ac_check_krb5 - comments requested

[Ken Hornstein]

>I've checked in a preliminary example autoconf macro for using
>krb5-config (both MIT and Heimdal) for setting build system paramters
>for building applications.  I have a few points on which I would like
>comments.
>
>(1) Should it set LIBS rather than LDFLAGS (I am leaning towards yes)

The autoconf macros I wrote have always set LIBS (I'm assuming you mean
for the output of krb5-config --libs).  Seems to make sense, and it works
reasonably well.

>(2) What should be done about runpaths?  (In MIT krb5, they are
>    currently emitted if the build system linked the krb5 binaries
>    with runpaths.)  Note that actually altering current behavior
>    might break some things, and requires altering krb5-config's
>    behavior in any case.

I never ran into an issue with putting the runpath info simply into LIBS,
but I was always compiling using the same compiler on every system to
compile Kerberos and the various applications.

>(3) Related to (2), what should be done about differences between
>    compilers on the same platform?

I guess I see a couple of options:

- Somehow make krb5-config know about the different runpath options supported
  by different compilers.  I guess dealing gcc and a vendor-supplied compiler
  would be the big ones.

- Always return the same flags (the current behavior, I guess), and let the
  user have to manually fix this up.  Alternatively, you could supply the
  compiler to krb5-config, and if it doesn't know about it, you could return
  an error and the autoconf macros could notify the user that he will have
  to configure Kerberos manually.

I still think it would be reasonable to support a manual configuration,
especially if krb5-config output is wrong (which does happen on occasion).

--Ken