fix bare configure build

[elliefm]

This re-enables Cyrus builds configured with just ./configure, mostly by detangling some header mess and adding some judicious #ifdefs where necessary.

Also annotates a bunch of Cassandane tests with their dependencies so that they won't try to run if the feature they need wasn't compiled in. I did this in the simplest way possible because my goal right now is something backportable to 3.8, but I don't really like it. (For example, lots of tests needed to be individually annotated ":needs_component_httpd", even though that test doesn't need it, just because the suite itself starts the httpd service. In the future it would be nice to handle this better, but that's a non-trivial refactor.)

Fixes #4409

Reviewers: please build and test this with both:

• your usual configure arguments, to make sure I haven't broken anything any of us rely on
• a minimal configure (I used ./configure --enable-silent-rules --prefix=/my/cass/prefix), to make sure the fixes are adequate with your compiler/environment and not just mine

If you run Cassandane on the minimal build, you'll (probably) find a bunch of Foo.relocate_* and RelocateByID.* tests have errors. I don't know why, but I've put details in #4409. Since this PR is a prerequisite for even building like that at all, it seems sensible to get the build problem fixed first, and then investigate the relocate failures.

[ksmurchison]

@elliefm My normal build still works fine.

I tried a bare build, but mbpath.c failed to compile due to an unused function complaint. I fixed this on master.

The next problem is that master.c uses pow() and therefore needs to link with libm. We should probably just add this to LIBS unconditionally and be done with it.

The RelocateById failures can be fixed by removing the #ifdef WITH_DAV from mboxname.c:mboxname_metapath()

The Sieve.date_iana_tzid failure (when rebased on master) can probably be fixed by adding :needs_component_httpd since since this test requires libical

[ksmurchison]

This looks good after the pow() compile issue is fixed.

[elliefm]

For my future reference: the mbpath fix Ken mentioned was 9034a59 -- I'll need to grab this when I backport to 3.8

[elliefm]

@ksmurchison

The next problem is that master.c uses pow() and therefore needs to link with libm. We should probably just add this to LIBS unconditionally and be done with it.

It already links with libm if required, thanks to 4830785, using LT_LIB_M...

But LT_LIB_M fails when compiling with CFLAGS=-Werror, and as a result doesn't detect/set LIBM='-lm'.

The HTTPD section handles this differently, with a comment describing a similar problem. So I guess your minimal build still has CFLAGS=-Werror, and that's why it fails to detect LIBM, and fails to build master. Whereas my minimal build doesn't have CFLAGS set, because that's usually set by my wrapper script, which I'm not running.

But that means in both of our usual builds, we don't get LIBM, and yet master builds fine in that case. That's weird.

I have an idea for making this work (save CFLAGS, set CFLAGS=$CFLAGS -Wno-error, run LT_LIB_M, restore CFLAGS) which I'll tinker with in a bit. If that works, we can clean up the HTTPD section too

[elliefm]

@ksmurchison: I've rebased on master, added your suggested fixes for the RelocateById and Sieve test failures, and tidied up our libm handling in the way I proposed in my previous comment. Seems to work properly for me now, in both full builds and minimal, and with and without -Werror. Can you try it out again please, and confirm whether it's fixed on your system now too? Especially the pow() bit. Thanks

[ksmurchison]

@elliefm The LIBM hack didn't work for me with -Werror in either config. If I remove -Werror, its good in both.

[ksmurchison]

LIBM still not linked in for me with -Werror.

@elliefm In my script, I set CPPFLAGS rather than CFLAGS. If I do the same save/set/restore magic with CPPFLAGS, everything works fine. I suggest we do the magic with both variables.

[ksmurchison]

I approve once we add save/set/restore magic for CPPFLAGS around LT_LIB_M

[elliefm]

Ahh yeah, right!

[elliefm]

@elliefm 3.6 needs at least 7b8c76f, which will fix #4848. Also check the other commits on this PR