Ian's "Merging feature work to Mesa master" calls for not merging any features to master until a vertical slice of functionality is complete. In other word, no Mesa features would be merged (they would just be sitting in separate Mesa branches or in patch form) until something can be demonstrated that stresses all components that the given work deals with. This would eliminate broken features sitting around in Mesa or where a feature isn't fully implemented and just cluttering the code-base and making it more challenging for other developers to find where another developer left off.
Ian's other proposal, which he is seekign comments on as well, is that no feature should be merged to master under "a robust set of tests" are implemented or the feature at least documented. This could be Piglit regression tests or at least simple demos that take advantage of the newly-presented functionality.
Assuming most of the developers agree with the two proposals, this would affect how Mesa 8.1 is developed in the months leading up to its summer release.