> On Feb 23, 2015, at 4:05 PM, Jed Brown <[email protected]> wrote: > > Barry Smith <[email protected]> writes: >> The directory include/petsc-private/matimp.h is always provided >> with a PETSc installation so though _p_Mat is "private" to PETSc >> you can extend PETSc "out side of the library proper". > > What you don't get when accessing the private headers is a guarantee of > backward compatibility (even across subminor releases) or documentation > of changes in the release notes. If people are using private headers > frequently, we should strive to make a public interface to fulfill their > needs.
If they are writing code that they intend to share with the PETSc community (i.e. eventually it becomes a pull request into petsc so the user will not maintain it themselves for a long time)*; like I assumed in this case, then it is fine to use the private headers. Barry * For example providing a piece of functionality that we frankly just forgot about.
