fatal error: Don't know how to read Double about hdf5kit

[alejandro-isaza]

Are you working on a OS X app? We ran into this same problem and I suspect the warnings you get when you launch are the root cause:

Class SwiftObject is implemented in both .../libswiftCore.dylib and .../MyApp. One of the two will be used. Which one is undefined.

Looks like it's loading two versions of the swift standard lib. The only way to resolve it was to use Dataset.select() and readDouble() instead of the subscript helpers. At least until they fix these Swift issues.

[timburgess]

Thanks. Yes, it's an OS X command-line app. From what you say, it seems like avoiding swiftcore references means avoiding the two version loading issues. I have tried a few alternative ideas but unfortunately none successful. I have also hit http://www.openradar.me/23551273 which is what is causing the Xcode crashes.

We should probably keep this issue open until there's a resolution for the double library loading.

[alejandro-isaza]

This commit should fix this: 954561f

Also let me know what you think of that new approach.

[timburgess]

I've had the swift-build branch working fine on OS X with the Swift 2.2-dev snapshot that Apple released. The only slight quirk is that there is a import CHDF5 required but that's no biggie.
Were you planning on merging that branch into master or bringing these changes to the build branch?

[alejandro-isaza]

I haven't figured out hot to support iOS from swift-build, so I'll leave that in a branch until that's resolved. Merged latest changes into swift-build.

[timburgess]

The changes are an improvement I think to the prior methods and it all builds and works for me on swift-2.2-dev.

let dataset = file.openFloatDataset(nodeName) is better than let dataset = file.openDataset(nodeName, type: Float.self) and of course eliminates the cast required from Any to [Float] on the dataset read (which is a good thing!). The ultimate ideal of course would be just let ds = file.openDataset(nodeName) or even better let ds = file["nodeName"]. IMO the latter enforces the concept that an HDF5 file is essentially a dictionary containing datasets.

With swift-2.2-dev, data.read() now produces an ..is not marked with try issue but that doesn't concern me as I think that we should point people towards using subscript access rather than calling data.read(). So it would be nice to have support for let foo = dataset[] rather than let foo = dataset[0..]

My experience with scientist/progammers is that they know what the HDF5 names are that they want and they have a sense of what the data type is, and most of the time, they just want to get all the data in memory.

There is a long term issue with supporting complex HDF5 types i.e. tuples containing strings and decimals but IMHO those datasets are pretty few and far between.

Lastly, this all works for me so I have no problem with closing the issue.

[alejandro-isaza]

I added throws because I think it makes more sense than silently failing. subscript doesn't support that but I think it's good to have a way of not having to deal with the throws. I agree with reading via dataset[], I'll add that.

👍