Why is 6.0.2 the min FSharp.Core version requirement? about fsharp.control.taskseq HOT 6 CLOSED

I think I can update the code and make the overload resolution work without needing the NoEagerConstraintApplication attribute.

The problem of using that attribute is worst than just forcing everyone to 6.0.2. AFAIK that attribute won't be recognized by earlier F# compilers.

from fsharp.control.taskseq.

Good points, @gusty. But considering that task uses the same attribute in FSharp.Core, and this library cannot be used without task, the backward compat issue with earlier compilers is irrelevant. Same is true, for instance, for the resumable code in this lib, which isn't supported either by earlier compilers.

@bartelink I've downed the version to 6.0.1. I may reconsider at some point to go to 6.0.0, but that requires a different approach for the overload resolution that I'm not to keen on trying.

This will be part of whatever the next version is (either 0.3.1 or 0.4.0).

from fsharp.control.taskseq.

Tried to reduce the dependency:

• FSharp.Core v 6.0.1 is required for the NoEagerConstraintApplication compiler hint
• the tests pass with 6.0.1

Would still be interested to know the nature of the hard requirement for 6.0.2...

from fsharp.control.taskseq.

I tried the “earliest version possible”, and the NoEagerConstraintApplication is a requirement, the overload resolution won’t work without it.

The earliest I could make work with everything in the resumable code was 6.0.2, but I’ve changed quite a few things since the first TaskSeq versions, it’s possible I removed something that doesn’t require that anymore. I’d have no problem making the dependency on Core a lower version. In fact, that’d be a plus.

I’ll have to dive into history though, hopefully I recorded the ‘why’ of the decision.

from fsharp.control.taskseq.

It was done as part of #87, and from the text, it suggests I tried to go as low as possible. Let me create a quick PR and see what happens if I use 6.0.1.

See #123, running now.

from fsharp.control.taskseq.

Thanks for looking - tbh if I'd seen the need for the attribute, I probably wouldn't have asked in the first instance, but I do appreciate you investigating and the fact that we now have a paper trail ;)

While rippling dependency updates are not ideal, it's clear that there's a reason so this is not a real problem. It's not inconceivable that there are ancillary perf and stability benefits to 6.0.2, so once the option of targeting 6.0.0 is gone it's less critical.

In Equinox and Propulsion, I've since bitten the bullet and upped their min req to 6.0.7 (in order to ensure 6.0.6's fix for the Async.Parallel stack overflow edge case is forced in)

Happy for this to be closed as and when you see fit.

from fsharp.control.taskseq.