Re: Proposal to make better use of 5th arg to AC_CHECK_LIB and AC_SEARCH_LIBS

[Erez Zadok]

In message <address@hidden>, Paul Eggert writes:
[...]
> What if there were multiple -lxxx args, e.g.: AC_CHECK_LIB(hesiod,
> hes_init,,, -lresolv -lsocket) Would you want it to try all
> combinations ('', '-lresolv', '-lsocket', '-lresolve -lsocket')?
> Seems like that would be a win in some cases.
> 
> 
> My own impression is that there are a lot of possibilities here.  I
> agree that AC_CHECK_LIB does not have good semantics.  However, I
> think your proposal might be improved and/or generalized too.  If it's
> an incompatible change, I'd rather see an improved/generalized
> proposal now rather than install one incompatible change now, and
> another one later.

Hi Paul, I wrote the original AC_CHECK_LIB2 for am-utils b/c it served an
important purpose for the am-utils package, exactly as Rainer stated.  I do
not think that the current AC_CHECK_LIB behavior makes sense.  I think
autoconf need to modify that behavior or provide a new, more powerful macro.

I like your idea of generalization of this macros.  I would suggest against
a fancy macro that will test any permutation of N libraries: that would be
2^N checks and is just too costly, plus I don't think it's needed.  Many
autoconf macros simply search in order, and that's probably what I'd
prefer to see.  Here are two ideas for new/revised macros:


AC_CHECK_FOO(fxn, lib1, lib2, ..., libN)

        Checks if function "fxn" can be satisfied without linking with any
        libraries (other than those already appended to $LIBS).  If not
        possible, checks if "fxn" can be satisfied by linking only with
        lib1.  If not, try only lib2, and so on.  This is a serial test for
        a single library at a time.  If no single library can satisfy fxn,
        then try to link against lib1+lib2, then lib1+2+3, and so on until
        all libraries are tried.  This will require exactly 2N tests, much
        less time than 2^N and will satisfy the most demanding applications
        (like the crazy linking rules we have to put in am-utils,
        specifically to avoid bad libraries).

AC_CHECK_BAR(fxn1, fxn2, ..., fxnN,  -lib1, -lib2, ..., -lib3)

        Same as above, but check if all of fxn1 through fxnN can be
        satisfied by some of the libraries listed, in the same checking
        order as AC_CHECK_FOO.  This is useful IMHO b/c there are times
        where I need a series of functions to be found, or not at all.  That
        is, checking for hes_init is not enough to tell me if hesiod is
        fully working on a given system (there are different versions of
        hesiod out there, Solaris 7 has a partially working LDAP system,
        etc.)  A macro like this would ensure that I will only use the
        functions listed above if all of them can be found.   Otherwise, I'd
        rather not have a half-working implementation: it's better to have
        none of it.

        An alternative to this macro would check for any one of fxn1--fxnN
        in a series of libraries.  If the first idea checked for all of them
        (logical AND), this one will check for at least one (logical OR).
        This is useful b/c some times I need to figure out if a feature
        exists (say, ldap) on a system, but I have to check for it in
        various ways b/c of incompatible functions and different generations
        of (ldap) libraries.  I'm not sure if this OR alternative needs a
        separate macro: we could just execute a series of AC_CHECK_BAR
        macros with only one fxn in them, and the same series of libraries.

Thanks,
Erez.