mokus0 / bindings-hdf5 Goto Github PK

Specifically, the same version of the bindings library can have several different interfaces depending on the version of hdf5 that is being wrapped.

1. Is that a problem?
2. What should be done about it?

One option is to have a different cabal file for each supported version of the hdf5 library. That could make dependency resolution really messy though, depending on how cabal handles pkgconfig dependencies (and whether pkgconfig is even the way to go)

Another option is to just let it be, and let that be something the "high level" wrapper must abstract over.

Another option is to include in the bindings-hdf5 package a "minimal" abstraction over it, which wraps all versions into the same interface (presumably the latest). This seems like the best option overall, but I'm not sure offhand how it should be done. For example, what to do about enumeration constants or macros that don't exist?

First, what do I want the types to express?

• the underlying C type of the pointer
• whether the pointer refers to a single cell or to an array (and if an array, how its size is determined)
• whether the caller needs to allocate a buffer
• whether the caller needs to free the returned pointer

Second, can we come up with a nice way of automating the wrapping? Something like [1] but with less type-level clutter for the user. Ideally it would automatically unwrap all the extra parameters and return types. TH is an option but if it can be done dynamically I think that's better.

Third, something like this would be more widely useful than just in this project, so it should eventually be separated into its own project.

[1] https://github.com/mokus0/junkbox/blob/master/Haskell/FFI/With.hs

HDF5 1.8.11 removed functions H5Pset_dxpl_multi and H5Pget_dxpl_multi. Please remove them from src/Bindings/HDF5/Raw/H5FD/Multi.hsc as they prevent building hs-hdf5.

See the changelog at http://www.hdfgroup.org/HDF5/doc/ADGuide/Changes.html#1811

Thanks for your work.

I don't have an MPI-enabled build of HDF5 right now, so I don't even know whether the Bindings.HDF5.H5FD.MPI module will compile or not.

For now I've got these hard-coded in the cabal file. Figure out what it takes to set up something like pkgconfig.

I'm not sure the best way to handle this. It needs to be a global transformation, but aware that some prefixes are longer. It might work to simply take the maximal prefix of uppercase letters (and '5'), because it seems like every identifier has the prefix followed by either a lowercase letter or an underscore. It would also be nice if it added underscores when lowercasing names such as "H5Tfoo", so that the boundary between prefix and the rest remains well-defined.