On May 4, 2014, at 11:39 PM, Michal Orynicz <[email protected]> wrote:
> Personally, I'd think that it should fail, as the features that were
> specified to be active cannot be activated. This way the user is forced to
> consciously change desired configuration to one not containing the packages,
> or install the required packages.
Personally, I think that, for many features requiring third-party libraries,
there are *THREE* choices:
1) include the feature iff the third-party library is present, which is
the default with autotools, and should be the default for CMake;
2) fail if the third-party library is present;
3) don't even try to link in the third-party library.
Autotools supports all three for those features - you can:
give no --with-XXX or --without-XXX option, and get the first choice;
give --with-XXX, and get the second choice;
give --without-XXX, and get the third choice.
We currently don't offer all three choices with CMake; I have filed bug 10074
about that.
> On the other hand, we could just make CMake turn off the features which
> findpackage could not find.
That's the best compromise, if bug 10074 isn't fixed, and was the behavior
before the recent change. I have checked in changes CMakefile.txt to
1) properly set the XXX_REQUIRED variables, so that for packages for
which XXX_REQUIRED is set, the configure will fail in find_package;
2) *not* fail merely because a non-required package is missing.
(Note also that, for third-party libraries that are platform-specific, the
build should *NEVER EVER EVER* fail if a library is absent because it doesn't
exist for, or isn't relevant to, your platform. Builds should, for example,
*NEVER EVER EVER* fail on any non-Linux OS if libnl is missing.)
___________________________________________________________________________
Sent via: Wireshark-dev mailing list <[email protected]>
Archives: http://www.wireshark.org/lists/wireshark-dev
Unsubscribe: https://wireshark.org/mailman/options/wireshark-dev
mailto:[email protected]?subject=unsubscribe
