Eliminate compatibility mode from HDF5 I/O

[matteobachetti]

As per #8888

• Eliminate compatibility mode writing
• Changelog etc.

@taldcroft, @pllim, @astrofrog, @bsipocz et al.: is there a use case for being able to still read the old format?

[pllim]

Thanks! I don't use HDF5, so I cannot comment. 😬

[bsipocz]

Please remove changes to the helpers.

I don't use hdf5 files so don't know how widely the old formats are used, and what the actual issue with it. Naively I would say we should be able to read in as many types of files as possible, but should not write non standard or obsolete versions of them.

[matteobachetti]

Oh, sorry for the helpers thing. Done

[matteobachetti]

@bsipocz the main problem I see is that the old reader now remains untested because we have no way to write old-format files. Besides that, I would also prefer to keep as many formats as possible

[pllim]

Re: testing old format reader -- Is it possible to write it out now while we can and store it as a test data file (either remote or local depending on file size and your philosophy about data files).

[astrofrog]

I agree with removing this, and also agree with @pllim's suggestion - we should basically take the example from the first half of test_preserve_serialized_compatibility_mode, write it out as a file, store it with the data files for tests (inside a data directory), and use it to continue to test the reading. I think the file will be very small, so I'm not worried about it, and we shouldn't have to change it in future since it's just a legacy format.

[astrofrog]

@matteobachetti - could you rebase, to get rid of the merge commits?

[matteobachetti]

@astrofrog: done, but 32-bit tests are falling and I'm not sure why

[matteobachetti]

Oh, I see the likely problem with the 32-bit build: #8934

[bsipocz]

@matteobachetti - yes, failures should be unrelated. Given that this needs a rebase anyway due to the unrelated commit, I would say rebase after #8945 (or alternative) is merged.

[matteobachetti]

Oh, I wonder why that commit is there at all. I already rebased in theory O_o.
Anyway, the plan sounds good. Waiting for solution for 32 bits, then rebase. Thanks!

[bsipocz]

@matteobachetti - the workaround in merged, so you can rebase. You can manually remove the commit if you rebase in interactive mode.

[matteobachetti]

@bsipocz rebased and dropped commit.

[bsipocz]

Review comments have been addressed.

review has been addressed

[bsipocz]

Thanks @matteobachetti