On Thu, Dec 12, 2013 at 5:30 PM, Satish Balay <balay@mcs.anl.gov> wrote:
On Thu, 12 Dec 2013, Dmitry Karpeyev wrote:

> On Thu, Dec 12, 2013 at 5:05 PM, Satish Balay <balay@mcs.anl.gov> wrote:
>
> > On Thu, 12 Dec 2013, Dmitry Karpeyev wrote:
> >
> > > Is there any reason 'master' has PETSC_VERSION equal to the latest patch
> > > release version,
> > > rather than the first upcoming release version?
> >
> > All changes to petscversion.h in maint get merged to master.
> >
> Yes, I see the problem.  I don't necessarily have a (good) solution, short
> of creating
> a separate branch for each patch release (e.g., maint-3.4.3).  Those could
> be ahead
> of maint by a single commit such as those that are currently used to
> increment the patch level.
>
> >
> > And 'PETSC_VERSION_RELEASE'  0 vs 1 is supporsed to identify petsc-dev.
> >
> > > Surely, the use of PETSC_VERSION_LT() makes it relatively easy to guard
> > > against API changes in derived software, but it's a bit odd to see
> > > #if PETSC_VERSION_LT(3,5,0)
> > > /* Works for petsc-3.4.x and earlier. */
> > > #else
> > > /* Supposed to work petsc-3.4.x as long as !PETSC_VERSION_RELEASE */
> >
> > It should be 'latest petsc master' here. 3.4.x doesn't have any meaning
> > here..
> >
> Yes, but master today has version number 3.4.3.

As mentioned 3.4.3 has no baring on master. only PETSC_VERSION_RELEASE==0 is relavent.

> In some code that depends on PETSc (e.g., libMesh) I have to use guards
> like these:
> #if PETSC_VERSION_LESS_THAN(3,5,0) && PETSC_VERSION_RELEASE
> /* releases petsc-3.4.x and earlier. */
> #else
> /* master */
> #endif

I don't see a problem here [except for the extra '&&
PETSC_VERSION_RELEASE' which shouldn't be needed.


> Note that I can't always use the more succinct PETSC_VERSION_LT(),
> since it doesn't exist before something like petsc-3.3.

Presumably PETSC_VERSION_LESS_THAN() is the macro you created. Why not
just copy over PETSC_VERSION_LT() exactly as-is [with dependent
macros] - and call it PETSC_VERSION_LESS_THAN() [or equivalent name] -
if you need to maintain compatibility for petsc-3.2 or older.
 
In this particular instance (libMesh) it's a libMesh macro. libMesh doesn't rely on petscversion.h, 
but rather detects the petsc version numbers during configure.  It doesn't use PETSC_VERSION_RELEASE, 
so it's not an entirely trivial fix (at least for me).
I build libmesh against petsc-master fairly regularly, but I don't know how much general interest there is 
in this.  I'm forwarding this to libmesh-devel just in case. 

Dmitry.



> I could guard for pesc-3.3
> and then use the shorter macro inside, but that defeats the purpose.
> Meanwhile, these additional PETSC_VERSION_RELEASE clauses look odd.

I don't understand which part looks odd - and how adding a version
numer to petsc-master helps here.

Satish

> They could be removed after petsc-3.5 has been released, but that's another
> maintenance headache...
>
> Dmitry.
>
> >
> > Satish
> >
> > > #endif
> > >
> > > Besides, master's "version" changes with every patch release.
> > > It might not be of much practical importance, but looks a bit strange, in
> > > my opinion.
> > > At least intuitively, I think of 'master' as a sort of 'release
> > candidate'
> > > for petsc-3.5.0.
> > > Wouldn't it be natural to name it that way?  Is there a technical reason
> > > not to?
> > >
> > > Dmitry.
> > >
> >
> >
>