Fork me on GitHub
#graphql
<
2021-10-30
>
hlship18:10:48

Try out 1.1-alpha-5a and report back, if you can. I’m still looking for more optimizations. In some cases, the optimizations shift logic from the execution phase back to the parse phase … that’s really good if you can cache parsed queries. Still, it should be marginally better used as-is, and if you want to enable some of the opt-in optimizations, even better. Definitely shifted some processing and object allocation out of execution and into the parse (or schema compile) stages.

thumbnail12:11:30

Perf changes are looking good! Unfortunately we weren't able to test the latest release yet because some internal lib uses selector-context/is-wrapped-value?. Seems to be an easy fix though.

hlship21:11:05

Fortunately, that just moved to a new, different, and equally undocumented/unsupported internal namespace.

thumbnail21:11:17

Haha; yeah we have an helper which creates a Relay style node resolver which relied on it. Small change though 🙂. Will put in some effort today to migrate to 1.1-alpha-5a soon.

thumbnail17:11:27

Trying it out this week. I suspect there's a bug with returning tagged values with errors in subscriptions. Working on a repro; will report back when I can

thumbnail22:11:58

So I got to repro this, and the test starts failing since https://github.com/walmartlabs/lacinia/commit/f6371f3bf3fc93d9f3623b6352bf0933bf198dc7#diff-46dc32633a1117a5f120346c5f90504cc1944e0c5cc9904407a2f1d8e779a72f If i reverse the change to src/com/walmartlabs/lacinia/executor.clj the error-key in subscriptions works as before. I'll create a PR with a test and the 'fix' so you can take a look: https://github.com/walmartlabs/lacinia/pull/392