[CMake] [PATCH] slightly modify cache variable behaviour, Was: Re: option() behavior

[Alexander Neundorf]

On Tuesday 12 May 2009, Philip Lowman wrote:
> On Mon, May 11, 2009 at 4:53 PM, Alexander Neundorf
> <a.neundorf-work at gmx.net
>
> > wrote:
> >
> > On Monday 11 May 2009, Bill Hoffman wrote:
> > > Alexander Neundorf wrote:
> > > > On Monday 11 May 2009, Bill Hoffman wrote:
> > > >> So, CMake has done what it does now from the start.   There was a
> >
> > short
> >
> > > >> period of time when it did not, and that was when a re-write was
> > > >> done, and it quickly broke some existing projects.  Here is the
> > > >> thread:
> > > >
> > > > Yes, I can imagine that.
> > > > Still the current behaviour is not very intuitive, especially that
> > > > set(CACHE) in the first cmake run overrides the "visible" variable,
> >
> > while
> >
> > > > the set(CACHE) in the second run basically does nothing, and so the
> >
> > value
> >
> > > > of the "visible" variable is different.
> > > > I think for CMake 2.8 or 3.0 it would be a good idea to make this
> > > > more consistent and easier to understand (especially since now more
> > > > and more people are using cmake).
> > > >
> > > >> http://www.cmake.org/pipermail/cmake/2007-March/013204.html
> > > >>
> > > >> I think the rule should be if you want to change a variable you have
> >
> > to
> >
> > > >> have the set AFTER the variable, or use CACHE INTERNAL.
> > > >
> > > > What do you mean exactly with "have the set after the variable" ?
> > > > Have the set(FOO <value>) after the set(FOO <value> CACHE), or
> >
> > something
> >
> > > > else ?
> > >
> > > Yes,
> > >
> > > This always has worked as expected:
> > >
> > > set(FOO 2 CACHE)
> > > set(FOO 1)  # this is always the value
> > >
> > > However, I think your change is OK.  Basically if the set CACHE comes
> > > after the set, it will always set the value.   We do need a policy for
> > > this, but it should be easy to detected.  You will have to see if the
> > > REMOVE is going to do anything.    I wonder how many warnings that
> > > policy will cause... I  guess we could create the policy code, and try
> > > it on VTK, Boost, SecondLife, and some other big projects to see what
> > > happens...
> >
> > Ok, I put it on my TODO, then we'll see what (would) happen.
>
> I'll be happy to test it out on our build system at work as well.
>
> I never thought to do anything crazy like define a local variable before a
> cache variable until CMakePorts... so I don't think this is going to impact
> us at all and I suspect it probably won't affect others much either.
> Wrapping the cache and option declarations with if(DEFINED...) is no big
> deal.

The patch attached at http://public.kitware.com/Bug/view.php?id=9008 
implements this changed behaviour, together with a new cmake policy CMP0014 
(OLD behaviour: ignore the value in the cache if the variable has been 
already set, new behaviour: set the variable to the value from the cache)

The cmake tests are all succeeding, and I'm also building kdelibs without 
problems with this patch.

If you think it's ok I can commit it to HEAD.

Alex