Re: autoconf and OS/2

[Tim Van Holder]

> That would have some really bad side effects:
> 
> if test -d "$dest_dir"; then
>   cp my_file "$dest_dir"
> fi
> 
> If x: is assumed to be a directory my_file would be copied somewhere.

That's not a bad side effect; that's expected behaviour.
The only bad thing is that Unix-minded people think that if $foo is a
dir
(other than '/'), then $foo/ is the same dir.  This does not hold for
a bare drivespec.

> mkdir -p works. However, I wonder why not using ginstall -d instead of
> mkinstalldirs?

Because ginstall is not present on all platforms; and on some platforms,
it's called 'install', not 'ginstall'.  AC_PROG_INSTALL is hairy enough
as it is.  Besides, then commands for automake's install rules would
have to depend on what kind of 'install' was found.

> > I don't understand this point.  "x:" is a relative path, just as
"x:y"
> > is.  Which directory "x:" stands for depends on the runtime context.
> > So "x:" shouldn't be treated as absolute, any more than "x:y" or "y"
> > should be treated as absolute.
> 
> May be. At least it works because all files are installed into
> $prefix/... which is a absolute path because of the first '/'.

This is a severe thinko.  'x:' is the same as 'x:.'; so if I use
'--prefix=x:', I do NOT want things installed in the root of drive X,
but in X's current directory.
So this should be seen as a relative path; at the VERY least, it
should be turned into 'x:.'; even better would be (cd x:.; pwd),
assuming
your bash isn't so brain-damaged that 'cd x:.' doesn't change the
current
drive.

> configure could change '\' into '/'. This isn't that difficult.
But not necessarily desirable either.