point taken; if this becomes an issue we should add a "are these two
descriptors open to the same file" syscall (taking two file
descriptors).

However, there's also the issue that "test -ef" doesn't actually test
that the immediate link target is pointing to the *string* which it
should point at...

If you're juggling object trees around like you seem to be you may
also be playing with the value of BSDOBJDIR .. and if the underlying
tree has multiple names due to symlinks, over time you may end up
creating links to different names for the obj tree, and you may run
into some inconsistancies if you "make obj", then move one of the
symlinks... and if this happened, it would not be immediately obvious...

I'm happier with the "compare link target text" approach.  Too bad
that there isn't a standard readlink(1) program (and, before anyone
goes off and implements one.. well, depending on it in the build
process would complicate cross compiles that much more...).

					- Bill