Ticket #4214 (closed Patches: fixed)
Use collection_size_type/version_type consistently (fixes Boost.MPI)
|Reported by:||dgregor||Owned by:||ramey|
|Version:||Boost Development Trunk||Severity:||Problem|
Boost.MPI has some special archivers (e.g., ignore_skeleton_oarchive) that ignore some of the structural aspects of serializing data. It does so by recognizing attempts to serialize values of type collection_size_type and version_type and then silently ignoring them which, in the narrow domain that ignore_skeleton_oarchive is used, works well. The current crop of failures in the MPI library (skeleton_content_test-* are failing sporadically) all appears due to inconsistent use of collection_size_type/version_type.
The attached patch updates the serialization library to use collection_size_type and version_type consistently. Tests pass for GCC and Clang with both the Serialization and MPI libraries, and this ends up fixing a pretty serious regression in the MPI library.
Currently, Boost.MPI is broken on both trunk and the release branch because
- Status changed from closed to reopened
- Resolution fixed deleted