httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From r..@ai.mit.edu (Robert S. Thau)
Subject Re: conditional HTML
Date Mon, 15 Apr 1996 19:08:40 GMT
One man's opinion:

* Should the default configuration files ("blah-conf.dist") be
  rewritten at build time to know the install location?  This is easy
  to do, if it is desirable.

IMHO, the right thing to do is to produce a set of config files which
are valid for the location where the user wants to install the thing,
but to keep "make install" from installing them if another set of
config files is already there (with a separate "install-config-always"
makefile target for anyone who wants to force the issue).

BTW, having looked your package over, there's are two questions I can
answer and another I'd like to ask.  First the answers:

ULTRIX_BRAIN_DEATH is used to simplify certain declarations which the
Ultrix vendor-supplied 'cc' chokes on.  I gather you have not had
the... ummm... *pleasure* of dealing with it.  Given that Ultrix is on
the outs anyway, it may make sense to just require use of gcc for
Ultrix installs (there are a heck of a lot more people running SunOS
at this point who live with the same restriction).

FCNTL_SERIALIZED_ACCEPT is needed on certain sysv-derived TCP stacks;
when two processes simultaneously do an accept() on the same socket,
on these systems, it can cause a kernel panic.  Since some of the
affected systems are widely used (e.g. Solaris --- have they fixed it
for 2.5?), you really do have to deal with it.  There probably ought
to be a comment in the base Apache code explaining this someplace.
(I'm not at all sure how to do an autoconf test for this behavior).

The thing I'm not sure of is your strategy for dealing with an issue
which is handled by the current config setup.  For SunOS and SGI, the
'Configuration' file tells the user how to select NIS or DNS for name
lookups.  This is difficult to determine with an autoconf test, since
there are sites where both of them work, but there are valid reasons
for preferring one or the other which depend on the site's
administrative policies.  It wasn't clear to me from looking over your
scripts how you're handling this.

(FWIW, I'm told the Kerberos folks wound up using configure
command-line options to cover this case, and maybe a few other things
that sleaze in around the edges).

rst

Mime
View raw message