How does CSnakes compares to Python.NET?

[rkpasia]

First thing first, awesome library and speech at MSBuild this year.

I'm deeply interested in the inner working of this library and I'm looking forward to develop a PoC for a new industrial application in a manufacturing environment (Computer vision and other stuff related to data science where C# + Python it's a plus).

I really see this library as a possible bridge for data scientists and software engineers where the need for a integrated solution with advanced data analysis capabilities is required to enhance a step in a raw material processing pipeline.

Nowadays in our research and dev settings we found this "bridge" in the Python.NET package, which kinda works as a two-way bridge...
CSnakes seems more a "one-way" bridge to me.. but I'm curious to understand the point of view of the authors of this package.

Did you know in advance of Python.NET? Did you take as a reference by any mean Python.NET?

What's your take on the differences between these two packages?

Thanks for the awesome work!

[tonybaloney]

Thanks for the kind words.

Yes, there are several differences. Early prototypes of CSnakes used Python.NET so I'm very familiar with it. We found several things missing -

• Python.NET doesn't have any support for downloading Python
• Python.NET didn't handle Linux dynamic module loading well
• The C-API wrapper was outdated and needed a lot of work
• Python.NET used Binary Formatter, which is deprecated in .NET (and shouldn't be used regardless)
• The GIL wrapper is quite basic, in CSnakes we have concurrent queues to avoid the GC collector deadlocking.

We also focused on some different design goals, which are

1. Compatibility
2. Performance
3. Ease-of-use

Or, more specifically

• Minimal boilerplate using the source generator
• Native support for virtual environments and package installations
• Support for the latest Python and the latest .NET versions
• Support for async functions
• CSnakes' marshalling (C# <-> Python) is benchmarked to be highly performant.
• We support AoT as well, although that's still in early development

[GeorgeS2019]

@tonybaloney

We support AoT as well, although that's still in early development

Very interesting, where to look for the AOT support part?