Conversation
Newer GNU linker refuses to use transitive linking; if netcdf calls MPI_Allreduce, it must link the providing library directly. Also switched to CMake because I don't know autoconf well enough to know how to put MPI stuff into it.
|
🎉 |
|
The initial netcdf release with CMake support is 4.2.1. |
|
@mathstuf: Can the MPI patch be done with a |
|
For something like this, I think a patch is better. Adding lines with no context will only mean that we add duplicate lines or add it to the wrong place if things change between versions. |
|
Will a patch still apply if the context line numbers don't match? I seem to recall there being a threshold, so maybe it's ok. I think it's fine to leave it as a patch for now. |
|
There's a "fuzz" factor it uses to try and "wiggle" the code in by removing context lines. By default, it is 2. |
* Refactoring neurodamus building for python support * Several neurodamus improvements - syn2 variant renamed to syntool, and made default - exporting location of NEURODAMUS_PYTHON, not PYDAMUS - more refactoring
* Refactoring neurodamus building for python support * Several neurodamus improvements - syn2 variant renamed to syntool, and made default - exporting location of NEURODAMUS_PYTHON, not PYDAMUS - more refactoring
…ild_type_prevent_overlinking Prevent overlinking against MPI libraries in ECMWF software and add support for different build types
3 participants
Newer GNU linker refuses to use transitive linking; if netcdf calls
MPI_Allreduce, it must link the providing library directly.
Also switched to CMake because I don't know autoconf well enough to know
how to put MPI stuff into it.