Please add a changelog entry, fairly important to highlight this feature config change :)

Handling of serial-pass looks broken to me but I'm a bit on the fence whether we still need this one 🤔. I'm actually overall here with what @cwfitzgerald's comment on the issue and would suggest joining all of serde, serial-pass, serialize and deserialize to one feature, I doubt anyone needs this level of granularity realistically.

Anyways, thanks so much for looking into this! Happy to merge once the serial-pass story, docs and changelog is sorted out! :)

Thanks for the heads up about serial-pass. Should be fixed now, I hope.
Wasn't sure where to put stuff for the Changelog. Let me know if the current version works for you.

Oh and about merging everything to serde. I am also fine with that. Just let me know and I will try to do the necessary changes.

thanks for the fixes! changelog looks great Yeah if you don't mind changing it again right away I'd say we should go with just serde, it makes all the cfgs so much simpler and reduces the number of feature flags and thus possible permutations by a lot.

Yeah, I will try that. On wgpu-types level it is trivial, but I am not sure how to approach it for wgpu-core and wgpu.
serial-pass is separate but should be part of serde. For that reason trace/replaycan not imply serde and I can not use serde in the source code.

Edit: I could replace feature = "serialize" with any(feature = "trace", feature = "serde") and feature = "deserialize" with any(feature = "replay", feature = "serde"). Is that what you had in mind?

Edit: I could replace feature = "serialize" with any(feature = "trace", feature = "serde") and feature = "deserialize" with any(feature = "replay", feature = "serde"). Is that what you had in mind?

I was thinking simpler still that both replay and trace would just imply the serde feature and all checks where we decide whether a thing implements the serde serialization/deserialization would only check for the serde feature

fyi I just posted about this also on the wgpu Matrix chat room just in case anyone has a good reason that we're not seeing right now on why deserialize & serialize should stay separate :)
https://matrix.to/#/!FZyQrssSlHEZqrYcOb:matrix.org/$DnwtH2xxPKoTq7OjVYEVoyUHzZMR3mVRIAvOhy3_GeI?via=matrix.org&via=mozilla.org&via=envs.net

Edit: I could replace feature = "serialize" with any(feature = "trace", feature = "serde") and feature = "deserialize" with any(feature = "replay", feature = "serde"). Is that what you had in mind?

I was thinking simpler still that both replay and trace would just imply the serde feature and all checks where we decide whether a thing implements the serde serialization/deserialization would only check for the serde feature

With your earlier comment about serial-pass this means that trace/replay implies serde which implies serial-pass. So trace/replay implies serial-pass. Is that fine?

Ah apologies for the confusion! If we go with serde for both serialize & deserialize then serial-pass should also go away I figure. Essentially we'd bulldozer everything over to only have serde, trace->serde, replay->serde.
Apparently, the original concern was compile time performance #619 (comment) but it looks like we don't have numbers for this and the setup with just serde implying both serialization and deserialization is much more common so I'll happily take all the blame (: