[903] in Kerberos_V5_Development
Re: "Two copies of DES library" change breaks K5 compat w/Cygnus K4
daemon@ATHENA.MIT.EDU (eichin@MIT.EDU)
Mon Mar 20 22:28:02 1995
Date: Mon, 20 Mar 95 22:26:19 -0500
To: tytso@MIT.EDU
Cc: gnu@cygnus.com, krbdev@MIT.EDU
In-Reply-To: <9503210148.AA18498@dcl.MIT.EDU> (message from Theodore Ts'o on Mon, 20 Mar 1995 20:48:46 +0500)
From: eichin@MIT.EDU
Last time I ran into this, I just tweaked the makefile to make sure
that KRB4_whatever was both -lkrb and -ldes. Note that the *only* time
that the problem shows up is if you've got the CNS V4 libkrb -- which
pretty much guarantees you've got the corresponding CNS V4 libdes, so
linking it in is no big deal (and a one line change to either pre.in
or the relevant makefiles, I'm not sure which.) I'm pretty sure that
having it be -lkrb -ldes425 -lcrypto -ldes works just fine; it may
look "wrong" but it'll get you (John) around your immediate problem...
It's not general -- for example, Solaris 2.x ships with a v4 libkrb.a
*only*... but the autoconf tests we use for cvs (which test for libkrb
and then seperately for libdes) cover that, and could be used here.)
I think the argument about building libdes.a is specious -- because if
you've got the v4 libkrb.a built, then the libdes.a was the easy
part. (Also, to address jik's comment, the V5 libdes *is* the CNS V4
libdes, ever since we switched to configure.)
There's also a bug in the idea that the V4 krb and des interfaces are
distinct... sure, they can be divided up, but the V4 libkrb *use* of
the des interface is not parameterized enough to withstand any real
change (such as doubling the keyschedule size) so this will all be
half broken until the v4 lib is folded into the v5 tree.
_Mark_
