Raw memory passing

[Scooletz]

I did my best to understand JsValue and its semantics of value passing. Has it been considered to provide another JsValue that could represent an ArrayBufferInstance that would be based on ReadOnlyMemory<byte>. This could be beneficial for scenarios where raw data are loaded in .NET runtime and passed to some script for executing logic over it. Additionally, depending on how closely Jint needs to be in regards to the buffer detachment, an ArrayBufferInstance that is based on ReadOnlyMemory<byte> could allow a clone-less copying, just by slicing over the underlying ReadOnlyMemory<byte>.

[lahma]

This sound interesting. So do you have an actual use case that you are building?

ArrayBufferInstance is by nature read-write memory so it doesn't directly map to ROM. How would you use this type, constructed on C# side and then fed to engine via engine.SetValue("variable", myInstance) or would this require deeper customization like that you would control in instantiation of default ArrayBufferInstance? ArrayBufferInstance could probably model it's backing store using Memory<T> (write required). Would your custom type be inheriting from default ArrayBufferInstance and overriding some methods and throwing InvalidOperationException in case of writes?

There's also DataView to consider.

[Scooletz]

I'm currently sketching a filtering engine that would either run the filter in .NET or in JS. As the data are uint32, which falls into TypedArray and ArrayBuffer. If the latter option is selected, it would be nice to be able to pass large slabs of data over without copying at all. I'm somewhat familiar with the DataView but as it's created on top of an ArrayBuffer and in Jint it's internal, I don't know how this helps or prohibits ArrayBuffer considerations.

If the idea seems plausible, maybe I can dive a bit deeper into Jint and try to sketch a PR with it :-)

[lahma]

A PR would be most welcome and appreciated! Like any OSS project, we happily accept help and appreciate it, there's never too much of help available. Many things in Jint are modeled based on JS spec but we would love to have first class .NET experience with comfortable APIs that "make things just work".

Some background:

• We make things internal at first to allow some wiggle room, when a plausible use case is found things can be made public and just require some more scrutiny to ensure we don't create "bad APIs", we have Jint.Tests.PublicInterface for the API promises we've made and try to maintain
• We go spec first, so EcmaScript has all sorts of weirds things that need to be supported, some code isn't optimal at first (hopefully compliant!), but we try to optimize when performance problems are found - as an example arrays in JS have a lot of special casing but we try to make them fast

With your track record, any PR to improve the project is most welcome 😉

[Scooletz]

Sounds like a plan. I'll dig deeper then and if in doubt will raise it here.

[Scooletz]

Will be implemented in PR #1590. Closing the discussion as the rest will be implementation specific.