Adding H5MDWriter to H5MD.py

[edisj]

Is your feature request related to a problem?

Currently in my pull request (PR#2787), we've only implemented an H5MD file reader. I'd like to implement a writer as well using H5PY's libraries for writing HDF5 files. H5MD files follow specific rules for the structure of groups and datasets. There's already a package, pyh5md, that can write H5MD files nicely, but it's no longer being maintained by its author, so we've decided to stick to H5PY for writing the files.

Describe the solution you'd like

We've been given permission from Pierre de Buyl, the author of pyh5md, to borrow code from his library and implement it into H5MDWriter. I've already begun a draft that's mostly working (uploaded to my github here), where I modeled it after NCDFWriter and added/tweaked code from pyh5md. I would like to open a pull request as soon as the reader is merged.

Describe alternatives you've considered

We could write the writer from scratch without using pyh5md, but that sounds a lot harder.

Additional context

H5MD format
NCDFWriter - the writer I used as a template to write mine
pyh5md code - used TimeElement class, create_box(), default_chunks(), and element()

[orbeckst]

• make a PR (or does it require add h5md format #2787 to be merged first?)
• When we use pyh5md code then we
  • add it as separate commit and set the author to @pdebuyl --author="Pierre de Buyl <...>
  • add pyh5md license information to the LICENSE file

[edisj]

It doesn't have to be merged first, I guess. I just thought it would be awkward to keep pushing commits to both PRs from the same file. Is there a more convenient way to handling that than, say, deleting H5MDWriter, pushing to reader PR, then deleting H5MDReader and pushing to the writer PR?

edit: nevermind, I think this is all handled by creating a new branch

[orbeckst]

Yes, indeed, branches!

[orbeckst]

FYI: discussion on empyty boxes #1738

[pdebuyl]

H5MD allows "boundary" (a length D vector attribute) to be none.

https://nongnu.org/h5md/h5md.html#simulation-box

[orbeckst]

@edisj is your writer PR out of draft stage?

[edisj]

@edisj is your writer PR out of draft stage?

yes - I've just pushed a working cleaned up version to the pull request #2869