Portable openssh build problems.

[mouring at etoh.eviladmin.org]

[Moved to openssh port from openssh at openbsd.org for broader comments]

On Wed, 18 Apr 2001, Mike Crowl wrote:

>
> To whom it may concern,
>
> I've been building openssh for various versions of solaris and noticed
> that I'm having to jump through hoops to get it done because the build
> process for openssh is pretty sloppy.  Specifically there are several
> option flags in the configure script which do not work at all.
>
> --includedir
> --libdir
>
> (do not place the correct locations of zlib into the make when I
> specify the include directory and lib directory under my installed zlib)
>

Hmm..Can't comment on either one of theses because I never use them.  Even
under Solaris, I tend to install common libraries in /usr/local.

> --with-pid-dir
>
> (does not do anything...but is mentioned in the install page on
> website.  I tried this, but it ignores my input entirely.  Also it does
> not place the pid file in a uniform location across different OS
> versions of solaris.  I.E. if /var/run exists, then pid goes there...if
> not then it goes into the dir with the local configs.  It should go to a
> uniform place if you are not going to have this option work correctly.)
>
Thank you very much for pointing this out.  I've corrected it.  Somewhere
in the Makefile.in clean up it was lost.  If you add to your Makefile.in:

PATHS=
	[..] \
	-D_PATH_SSH_PIDDIR=\"$(piddir)\"

Then re-run your ./configure it should compile with the correct path
for the PID directory.

> make install - always creates a host key....assumes that you always
> want to do so.  This is not a good idea when you have just
> cross-compiled for multiple architectures and need to install the
> binaries into you apps tree for later dist to various OS/server types.
>

make install tests to see if you have keys.  If you don't it will
automaticly generate them.  This is the correct behavior.  There was
a long talk about this long time ago, and the agreement was that this was
the correct behavior.

I'm not sure if it's worth adding a 'make install-nokeys' for those who
don't wish to generate keys at install time.

> Also the location of tcpd.h and libwrap.a cannot be fed in via the
> configure script.  The build blindly assumes they are installed in
> /usr/include and /usr/lib.
>

/usr/include, /usr/local/include /usr/lib, /usr/local/lib  are currently
the default search path for the installer (for Solaris).

--with-cppflags="/path/to/wrapperinclude/" \
--with-ldflags="/path/to/wrapperlib/"

should work fine to add a new Inlcude and Library paths.


> I know I'm not offering anything in the way of a fix here, but maybe
> noting this stuff as a bug will allow someone to have justification to
> work on it.
>
> I'm a little surprised that I'm having this much trouble.  I run
> openssh on my three linux servers at home and had no problems getting it
> built and running.
>



- Ben