Q: System.IO.Pipelines - Performance and Memory Allocation

[grahamehorner]

Great work, give the improvements in performance and memory allocations are there any plans to provide new/additional implementations for features like XmlSerializer and other features that currently use System.IO. Stream etc.. IMHO it would be good if implementations of Text, Xml, Json, and/or Csv reader/writers where included especially given the memory improvement for when a developer is targeting IoT devices.

[khellang]

It's already being worked on, with JSON as first priority (see dotnet/corefx#33216 and dotnet/corefx#34485).

I doubt XML and CSV will come around in the near future as most things are JSON nowadays anyway.

Maybe a nice opportunity for a community project?

[grahamehorner]

@khellang that is great to hear; I do think however XML serialization is important as a large number of services are very much WSDL/SOAP based and XML is IHMO the performance of SOAP client/server performance would benefit greatly from using the pipeline version of the serialization; while maintaining backwards compatibility.

I'll have a look at the WIP referenced and possible start of an XmlParser/CsvParser etc. based on the concepts/interfaces around the Json implementation using pipelines

[ahsonkhan]

Maybe a nice opportunity for a community project?

Definitely.

I'll have a look at the WIP referenced and possible start of an XmlParser/CsvParser etc. based on the concepts/interfaces around the Json implementation using pipelines

If you have a concrete usage scenario and can showcase significant performance improvements, feel free to file an API proposal within corefxlab (following up with a project with the implementation).

I am sure we can find some steward to review it and provide feedback along the way. Feel free to ping me on the issues as well.

cc @danmosemsft, @joshfree, @krwq, @davidfowl, @pakrym

[grahamehorner]

@ahsonkhan thanks I will do, keep up,the great work everyone loving that the teams are looking at memory footprint and performance at every level and seeing the hardwork show up in the techempower benchmarks