Bug: Duplicate can't copy plates with runs

[mtbcarroll]

If a plate has a run then Duplicate errors in duplicating the plate or run.

[mtbcarroll]

Does work for simple plates but fails for the MIAS plate 001-365700055641.

[mtbcarroll]

Duplication seems to work fine for whatever richer plates I construct in integration testing. However, the MIAS plate failure is reproducible: it may point to a general problem with attempting to persist a transient model graph with Hibernate. On looping through the duplicates to persist them I get exceptions like,

not-null property references a null or transient value: ome.model.screen.WellSample.image

but all the well samples do have images, probably just not ones that are yet persisted themselves. While I'd hoped that the twin magic of FlushMode.COMMIT and @Cascade(CascadeType.PERSIST) would suffice, my guess is that I may need to session.persist the new subgraph nodes in an order that has no property refer to a not-yet-persisted one, even though for other plates it works fine even when that constraint is repeatedly violated.

[mtbcarroll]

With some pain I've introduced rather more ordering into how the newly duplicated subgraph is persisted. However, I haven't yet determined what the suitable ordering would be. For example, if I,

1. create a new Image
2. fill in its mandatory properties
3. persist the image
4. create a new Pixels
5. fill in its mandatory properties with its image set to the persisted image
6. persist the pixels

then I get,

not-null property references a null or transient value: ome.model.core.Pixels._pixelsBackref

This is also true if I use image.addPixels(pixels); instead. It fails differently if I use pixels.setImage(image.proxy());.

I assume I'm just forgetting some basic Hibernate stuff here. However, really there are two questions:

1. What should I have done to get that simple scenario working?
2. How can that be automatically determined in the general case?

[mtbcarroll]

I was mistaken: if I instead use image.addPixels(pixels) it works fine. Pixels.getImage() is annotated with insertable=false, updatable=false. Will accordingly look at further revising code.

[mtbcarroll]

Still working through interesting Hibernate issues and making steady progress. The latest is that if I want to add a channel to the beginning of a list of channels on Pixels then I can't just do,

pixels.addChannel(channel1);
session.persist(channel1);
...
pixels.clearChannels();
pixels.addChannel(channel2);
pixels.addChannel(channel1);
session.persist(channel2);

because Hibernate performs the update in a manner that violates channel_pixels_pixels_index_key.

[mtbcarroll]

Fixed by ​https://github.com/openmicroscopy/openmicroscopy/pull/4546.