fixincl.x being blown away

[Richard Earnshaw]

> so you would have an empty file only if fixincl.x is gone.

That was, of course, the problem.  And it manifested itself as follows.

Since I don't do a fresh configure each time I build gcc (configure tends 
to be very slow on StrongARMs because of the number of forks that are 
needed -- I also want to keep some of the log information from previous 
builds), I tend to run

  gnumake clean-target && gnumake clean && gnumake bootstrap && gnumake 
check

This was deleting the fixincl.x file.  Autogen was not installed, so the 
touch command created an empty one, and it then *appeared* to have stamped 
on the old one.  The build only failed when it came to compile fixinc.c, 
since the header file was empty.

Maybe genfixs should exit with an error if autogen is not installed and 
the output file does not exist.  Then at least the build error would occur 
at the point of failure, rather than sometime later.

R.