🚀 go-pugleaf

In article <1992May13.174308.21906@u.washington.edu> donn@carson.u.washington.edu (Donn Cave) writes:

   Do you happen to know how the patch/perl approach manages to outdo
   fileutils?  You're right, this is a source of problems, but I seem
   to recall having similar problems configuring perl.  Does its
   Configure benefit from a conceptually different approach, or does
   it just happen to catch more outlying cases because it has been
   worked over more thoroughly?

Both metaconfig and autoconf are feature testing systems.  That is,
their first heuristic is to "look" at the environment around them in
an attempt to derive and appropriate feature set.  Both systems also
have database heuristics.  That is, in some cases, something else must
be done.  In general, metaconfig is a little further evolved than
autoconf.  Hence, yes, I think metaconfig just catches more outlying
cases.

   | As someone who does a lot of GNU compilation, and who is getting into
   | cross compilation gradually, I don't mind editing Makefiles, but someone
   | who wants automatic configuration to take place after one or two commands
   | would find this method unsatisfactory.

   That's true enough.  I suppose that means that a configuration script for
   groff would ultimately need to ferret out the answers to questions like

     # Uncomment the next line if you have the mmap() system call (and you
     # want to use it).
     #D4=-DHAVE_MMAP

   which sounds like a job for Larry Wall's Configure.  Why do I find
   editing the Makefile so much more straightforward?

Because you only do one at a time.

Imagine that I've just hired you to bring up 80 some GNU programs on
my widgetbox, my foobox, my PowerWhatSis, my X86 running BarNix, and
my own internal hardware running my own internal *nix (which is based
mostly on SysVr1 with patches to bring it up to Vr4, but still using
v7 a.out, with bsd networking thrown in, and a posix library, but no
ansi library.  Oh, and strchr() and rindex() are currently broken,
sorry. :-)

Now, you know the gnu programs, but you have no clue what the feature
sets look like on my boxes.  If everything is Makefile based, you're
looking at 80 x 5 Makefile edits, plus research time.  In short, a
coupla busy weeks.

With a tolerable configure system, and a little luck, you're looking
at five (well, ten what with a few misteaks) invocations of configure,
20-40 edits for minor corrections, and a nice pleasant weekend with a
book, or Significant Other, with occasional dial in's to check on
progress.  Perhaps another day finding new bugs in my internal *nix.

Now if you take that time savings and multiply it by tens or hundreds
of thousands of GNU admins, many of whom *don't* know the feature sets
of their machines as well as you do, well, that's a lotta time saved.