[MPlayer-dev-eng] autoconf

[D Richard Felker III]

On Mon, Mar 24, 2003 at 01:34:53PM -0500, Paul Jarc wrote:
> Arpi <arpi at thot.banki.hu> wrote:
> > Now i see no advantages of switching to autoconf,
> 
> Autoconf's generated code is used *much* more widely than your
> hand-written code, and so is more widely tested and less likely to be
> buggy.  It also provides additional features that users are accustomed
> to having, from other packages.

If our script is buggy we'll fix the bugs. Yes, every once in a while
someone commits dumb bash2-specific code to it, but the same mistake
can be made with configure.in! Just drink the cola, fix it, and move
along.

If there are missing features in our script, feel free to add them.

> > especially that no one of the active developers knows autoconf
> > syntax,
> 
> So this takes us back to "someone must volunteer to do the work".  I'm
> considering it.

NO! This is a reason it should not be done, since it's not a one-time
thing, but something the developers have to update every time they add
new optional features!!

> > and in my experience autoconf versions are not backward compatible
> > sometimes, so running it is a nightmare, and as mplayer is used by
> > users as cvs version, it may be a problem for them.
> 
> I've heard that the situation has improved in recent versions, but I
> don't really know.  Regardless, there are some projects that store the
> autoconf-generated files in CVS.  That way, everyone gets an
> up-to-date configure script without having to run the right version of
> autoconf themselves.

This still means any developer who wants to update the configure
script must have the right version of autoconf installed (bad), and it
pollutes the CVS logs immensely since autoconf insists on hard coding
line numbers all over the generated file, making it so even a small
modification alters the entire file.

Rich