On Mon, 27 Apr 2009, Tim Jenness wrote:
> On Apr 27, 2009, at 8:30 AM, Malcolm J. Currie wrote:
> >> No idea about that one. It must have been a transient of some kind,
> >> interrupted build or system glitch I'd guess.
> > I've seen a few glitches today. There's some obscure error in "make
> > world", but going to the directory itself and running "make; make
> > install" works fine. Odd. I started from a total clearout of the
> > installed tree and a fresh git clone to avoid issues from failed
> > builds or the presence of old executables.
> > There was an issue with ifd2star not being found. It was trying to
> > access my older private copy in my bin directory that was accessing a
> > now removed tcl, hence ifd2star failed. I'm puzzled why "./ configure
> > -C" would configure using that rather than a version built from the
> > repository code. Yes I had run "./bootstrap --buildsupport".
> > Reconfiguring the applications fixed the problem.
> The system has always been designed such that if it finds application X
> in the path at configure it will use that explicit version. If it
> doesn't find it it will assume that the file will turn up in the path by
> the time it was needed. Norman did it this way to prevent a large chunk
> of applications ending up as configure dependencies (you don't want make
> configure-deps to build 3/4 of the starlink software before configure is
> run). I suppose you could argue that it could always just assume it is
> going to turn up in the path.
Also if you run "./configure -C" and a version of ifd2star (or whatever)
is already recorded in "config.cache" it will be just reused, without
further checking, but since you changed to gfortran I expect you had to rm