variableBased Encoding should be limited to ADIOS

[guj]

Describe the bug
variableBased encoding with HDF5 produced error for not able to create (an existing) dataset.

To Reproduce
in WarpX, let drag use both h5 and v; e.g.
diag1.openpmd_encoding=v
diag1.openpmd_backend=h5

Expected behavior
HDF5 will throw error when written second step.

** comments **.
I think possible solution can be at setIterationEncoding(inputEncoding), it first check with I/O handler whether inputEncoding is allowed. If not, fall to a default one (groupBased)

[franzpoeschel]

Thank you for opening an issue for this

I think possible solution can be at setIterationEncoding(inputEncoding), it first check with I/O handler whether inputEncoding is allowed. If not, fall to a default one (groupBased)

I think that we should recognize this situation and throw an error accordingly. It's simpler to do and does not do unexpected things. Also note this test: Variable-based encoding is supported for all backends as long as only one iteration is written. The intended use case is for datasets that don't need iterations and can hence forgo one layer in the file hierarchy.

[guj]

True, files with one iteration should always to allowed.
Then defaulting variable encoding in h5 to file encoding would cover it.

I am not sure whether it is serious enough to exit with error just because variable encoding in not supported.
If I were a client, I might prefer to continue with a supported encoding.

I think that we should recognize this situation and throw an error accordingly. It's simpler to do and does not do unexpected things. Also note this test: Variable-based encoding is supported for all backends as long as only one iteration is written. The intended use case is for datasets that don't need iterations and can hence forgo one layer in the file hierarchy.

[franzpoeschel]

I am not sure whether it is serious enough to exit with error just because variable encoding in not supported.
If I were a client, I might prefer to continue with a supported encoding.

This situation means that a user configured openPMD in a way that implies data loss. We do have error::OperationUnsupportedInBackend for exactly those cases.

This situation can only be detected at the point where the user tries opening a second step, since we support writing a single step. At this point, it's too late to fix things since the first step is already written.

[guj]

OK, sure.

This situation means that a user configured openPMD in a way that implies data loss. We do have error::OperationUnsupportedInBackend for exactly those cases.

This situation can only be detected at the point where the user tries opening a second step, since we support writing a single step. At this point, it's too late to fix things since the first step is already written.