extent that .rpm and .deb packages can
be installed on each system in a maintainable fashion).

Linus has just released 1.3.49. There are still many bugs in this
developmental series of kernels (1.3.xx), but several new features
are available. SMP (Symmetric Multiprocessor Support) by Alan Cox
is minimally functional; most of the kernel drivers can be selected
as loadable modules (in make config); many new drivers are available
including appletalk, Token Ring, Arcnet, SMB support for WFW volumes,
AX.25 for amateur radio, EQL (serial line load balancing), and Stallion
multiport serial card support among others; there is direct support (in
a single source tree) for the alpha, i386, mips, and sparc processors;
and finally, the kernel is now over 400,000 lines of code (for better
and for worse). Still no one knows where the train is headed, but we
are all enjoying the ride!

In my efforts to maintain mawk and gawk (two GPL'd versions of the
AWK programming language), I have uncovered some curious differences.
Which raises the issue of whether to make /usr/bin/awk a symbolic
link to mawk or to gawk (in either case some code will work for one
implementation and not the other). Now, the 1.3.xx kernel releases use
awk to calculate dependencies in C code for the Makefile. But when mawk
is used instead of gawk, gobs and gobs of error messages are reported
(no such file or directory). It turns out that this is due to a
peculiar mis-feature in mawk's implementation of the getline() function.
This small script will spit out an error message when the file test
doesn't exist: { ERRNO = getline dummy < "test" }. It also sets ERRNO
to -1, so I don't see any reason for it to spit out anything onto stderr.
I've contacted the mawk author and I'm hopeful that there is some way to
turn off this error message. The source code for mawk and gawk contain
tests to verify that the program performs adequately after a new build.
In the gawk test suite there is this odd script: { gsub(/^[ ^I]*/,
"", $0) ; print } where ^I indicates a [TAB] character.  Running gawk
with this script on the input: "  ^IThis is a test, this is only a
test." produces "This is a test, this is only a test." whereas mawk
produces "Thisisatest,thisisonlyatest.". Of course it is a bit ridiculous
to use gsub on a regular expression with a ^ (beginning of line anchor) in
it or with an * (zero or more "leftmost longest" matches). As such I'm
not sure which program behaves more responsibly here. But the difference
is a bit unsettling.

I have also been compiling apache 1.0.0 (if I ever get it into a "good"
state, I plan on releasing this work to the Debian Project as well).
Apache has an option to use the GNU dld (dynamic link editing library)
to load it's modules at run-time. Since I'm building it for Debian's
pre-release ELF system, I might also build a shared library of all it's
modules (building shared libraries under ELF is trivial: build all of
the object files with -fPIC then when building the shared object use gcc
-shared -Wl,-soname,$(LIBSO_MAJOR) -o $(LIBSO) $(OBJS) and it should
work). The question arises which approach is best: dld, ELF shared
libraries, or a monolithic apache binary (since Linux supports shared
copy-on-write executables this isn't at all a bad option either). Now,
the givens are that apache is the only program that can take advantage
of any run-time modularity provided by dld and shared libraries and that
there may be dozens of apache-httpd processes in core at any given time.
I'm not a compiler expert, so I'm asking: What approach do you think
is best?

There have been two meetings of the Philadelphia Linux User's Group
(PLUG). PLUG was formulated at a PACS meeting, but has no formal ties
to PACS. They have been quite fun - hanging and talkin' shop with Linux
users. The third meeting was cancelled due to the big storm. The next
meeting is January 9. Look for announcements in phl.announce, or send
me e-mail <cjf@netaxs.com> requesting to be on the PLUG mailing list.
Many people have suggested we move to a bi-weekly format (we've been
trying to meet weekly). What do you think?

Next month at PACS we will have another Q & A session on Unix and
Linux.