Async-await status report
1 March 2019
I wanted to post a quick update on the status of the async-await effort. The short version is that we’re in the home stretch for some kind of stabilization, but there remain some significant questions to overcome.
Announcing the implementation working group
As part of this push, I’m happy to announce we’ve formed a async-await implementation working group. This working group is part of the whole async-await effort, but focused on the implementation, and is part of the compiler team. If you’d like to help get async-await over the finish line, we’ve got a list of issues where we’d definitely like help (read on).
If you are interested in taking part, we have an “office hours” scheduled for Tuesday (see the compiler team calendar) – if you can show up then on Zulip, it’d be ideal! (But if not, just pop in any time.)
Who are we stabilizing for?
I mentioned that there remain significant questions to overcome before stabilization. I think the most root question of all is this one: Who is the audience for this stabilization?
The reason that question is so important is because it determines how
to weigh some of the issues that currently exist. If the point of the
stabilization is to start promoting async-await as something for
widespread use, then there are issues that we probably ought to
resolve first – most notably, the await
syntax, but also other
things.
If, however, the point of stabilization is to let ’early adopters’ start playing with it more, then we might be more tolerant of problems, so long as there are no backwards compatibility concerns.
My take is that either of these is a perfectly fine answer. But if the answer is that we are trying to unblock early adopters, then we want to be clear in our messaging, so that people don’t get turned off when they encounter some of the bugs below.
OK, with that in place, let’s look in a bit more detail.
Implementation issues
One of the first things that we did in setting up the implementation working group is to do a complete triage of all existing async-await issues. From this, we found that there was one very firm blocker, #54716. This issue has to do the timing of drops in an async fn, specifically the drop order for parameters that are not used in the fn body. We want to be sure this behaves analogously with regular functions. This is a blocker to stabilization because it would change the semantics of stable code for us to fix it later.
We also uncovered a number of major ergonomic problems. In a
follow-up meeting (available on YouTube), cramertj and I
also drew up plans for fixing these bugs, though these plans have
not yet been writting into mentoring instructions. These issues
include all focus around async fns that take borrowed references as
arguments – for example, the async fn syntax today doesn’t support
more than one lifetime in the
arguments, so
something like async fn foo(x: &u32, y: &u32)
doesn’t work.
Whether these ergonomic problems are blockers, however, depends a bit on your perspective: as @cramertj says, a number of folks at Google are using async-await today productively despite these limitations, but you must know the appropriate workarounds and so forth. This is where the question of our audience comes into play. My take is that these issues are blockers for “async fn” being ready for “general use”, but probably not for “early adopters”.
Another big concern for me personally is the maintenance story. Thanks to the hard work of Zoxc and cramertj, we’ve been able to standup a functional async-await implementation very fast, which is awesome. But we don’t really have a large pool of active contributors working on the async-await implementation who can help to fix issues as we find them, and this seems bad.
The syntax question
Finally, we come to the question of the await
syntax. At the All
Hands, we had a number of conversations on this topic, and it became
clear that we do not presently have consensus for any one syntax.
We did a lot of exploration here, however, and enumerated a number
of subtle arguments in favor of each option. At this moment,
@withoutboats is busily trying to write-up that exploration into a
document.
Before saying anything else, it’s worth pointing out that we don’t
actually have to resolve the await
syntax in order to stabilize
async-await. We could stabilize the await!(...)
macro syntax for the
time being, and return to the issue later. This would unblock “early
adopters”, but doesn’t seem like a satisfying answer if our target is
the “general public”. If we were to do this, we’d be drawing on the
precedent of try!
, where we first adopted a macro and later moved
that support to native syntax.
That said, we do eventually want to pick another syntax, so it’s worth thinking about how we are going to do that. As I wrote, the first step is to complete an overall summary that tries to describe the options on the table and some of the criteria that we can use to choose between them. Once that is available, we will need to settle on next steps.
Resolving hard questions
I am looking at the syntax question as a kind of opportunity – one of the things that we as a community frequently have to do is to find a way to resolve really hard questions without a clear answer. The tools that we have for doing this at the moment are really fairly crude: we use discussion threads and manual summary comments. Sometimes, this works well. Sometimes, amazingly well. But other times, it can be a real drain.
I would like to see us trying to resolve this sort of issue in other ways. I’ll be honest and say that I don’t entirely know what those are, but I know they are not open discussion threads. For example, I’ve found that the #rust2019 blog posts have been an incredibly effective way to have an open conversation about priorities without the usual ranchor and back-and-forth. I’ve been very inspired by systems like vTaiwan, which enable a lot of public input, but in a structured and collaborative form, rather than an “antagonistic” one. Similarly, I would like to see us perhaps consider running more experiments to test hypotheses about learnability or other factors (but this is something I would approach with great caution, as I think designing good experiments is very hard).
Anyway, this is really a topic for a post of its own. In this particular case, I hope that we find that enumerating in detail the arguments for each side leads us to a clear conclusion, perhaps some kind of “third way” that we haven’t seen yet. But, thinking ahead, it’d be nice to find ways to have these conversations that take us to that “third way” faster.
Closing notes
As someone who has not been closely following async-await thus far, I’m super excited by all I see. The feature has come a ridiculously long way, and the remaining blockers all seem like things we can overcome. async await is coming: I can’t wait to see what people build with it.