Oops, I replied to Eric instead of Reply-All.  Here is what I said:

----- Original Message -----
From: <[EMAIL PROTECTED]>
To: Eric W. Biederman <[EMAIL PROTECTED]>
Sent: Thursday, December 06, 2001 1:32 PM
Subject: Re: NLBConfig and other changes


> From: Eric W. Biederman <[EMAIL PROTECTED]>
> > [EMAIL PROTECTED] writes:
> > > - The "include cpuflags" in the Makefile was broken: namely, the
> > > cpuflags file was generated long after it was needed.  So I eliminated
> > > the cpuflags file and used a different mechanism to compute CPUFLAGS.
> > > Yet another way to do this would be to just have NLBConfig write
> > > "CPUFLAGS := ..." in Makefile.settings.
> >
> > Was it really broken.  It works here and I haven't heard complaints.
> > GNU make is pretty smart about building makefile depenencies.
>
> I may check it again sometime, but what I remember is that the _first_
time
> make was run after doing NLBConfig (no existing cpuflags file), make would
> warn about the missing cpuflags file, then would make it later on.  So
next
> time make was run, it would (I _assumed_) use the previously made cpuflags
> file.  Sort of one step behind.
>
> I must admit my solution is a bit obscure.  I had to study the make
> documentation and fiddle with it for a few hours to find something that
> works.  Make wasn't designed to be a scripting language.  Or be readable.
> :-)
>
> > Ideally I'd like to have 2 tools.
> > Build-Makefile
> > Build-Configuration
> >
> > And with everything being derived from Makefile.settings and regenerated
> > we don't have a problem seperating things out.
>
> Are you saying that Makefile.settings should be the only "source" file for
> configuration, and the top-level config file that is now passed to
NLBConfig
> would be derived from Makefile.settings?
>
> I'm afraid I don't see a clear advantage to that scheme, compared to what
is
> done now.  Could you explain?
>
> The current scheme make sense to me:  There is a top-level config file
that
> can be easily altered by the user.  Makefile is derived from the config
> files, and it controls the overall flow or "shape" of the build.
> Makefile.settings should contain parameters that don't affect the
Makefile,
> but do affect the .o and later files.  Do you want to have a way to
specify
> parameters that _do_ affect the makefile?  If so, NLBConfig could be
altered
> to allow such parameters to be set in the config files (I think
[N]LBConfig
> might have previously allowed this, based on a few lines of "unreachable
> code" that I removed.)
>
> If you can explain what you want, and can convince me that it is an
> improvement, I might try to implement it.
>
> > And you missed the number one detail on my todo list.  Add header file
> > dependencies so we have complete dependency information.
>
> Do you have an idea about how you would want to do that?  There are a few
> tools (mkmf, etc.) that can figure out what the dependencies are by
reading
> the source files, but I'm not all that familiar with them and not sure
what
> would best suit your needs.
>
> Thanks for your feedback!
> - Jan
>
>

Reply via email to