general questions about asdf HOT 5 CLOSED

One thing that can be done now is perhaps write out the version of Pyasdf that created the file.

Yes -- the size of the block header recently changed to support compression. Once we have a first stable numbered release of ASDF, then all versions of pyasdf going forward will support all versions of ASDF, but right now, I'm not dealing with such complications. The ASDF version number is at the top of every ASDF file, and each tag within an ASDF file is individually versioned, so this won't be difficult going forward. One of the major drawbacks of FITS is that it has never had a version number.

Why the big difference?

I suspect the difference is the inclusion of arrays in dist.asdf, but I can't really say. Can you share the files with me?

from asdf.

On the performance issues. There was one obvious bug in pyasdf that was causing runtime to increase exponentially with the number of items in the tree. That should be fixed by #92.

After doing that, the breakdown on the image.asdf example is essentially:

• 25% YAML parsing (in pyyaml)
• 25% JSON schema validation (about 25% of which is in pyasdf, the reset in jsonschema)
• 50% Construction of models

There may be some low-hanging fruit for optimization in models. And there's probably some extra in pyasdf, but not as much.

from asdf.

This sounds great. Thanks.

from asdf.

I'm making a little bit of progress on the construction of models. I just eliminated some overhead in creation of model classes, which happens every time you build a compound model. Will do better still once it's possible to convert an expression tree directly to a model class without making intermediate classes.

from asdf.

I'm going to close this because I think I've got all of the low-hanging performance fruit in pyasdf here.

from asdf.