[v14.x-backport] node-api: rtn pending excep on napi_new_instance

[legendecas]

When there are any JavaScript exceptions pending,
napi_pending_exception should be returned.

PR-URL: #38798
Reviewed-By: Anna Henningsen anna@addaleax.net
Reviewed-By: Michael Dawson midawson@redhat.com

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/39260/

[addaleax]

So … this looks good, but:

When I reviewed the original PR, the reason I approved that this was a bugfix that aligned this call site with other call sites.

However, this prompted me to take another look at the source, it seems that CHECK_MAYBE_EMPTY most often uses napi_generic_failure – which seems pretty wrong? An empty maybe by definition indicates a pending exception.

[legendecas]

@addaleax previously we consider this returning status change as a possible ABI breaking change. However, the original PR does resolve a real problem, which we considered as a bug of node core. I believe it is time to re-evaluate if such a change is breaking anymore. If it is not been categorized as an ABI breaking change, we should apply the change on all other similar call sites.

/cc @nodejs/node-api

[addaleax]

@legendecas Yeah, I think this is either an all-or-nothing situation.

However, the original PR does resolve a real problem

But the same goes for the other call sites…?

[mhdawson]

LGTM

[targos]

Can you please rebase?

[targos]

ping @legendecas

[legendecas]

@targos updated! :)

[targos]

Landed in a74032a