httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ralf S. Engelschall" <>
Subject Re: Allow modules to add libraries
Date Sun, 06 Jun 1999 19:16:11 GMT

In article <> you wrote:
> Ralf S. Engelschall wrote:

>> Hmmm... I've no great objection to Ben's patch and idea in general, except
>> that it looks inconsistent. When we allow to adjust the libraries via "Libs:",
>> we also should allow "CFlags:", "LdFlags:", etc. But this OTOH is not really
>> necessary, because inside ConfigStart...ConfigEnd you're more flexible (you
>> can expand variables, etc.). So I still don't see a great advantage from
>> having "Libs:" in contrast to ``LIBS="$LIBS ..."''. Instead I've to conclude
>> that just providing "Libs:" is only a half-way solution and will confuse
>> people more than it provides advantages. Ben, where exactly is the benefit
>> from using "Libs:" above a config section instead of changing LIBS in the
>> config section itself? I guess I've overlooked an important intention here,
>> right?
> The important point is that changing LIBS= and other stuff requires
> knowledge of the guts of the configuration process, whereas declaring it
> with Libs: does not.
> There are two obvious advantages to the declarative approach:
> 1. We restrict what is possible, which allows us greater room for
> manoeuvre later.
> 2. We don't force ourselves to maintain a particular implementation for
> Configure by forcing people to rely on the implementation.

Oh, I've actually overlooked these two points. Yes, I've to admit that they
sound reasonable. Thanks for explaining them, Ben.

> I absolutely agree about the inconsistency, it is my fond hope that
> having started the ball rolling with Libs: others will follow suit and
> add the rest. For example, Greg as already expressed a desire for adding
> rules. However, as a general point, it doesn't seem to me that one
> should require the submitter of an improvement to also follow through
> and provide every other related improvement.
> Ideally, the need for ConfigStart/End should be eliminated entirely,
> because, as far as I can see, it is, at best, a kludge, and at worst a
> maintenance problem.

Hmmm.. I personally wouldn't go so far and say it's not needed, of course.
You're right: it's a kludge. But for complex modules like mod_perl, mod_ssl,
etc. it's a very important facility.  Only with it they can adjust the built
environment correctly.  They've to do some checks for which one need some
scripting language. And here the ConfigStart...ConfigEnd facility with the
embedded shell scripts is a very useful thing.

But ok, I know understand your intentions, so +1 for the idea of declarations.

                                       Ralf S. Engelschall

View raw message