I think to make our case work, we'd need to get rid of pulumi.runtime's shared state, and use a local variable that gets passed to all calls to Pulumi. Which is not a small job... So I'll have a look at running Mocha once per .spec.ts file, and having a completely new Pulumi engine each time.

It looks like the actual problem is that the object passed to

setMocks

is visible across all test files and each call to

setMocks

blats whatever was already in there. For example, if in a.spec.ts I call

pulumi.runtime.setMocks({ newResource: new_resource_a; call: call_a; })

and in b.spec.ts I do same except with

new_resource_b

and

call_b

, then the tests in a.spec.ts see the "correct" resources but they've been processed by

new_resource_b

and

call_b

. I need to scope Pulumi (or at least its runtime) to a single file, since the implementations of those faking functions are contradictory and can't be merged.

I think I need to scope the top-level variable

options

in runtime/settings.ts. I'm a JS/TS beginner, I don't know what to search for to figure out how to achieve that 😞

Ok got it sorted for now, though there is a minor risk of race conditions if run in parallel. I've moved setMocks to before(), instead of just at the top of the class. For now I'm running my tests serially and it's all good.