Gssapi Questions
Allen McWongahey
allenmcw at comcast.net
Wed Aug 24 15:03:24 EDT 2011
Hi Marcus,
Thanks very much. This definitely got me further and I removed all the
changes I made to configure so it completed as it was designed to do.
Now I'm having a different problem though. The order I should do things are:
"configure", "make", "make install" - is this correct?
When I run "make" I am getting the following error:
cc1: unrecognized option `-Wdeclaration-after-statement'
cc1: unrecognized option `-Wvariadic-macros'
The configure script apparently thinks my compiler supports these options
but it does not. I am using gcc version 3.2. From comments in the configure
script I thought I could specify "error=foo" to eliminate this CLFAG in the
Makefiles This does not work. I also tried "Werror=foo" but it does not work
either.
If I am understanding your previous message correctly "make install" should
add all the appropriate header files, correct? But since "make" precedes
that will make even succeed since those same header files are still not
there?
Thanks much
Allen
-----Original Message-----
From: Marcus Watts [mailto:mdw at umich.edu]
Sent: Wednesday, August 24, 2011 9:03 AM
To: allenmcw at comcast.net
Cc: kerberos at mit.edu
Subject: Re: Gssapi Questions
"Allen McWongahey" <allenmcw at comcast.net> writes:
...
> that it would not bomb out because I'm doing cross compilation.
> Specifically when it checks for "constructor/destructor", "regcomp"
> and "printf positional". Basically I just modified the script to
> disable the checking of these 3 things and the script then ran fine.
...
For autoconf / configure, rather than commenting tests out, there is a
better way. autoconf looks at internal variables to see if it's already run
a test, and suppresses it with a special "(cached)" log message when it does
so. You can set those variables in advance which means:
you can tell it what answer to use.
you don't need to touch configure at all.
for instance, gnu parted 3.0 can bomb when it tests for the existance of
linux/ext2_fs.h. To fix that,
./configure ac_cv_header_linux_ext2_fs_h=yes
In your case, for MIT kerberos, something like,
ac_cv_func_regcomp=yes
krb5_cv_attr_constructor_destructor=yes,no
ac_cv_printf_positional=no
(you'll need to to adjust yes|no to match your configuration.)
If you need a list of the cache variables, configure a scratch copy of
kerberos (say, for your native host), then look at config.cache.
To find out what a particular test wants, look at configure.
If you're really not sure what the answer is, you can rip the test out of
configure, cross-compile it, and run it by hand. Note that \$ in conftest.c
in configure means $ in the actual test source (this matters for your
ac_cv_printf_positional test!)
...
> 2) When running configure for OpenSSH it tries to include a header
file
> "gssapi_krb5.h" or "gssapi/gssapi_krb5.h" depending on which of two
> GSSAPI compile flags are set. Neither of these exist in the Kerberos
distribution.
> The nearest match I can find is "gssapiP_krb5.h" in "src/lib/gssapi/krb5".
> Am I missing something here or is OpenSSH looking in the wrong place?
You're looking in the wrong place, apparently your source or build area.
When kerberos is *installed*, it will make
include/gssapi
include/gssrpc
include/kadm5
include/krb5
wherever you told it (configure --prefix etc, as modified by make install
DESTDIR=).
Looks to me like you'd tell openssh
./configure --with-kerberos5=/usr
if you have /usr/include/gssapi/gssapi_krb5.h It wants to see
$KRB5ROOT/bin/krb5-config so that's a good starting point.
If you're telling mit kerberos to install in some spectacularly non-standard
way, the logic in openssh's configure probably won't work too well.
If you're cross-compiling, you probably *should* be using DESTDIR to put
kerberos out of the way of your running system. In that case, krb5-config
will *definitely* produce the wrong answer. For your cross-compile
environment, you may want to do something about that.
-Marcus Watts
More information about the Kerberos
mailing list