DefaultCheck with context.Context

[iwahbe]

Following up on #252, this allows DefaultCheck to safely and accurately handle secrets application. The new signature for DefaultCheck is easier to extend without additional breaking changes. The downside to this design is that this makes DefaultCheck special. I think it's worth living with that until #212 is resolved.

I'd like to merge shortly after #252 (and before #252 is released) so that user's don't rely on #252 injecting secrets when CustomCheck is implemented and DefaultCheck is not called.

[mjeffryes]

I'm not sure I follow the motivation. Why do we want authors to apply secrets themselves rather than relying on the framework to apply them?

[iwahbe]

I'm not sure I follow the motivation. Why do we want authors to apply secrets themselves rather than relying on the framework to apply them?

I added this to the commit message:

Ideally, most users will not need to implement check. By design, not implementing
CustomCheck has the same behavior as implementing check with
return infer.DefaultCheck(...). If users do implement CustomCheck, then we don't want
to make any assumptions about what behavior they want. This includes not applying defaults
and not injecting secrets.

[EronWright]

It is a breaking change (to DefaultCheck), right? Is that a concern?

The technique of using the context to pass information back up, I've not seen elsewhere. Maybe could use more commentary, like why the default check's encoder is so precious.

[iwahbe]

It is a breaking change (to DefaultCheck), right? Is that a concern?

It is. I think its a positive one though. This library is still in v0 and we tell users to expect breaking changes.

The technique of using the context to pass information back up, I've not seen elsewhere. Maybe could use more commentary, like why the default check's encoder is so precious.

ende.Encoders are the byproduct of going from resource.PropertyMap to C where C is some user supplied type. They carry a look-aside table of what is expressible with resource.PropertyMap but not expressible with C, such as secrets and unknown values. I'll add a comment to ende.Encoder to explain why it exists.

[EronWright]

Approving because it seems fine.

[pulumi-bot]

This PR has been shipped in release v0.21.0.