| Jdforrester-WMF | |
| Mar 7 2024, 4:44 PM |
| F94091792: image.png | |
| Jul 16 2026, 11:32 PM |
| F94091609: image.png | |
| Jul 16 2026, 11:32 PM |
Evaluator-Orchestrator callbacks is the ability for code implementations, which are executed by the Evaluator, to make function calls back to the Orchestrator as part of their execution. Previously, we've use the words "reentrant" and "reentrancy" to describe this ability (and they still may appear in some places), but we've decided not to use those words anymore.
This is an umbrella ticket to collect all tasks related to this capability.
I don't think this should be marked as a duplicate of T412952, which is a time-bound piece of work that might or might not accomplish all of this Epic?
Good point, @Jdforrester-WMF ; thanks. I interpreted this ticket according to its brief list of subtasks, which were either done or may be done as part of T412952. In other words, I didn't interpret this ticket to be all-encompassing with respect to supporting reentrancy. Either way, you are right it's not accurate to call these 2 tickets duplicates. I will re-open to remove the duplicate connection and also because there is still an open subtask.
Indeed it should. Thanks for calling this out. It refers to Evaluator-Orchestrator callbacks; that is, the ability for code implementations, which are executed by the Evaluator, to make function calls back to the Orchestrator as part of their execution. In fact, we've decided not to use "reentrant" and "reentrancy" anymore, so I'm changing the title of this ticket.
This is an umbrella ticket to collect all tasks related to this capability. Most of this work under here has been completed, but not yet deployed, primarily because we are awaiting completion of a Security review.
Perhaps deployment should be deferred if it has any chance of interacting with everything else? When the caching problems are fixed, we really need a few weeks of everything working.
As you might recall, we've been working on this for over a year precisely because folks keep writing ever-more-complex Functions that time out. The caching problems to which I think you're referring are mostly due to this (but not entirely). The key design of this feature is that the Wikifunctions community can choose to use it or not on a per-Function basis, so if it's unused it will have no effect on anything.
No, not really; Phabricator isn't externally-facing documentation, but a working venue. Instead, we should do a better job of highlighting the regular community updates of what's being worked on, which links to tasks like this with some context.
Yes, this is an important piece of work. I absolutely want it.
The caching problems to which I think you're referring are mostly due to this (but not entirely).
I don't think this is true. (I'm referring to rendering articles, as tracked in T430898.) Can you explain why you think so? I believe over 95% of failed responses on AW are not caused by long-running functions. The reason I can tell this is when we construct the article in the first place, each function call is added separately, and runs fine, returning a result. So, for example, the other day, we made this, with every single fragment working as expected:
The key design of this feature is that the Wikifunctions community can choose to use it or not on a per-Function basis, so if it's unused it will have no effect on anything.
Okay. If it cannot degrade anything existing, go for it. You're the ones giving the talk at WIkimania!
I do not think this is resolved since what I want is allow callback to run any function, not just some pre-defined one. (This also include running on-the-fly anonymous functions provided by function inputs.)
Anonymous function means https://www.wikifunctions.org/view/en/Z38546
Right. Only an initial phase of this work has been completed; This ticket will be kept open for possible further developments.
… but to be clear, we are extremely unlikely to work on this any time soon, given Security's concerns.
More info please. Does "this" mean anonymous functions, or does it mean any extension beyond the inital test set that is now available?