[Freeciv-Dev] Re: CVS doesn't compile

[Jason Short <jdorje@xxxxxxxxxxxxxxxxxxxxx>]

Per I Mathisen wrote:

> But there another issue: "make" doesn't know when to require another
> ./autogen.sh run. I think "make" / "make all" should read some kind of
> flag/counter to see if we need to rerun ./autogen.sh and abort, and have
> "make anyway" as a way to override this test. I am not sure how to
> implement this in a sane way, though.
I think this is conceptually wrong; autogen.sh is supposed to be a 
"developer" tool that isn't present in the final distribution.  From a 
practical standpoint, running "make dist" should give you a tarball that 
has the configure script and no autogen.sh, so checking for dependencies 
at this stage will not be useful.
The idea of a "make configure" doesn't seem workable either.  In this 
case the rules for "make configure" will be present in the distribution 
even though they wouldn't work, but so long as nobody ran it this 
wouldn't be a problem.  However, there would still be problems.  For 
one, "make configure" isn't any more automatic than ./autogen.sh, it's 
just faster.  Also (and more problematic), the behavior of "make 
configure" is created from the Makefile.am files after running automake 
&& autoconf && ./configure, so if changes are made (say, an added file) 
"make configure" wouldn't know about it anyway until autogen.sh is run.
The only thing that seems workable is to add dependency checking 
directly into autogen.sh.  However this is way more work than it's worth 
(dependencies would have to be checked on *everything* to be useful).
jason