[1466] in Kerberos-V5-bugs
Re: make install installs krb5.h but not autoconf.h
daemon@ATHENA.MIT.EDU (Theodore Ts'o)
Sat Jun 10 19:41:28 1995
Date: Sat, 10 Jun 1995 19:41:10 +0500
From: Theodore Ts'o <tytso@MIT.EDU>
To: epeisach@MIT.EDU
Cc: krb5-bugs@MIT.EDU
In-Reply-To: epeisach@MIT.EDU's message of Sat, 10 Jun 1995 08:59:58 -0400,
<9506101259.AA17006@kangaroo.mit.edu>
No, I don't want to install autoconf.h, and I don't want krb5.h to
depend on it. (I knew about it when I was fixing krb5.hin; I just never
got around to fixing it.) In general, we should avoid polluting the
namespace with symbols and macros that don't start with krb5/KRB5; and
if applications need autoconf defined symbols, they should test for them
themselves, instead of assuming that krb5.h will define some useful
subset of the autoconf defines. (Which might not be accurate anyway if
they are using a different compiler than the one used to compile the
library.)
[The other area where we are still polluting the namespace is the
encode_* and decode_* functions from the asn.1 library. It's one of
those things that we should fix; fortunately, they're internal
interfaces so changing them shouldn't break any application programs.]
As far as I know, the only dependency in krb5.h on autoconf.h, is
SIZEOF_INT, SIZEOF_SHORT, and possibly defining away "const". The const
issue has been discussed in krb5-bugs, with the thread starting at
transaction 1347, and I think that's the best way to handle it. For
SIZEOF_INT and SIZEOF_SHORT, those do tend to be platform dependent, and
they should be handled by a sed substitution on krb5.hin.
- Ted
