random_choice missing from fable library

[joprice]

Description

A runtime error is hit when trying to use Seq.randomChoice:

ImportError: cannot import name 'random_choice' from 'fable_modules.fable_library.seq'

It looks like this is a group of relatively recent helpers from this rfc (https://github.com/fsharp/fslang-design/blob/main/RFCs/FS-1135-random-functions-for-collections.md), and need to be added to the replacements:
dotnet/fsharp@54a1997#diff-e145f82a233516da994063cddbbe920e616e734ee6ab652eb4c8a21f76382ef9

Related information

• Fable version: 4.23.0
• Operating system: OSX

[MangelMaxime]

Thanks for the report, we should indeed add them. It looked like really interesting API when I was reading about them last week.

If I remember correctly, Seq, List and Array in Fable are implemented using F# directly so perhaps it will not be too difficult to do. I think the things we need to be careful to test is the that the Random seed works for reproducing the randomness.

On another note, I am a bit surprise that you have this error and not a "nicely formatted" Fable error message about not all BCL API being supported. But this perhaps, this is because we just forward the call without checking if we actually have an implementation in the Seq.fs, List.fs, etc.

[joprice]

Yes I expected the typical unsupported error as well. I guess to handle that every supported method needs to be listed out in the replacements to avoid new passing through.

[MangelMaxime]

I guess to handle that every supported method needs to be listed out in the replacements to avoid new passing through.

Indeed, I often think about if we should or should not do it.

On one side, have an exhaustive list of the support API make the DX better because Fable can capture the error.

On the other side, it makes adding a new API a bit more verbose and also means that for every API call Fable will need to check the list of supporting API. We would probably not want to use a list for that but a Set or Hashmap if I remember correctly to get a O(1) search.

IHMO I think the pros are worth considering.

[joprice]

Yes, luckily it results in a hard error that prevents the script from executing, but having a compile time error benefits library development as well to find unsupported usages without requiring full coverage.

It could end up being a lot of string comparisons in a large codebase. A hash or treemap/set lookup or even a trie could be used to optimize it.