hydro_lang/live_collections/keyed_singleton.rs
1//! Definitions for the [`KeyedSingleton`] live collection.
2
3use std::cell::RefCell;
4use std::collections::HashMap;
5use std::hash::Hash;
6use std::marker::PhantomData;
7use std::ops::Deref;
8use std::rc::Rc;
9
10use stageleft::{IntoQuotedMut, QuotedWithContext, q};
11
12use super::boundedness::{Bounded, Boundedness, IsBounded, Unbounded};
13use super::keyed_stream::KeyedStream;
14use super::optional::Optional;
15use super::singleton::Singleton;
16use super::stream::{ExactlyOnce, NoOrder, Stream, TotalOrder};
17use crate::compile::builder::CycleId;
18use crate::compile::ir::{
19 CollectionKind, HydroIrOpMetadata, HydroNode, HydroRoot, KeyedSingletonBoundKind, TeeNode,
20};
21#[cfg(stageleft_runtime)]
22use crate::forward_handle::{CycleCollection, ReceiverComplete};
23use crate::forward_handle::{ForwardRef, TickCycle};
24use crate::live_collections::stream::{Ordering, Retries};
25#[cfg(stageleft_runtime)]
26use crate::location::dynamic::{DynLocation, LocationId};
27use crate::location::tick::DeferTick;
28use crate::location::{Atomic, Location, NoTick, Tick, check_matching_location};
29use crate::manual_expr::ManualExpr;
30use crate::nondet::{NonDet, nondet};
31use crate::properties::ManualProof;
32
33/// A marker trait indicating which components of a [`KeyedSingleton`] may change.
34///
35/// In addition to [`Bounded`] (all entries are fixed) and [`Unbounded`] (entries may be added /
36/// changed, but not removed), this also includes an additional variant [`BoundedValue`], which
37/// indicates that entries may be added over time, but once an entry is added it will never be
38/// removed and its value will never change.
39pub trait KeyedSingletonBound {
40 /// The [`Boundedness`] of the [`Stream`] underlying the keyed singleton.
41 type UnderlyingBound: Boundedness;
42 /// The [`Boundedness`] of each entry's value; [`Bounded`] means it is immutable.
43 type ValueBound: Boundedness;
44
45 /// The type of the keyed singleton if the value for each key is immutable.
46 type WithBoundedValue: KeyedSingletonBound<UnderlyingBound = Self::UnderlyingBound, ValueBound = Bounded>;
47
48 /// The type of the keyed singleton if the value for each key may change asynchronously.
49 type WithUnboundedValue: KeyedSingletonBound<UnderlyingBound = Self::UnderlyingBound, ValueBound = Unbounded>;
50
51 /// Returns the [`KeyedSingletonBoundKind`] corresponding to this type.
52 fn bound_kind() -> KeyedSingletonBoundKind;
53}
54
55impl KeyedSingletonBound for Unbounded {
56 type UnderlyingBound = Unbounded;
57 type ValueBound = Unbounded;
58 type WithBoundedValue = BoundedValue;
59 type WithUnboundedValue = Unbounded;
60
61 fn bound_kind() -> KeyedSingletonBoundKind {
62 KeyedSingletonBoundKind::Unbounded
63 }
64}
65
66impl KeyedSingletonBound for Bounded {
67 type UnderlyingBound = Bounded;
68 type ValueBound = Bounded;
69 type WithBoundedValue = Bounded;
70 type WithUnboundedValue = UnreachableBound;
71
72 fn bound_kind() -> KeyedSingletonBoundKind {
73 KeyedSingletonBoundKind::Bounded
74 }
75}
76
77/// A variation of boundedness specific to [`KeyedSingleton`], which indicates that once a key appears,
78/// its value is bounded and will never change. If the `KeyBound` is [`Bounded`], then the entire set of entries
79/// is bounded, but if it is [`Unbounded`], then new entries may appear asynchronously.
80pub struct BoundedValue;
81
82impl KeyedSingletonBound for BoundedValue {
83 type UnderlyingBound = Unbounded;
84 type ValueBound = Bounded;
85 type WithBoundedValue = BoundedValue;
86 type WithUnboundedValue = Unbounded;
87
88 fn bound_kind() -> KeyedSingletonBoundKind {
89 KeyedSingletonBoundKind::BoundedValue
90 }
91}
92
93#[doc(hidden)]
94pub struct UnreachableBound;
95
96impl KeyedSingletonBound for UnreachableBound {
97 type UnderlyingBound = Bounded;
98 type ValueBound = Unbounded;
99
100 type WithBoundedValue = Bounded;
101 type WithUnboundedValue = UnreachableBound;
102
103 fn bound_kind() -> KeyedSingletonBoundKind {
104 unreachable!("UnreachableBound cannot be instantiated")
105 }
106}
107
108/// Mapping from keys of type `K` to values of type `V`.
109///
110/// Keyed Singletons capture an asynchronously updated mapping from keys of the `K` to values of
111/// type `V`, where the order of keys is non-deterministic. In addition to the standard boundedness
112/// variants ([`Bounded`] for finite and immutable, [`Unbounded`] for asynchronously changing),
113/// keyed singletons can use [`BoundedValue`] to declare that new keys may be added over time, but
114/// keys cannot be removed and the value for each key is immutable.
115///
116/// Type Parameters:
117/// - `K`: the type of the key for each entry
118/// - `V`: the type of the value for each entry
119/// - `Loc`: the [`Location`] where the keyed singleton is materialized
120/// - `Bound`: tracks whether the entries are:
121/// - [`Bounded`] (local and finite)
122/// - [`Unbounded`] (asynchronous with entries added / removed / changed over time)
123/// - [`BoundedValue`] (asynchronous with immutable values for each key and no removals)
124pub struct KeyedSingleton<K, V, Loc, Bound: KeyedSingletonBound> {
125 pub(crate) location: Loc,
126 pub(crate) ir_node: RefCell<HydroNode>,
127
128 _phantom: PhantomData<(K, V, Loc, Bound)>,
129}
130
131impl<'a, K: Clone, V: Clone, Loc: Location<'a>, Bound: KeyedSingletonBound> Clone
132 for KeyedSingleton<K, V, Loc, Bound>
133{
134 fn clone(&self) -> Self {
135 if !matches!(self.ir_node.borrow().deref(), HydroNode::Tee { .. }) {
136 let orig_ir_node = self.ir_node.replace(HydroNode::Placeholder);
137 *self.ir_node.borrow_mut() = HydroNode::Tee {
138 inner: TeeNode(Rc::new(RefCell::new(orig_ir_node))),
139 metadata: self.location.new_node_metadata(Self::collection_kind()),
140 };
141 }
142
143 if let HydroNode::Tee { inner, metadata } = self.ir_node.borrow().deref() {
144 KeyedSingleton {
145 location: self.location.clone(),
146 ir_node: HydroNode::Tee {
147 inner: TeeNode(inner.0.clone()),
148 metadata: metadata.clone(),
149 }
150 .into(),
151 _phantom: PhantomData,
152 }
153 } else {
154 unreachable!()
155 }
156 }
157}
158
159impl<'a, K, V, L, B: KeyedSingletonBound> CycleCollection<'a, ForwardRef>
160 for KeyedSingleton<K, V, L, B>
161where
162 L: Location<'a> + NoTick,
163{
164 type Location = L;
165
166 fn create_source(cycle_id: CycleId, location: L) -> Self {
167 KeyedSingleton {
168 location: location.clone(),
169 ir_node: RefCell::new(HydroNode::CycleSource {
170 cycle_id,
171 metadata: location.new_node_metadata(Self::collection_kind()),
172 }),
173 _phantom: PhantomData,
174 }
175 }
176}
177
178impl<'a, K, V, L> CycleCollection<'a, TickCycle> for KeyedSingleton<K, V, Tick<L>, Bounded>
179where
180 L: Location<'a>,
181{
182 type Location = Tick<L>;
183
184 fn create_source(cycle_id: CycleId, location: Tick<L>) -> Self {
185 KeyedSingleton::new(
186 location.clone(),
187 HydroNode::CycleSource {
188 cycle_id,
189 metadata: location.new_node_metadata(Self::collection_kind()),
190 },
191 )
192 }
193}
194
195impl<'a, K, V, L> DeferTick for KeyedSingleton<K, V, Tick<L>, Bounded>
196where
197 L: Location<'a>,
198{
199 fn defer_tick(self) -> Self {
200 KeyedSingleton::defer_tick(self)
201 }
202}
203
204impl<'a, K, V, L, B: KeyedSingletonBound> ReceiverComplete<'a, ForwardRef>
205 for KeyedSingleton<K, V, L, B>
206where
207 L: Location<'a> + NoTick,
208{
209 fn complete(self, cycle_id: CycleId, expected_location: LocationId) {
210 assert_eq!(
211 Location::id(&self.location),
212 expected_location,
213 "locations do not match"
214 );
215 self.location
216 .flow_state()
217 .borrow_mut()
218 .push_root(HydroRoot::CycleSink {
219 cycle_id,
220 input: Box::new(self.ir_node.into_inner()),
221 op_metadata: HydroIrOpMetadata::new(),
222 });
223 }
224}
225
226impl<'a, K, V, L> ReceiverComplete<'a, TickCycle> for KeyedSingleton<K, V, Tick<L>, Bounded>
227where
228 L: Location<'a>,
229{
230 fn complete(self, cycle_id: CycleId, expected_location: LocationId) {
231 assert_eq!(
232 Location::id(&self.location),
233 expected_location,
234 "locations do not match"
235 );
236 self.location
237 .flow_state()
238 .borrow_mut()
239 .push_root(HydroRoot::CycleSink {
240 cycle_id,
241 input: Box::new(self.ir_node.into_inner()),
242 op_metadata: HydroIrOpMetadata::new(),
243 });
244 }
245}
246
247impl<'a, K, V, L: Location<'a>, B: KeyedSingletonBound> KeyedSingleton<K, V, L, B> {
248 pub(crate) fn new(location: L, ir_node: HydroNode) -> Self {
249 debug_assert_eq!(ir_node.metadata().location_id, Location::id(&location));
250 debug_assert_eq!(ir_node.metadata().collection_kind, Self::collection_kind());
251
252 KeyedSingleton {
253 location,
254 ir_node: RefCell::new(ir_node),
255 _phantom: PhantomData,
256 }
257 }
258
259 /// Returns the [`Location`] where this keyed singleton is being materialized.
260 pub fn location(&self) -> &L {
261 &self.location
262 }
263}
264
265#[cfg(stageleft_runtime)]
266fn key_count_inside_tick<'a, K, V, L: Location<'a>>(
267 me: KeyedSingleton<K, V, L, Bounded>,
268) -> Singleton<usize, L, Bounded> {
269 me.entries().count()
270}
271
272#[cfg(stageleft_runtime)]
273fn into_singleton_inside_tick<'a, K, V, L: Location<'a>>(
274 me: KeyedSingleton<K, V, L, Bounded>,
275) -> Singleton<HashMap<K, V>, L, Bounded>
276where
277 K: Eq + Hash,
278{
279 me.entries()
280 .assume_ordering(nondet!(
281 /// Because this is a keyed singleton, there is only one value per key.
282 ))
283 .fold(
284 q!(|| HashMap::new()),
285 q!(|map, (k, v)| {
286 map.insert(k, v);
287 }),
288 )
289}
290
291impl<'a, K, V, L: Location<'a>, B: KeyedSingletonBound> KeyedSingleton<K, V, L, B> {
292 pub(crate) fn collection_kind() -> CollectionKind {
293 CollectionKind::KeyedSingleton {
294 bound: B::bound_kind(),
295 key_type: stageleft::quote_type::<K>().into(),
296 value_type: stageleft::quote_type::<V>().into(),
297 }
298 }
299
300 /// Transforms each value by invoking `f` on each element, with keys staying the same
301 /// after transformation. If you need access to the key, see [`KeyedSingleton::map_with_key`].
302 ///
303 /// If you do not want to modify the stream and instead only want to view
304 /// each item use [`KeyedSingleton::inspect`] instead.
305 ///
306 /// # Example
307 /// ```rust
308 /// # #[cfg(feature = "deploy")] {
309 /// # use hydro_lang::prelude::*;
310 /// # use futures::StreamExt;
311 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
312 /// let keyed_singleton = // { 1: 2, 2: 4 }
313 /// # process
314 /// # .source_iter(q!(vec![(1, 2), (2, 4)]))
315 /// # .into_keyed()
316 /// # .first();
317 /// keyed_singleton.map(q!(|v| v + 1))
318 /// # .entries()
319 /// # }, |mut stream| async move {
320 /// // { 1: 3, 2: 5 }
321 /// # let mut results = Vec::new();
322 /// # for _ in 0..2 {
323 /// # results.push(stream.next().await.unwrap());
324 /// # }
325 /// # results.sort();
326 /// # assert_eq!(results, vec![(1, 3), (2, 5)]);
327 /// # }));
328 /// # }
329 /// ```
330 pub fn map<U, F>(self, f: impl IntoQuotedMut<'a, F, L> + Copy) -> KeyedSingleton<K, U, L, B>
331 where
332 F: Fn(V) -> U + 'a,
333 {
334 let f: ManualExpr<F, _> = ManualExpr::new(move |ctx: &L| f.splice_fn1_ctx(ctx));
335 let map_f = q!({
336 let orig = f;
337 move |(k, v)| (k, orig(v))
338 })
339 .splice_fn1_ctx::<(K, V), (K, U)>(&self.location)
340 .into();
341
342 KeyedSingleton::new(
343 self.location.clone(),
344 HydroNode::Map {
345 f: map_f,
346 input: Box::new(self.ir_node.into_inner()),
347 metadata: self
348 .location
349 .new_node_metadata(KeyedSingleton::<K, U, L, B>::collection_kind()),
350 },
351 )
352 }
353
354 /// Transforms each value by invoking `f` on each key-value pair, with keys staying the same
355 /// after transformation. Unlike [`KeyedSingleton::map`], this gives access to both the key and value.
356 ///
357 /// The closure `f` receives a tuple `(K, V)` containing both the key and value, and returns
358 /// the new value `U`. The key remains unchanged in the output.
359 ///
360 /// # Example
361 /// ```rust
362 /// # #[cfg(feature = "deploy")] {
363 /// # use hydro_lang::prelude::*;
364 /// # use futures::StreamExt;
365 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
366 /// let keyed_singleton = // { 1: 2, 2: 4 }
367 /// # process
368 /// # .source_iter(q!(vec![(1, 2), (2, 4)]))
369 /// # .into_keyed()
370 /// # .first();
371 /// keyed_singleton.map_with_key(q!(|(k, v)| k + v))
372 /// # .entries()
373 /// # }, |mut stream| async move {
374 /// // { 1: 3, 2: 6 }
375 /// # let mut results = Vec::new();
376 /// # for _ in 0..2 {
377 /// # results.push(stream.next().await.unwrap());
378 /// # }
379 /// # results.sort();
380 /// # assert_eq!(results, vec![(1, 3), (2, 6)]);
381 /// # }));
382 /// # }
383 /// ```
384 pub fn map_with_key<U, F>(
385 self,
386 f: impl IntoQuotedMut<'a, F, L> + Copy,
387 ) -> KeyedSingleton<K, U, L, B>
388 where
389 F: Fn((K, V)) -> U + 'a,
390 K: Clone,
391 {
392 let f: ManualExpr<F, _> = ManualExpr::new(move |ctx: &L| f.splice_fn1_ctx(ctx));
393 let map_f = q!({
394 let orig = f;
395 move |(k, v)| {
396 let out = orig((Clone::clone(&k), v));
397 (k, out)
398 }
399 })
400 .splice_fn1_ctx::<(K, V), (K, U)>(&self.location)
401 .into();
402
403 KeyedSingleton::new(
404 self.location.clone(),
405 HydroNode::Map {
406 f: map_f,
407 input: Box::new(self.ir_node.into_inner()),
408 metadata: self
409 .location
410 .new_node_metadata(KeyedSingleton::<K, U, L, B>::collection_kind()),
411 },
412 )
413 }
414
415 /// Gets the number of keys in the keyed singleton.
416 ///
417 /// The output singleton will be unbounded if the input is [`Unbounded`] or [`BoundedValue`],
418 /// since keys may be added / removed over time. When the set of keys changes, the count will
419 /// be asynchronously updated.
420 ///
421 /// # Example
422 /// ```rust
423 /// # #[cfg(feature = "deploy")] {
424 /// # use hydro_lang::prelude::*;
425 /// # use futures::StreamExt;
426 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
427 /// # let tick = process.tick();
428 /// let keyed_singleton = // { 1: "a", 2: "b", 3: "c" }
429 /// # process
430 /// # .source_iter(q!(vec![(1, "a"), (2, "b"), (3, "c")]))
431 /// # .into_keyed()
432 /// # .batch(&tick, nondet!(/** test */))
433 /// # .first();
434 /// keyed_singleton.key_count()
435 /// # .all_ticks()
436 /// # }, |mut stream| async move {
437 /// // 3
438 /// # assert_eq!(stream.next().await.unwrap(), 3);
439 /// # }));
440 /// # }
441 /// ```
442 pub fn key_count(self) -> Singleton<usize, L, B::UnderlyingBound> {
443 if B::ValueBound::BOUNDED {
444 let me: KeyedSingleton<K, V, L, B::WithBoundedValue> = KeyedSingleton {
445 location: self.location,
446 ir_node: self.ir_node,
447 _phantom: PhantomData,
448 };
449
450 me.entries().count()
451 } else if L::is_top_level()
452 && let Some(tick) = self.location.try_tick()
453 {
454 let me: KeyedSingleton<K, V, L, B::WithUnboundedValue> = KeyedSingleton {
455 location: self.location,
456 ir_node: self.ir_node,
457 _phantom: PhantomData,
458 };
459
460 let out =
461 key_count_inside_tick(me.snapshot(&tick, nondet!(/** eventually stabilizes */)))
462 .latest();
463 Singleton::new(out.location, out.ir_node.into_inner())
464 } else {
465 panic!("Unbounded KeyedSingleton inside a tick");
466 }
467 }
468
469 /// Converts this keyed singleton into a [`Singleton`] containing a `HashMap` from keys to values.
470 ///
471 /// As the values for each key are updated asynchronously, the `HashMap` will be updated
472 /// asynchronously as well.
473 ///
474 /// # Example
475 /// ```rust
476 /// # #[cfg(feature = "deploy")] {
477 /// # use hydro_lang::prelude::*;
478 /// # use futures::StreamExt;
479 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
480 /// let keyed_singleton = // { 1: "a", 2: "b", 3: "c" }
481 /// # process
482 /// # .source_iter(q!(vec![(1, "a".to_owned()), (2, "b".to_owned()), (3, "c".to_owned())]))
483 /// # .into_keyed()
484 /// # .batch(&process.tick(), nondet!(/** test */))
485 /// # .first();
486 /// keyed_singleton.into_singleton()
487 /// # .all_ticks()
488 /// # }, |mut stream| async move {
489 /// // { 1: "a", 2: "b", 3: "c" }
490 /// # assert_eq!(stream.next().await.unwrap(), vec![(1, "a".to_owned()), (2, "b".to_owned()), (3, "c".to_owned())].into_iter().collect());
491 /// # }));
492 /// # }
493 /// ```
494 pub fn into_singleton(self) -> Singleton<HashMap<K, V>, L, B::UnderlyingBound>
495 where
496 K: Eq + Hash,
497 {
498 if B::ValueBound::BOUNDED {
499 let me: KeyedSingleton<K, V, L, B::WithBoundedValue> = KeyedSingleton {
500 location: self.location,
501 ir_node: self.ir_node,
502 _phantom: PhantomData,
503 };
504
505 me.entries()
506 .assume_ordering(nondet!(
507 /// Because this is a keyed singleton, there is only one value per key.
508 ))
509 .fold(
510 q!(|| HashMap::new()),
511 q!(|map, (k, v)| {
512 // TODO(shadaj): make this commutative but really-debug-assert that there is no key overlap
513 map.insert(k, v);
514 }),
515 )
516 } else if L::is_top_level()
517 && let Some(tick) = self.location.try_tick()
518 {
519 let me: KeyedSingleton<K, V, L, B::WithUnboundedValue> = KeyedSingleton {
520 location: self.location,
521 ir_node: self.ir_node,
522 _phantom: PhantomData,
523 };
524
525 let out = into_singleton_inside_tick(
526 me.snapshot(&tick, nondet!(/** eventually stabilizes */)),
527 )
528 .latest();
529 Singleton::new(out.location, out.ir_node.into_inner())
530 } else {
531 panic!("Unbounded KeyedSingleton inside a tick");
532 }
533 }
534
535 /// An operator which allows you to "name" a `HydroNode`.
536 /// This is only used for testing, to correlate certain `HydroNode`s with IDs.
537 pub fn ir_node_named(self, name: &str) -> KeyedSingleton<K, V, L, B> {
538 {
539 let mut node = self.ir_node.borrow_mut();
540 let metadata = node.metadata_mut();
541 metadata.tag = Some(name.to_owned());
542 }
543 self
544 }
545
546 /// Strengthens the boundedness guarantee to `Bounded`, given that `B: IsBounded`, which
547 /// implies that `B == Bounded`.
548 pub fn make_bounded(self) -> KeyedSingleton<K, V, L, Bounded>
549 where
550 B: IsBounded,
551 {
552 KeyedSingleton::new(self.location, self.ir_node.into_inner())
553 }
554
555 /// Gets the value associated with a specific key from the keyed singleton.
556 ///
557 /// # Example
558 /// ```rust
559 /// # #[cfg(feature = "deploy")] {
560 /// # use hydro_lang::prelude::*;
561 /// # use futures::StreamExt;
562 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
563 /// let tick = process.tick();
564 /// let keyed_data = process
565 /// .source_iter(q!(vec![(1, 2), (2, 3)]))
566 /// .into_keyed()
567 /// .batch(&tick, nondet!(/** test */))
568 /// .first();
569 /// let key = tick.singleton(q!(1));
570 /// keyed_data.get(key).all_ticks()
571 /// # }, |mut stream| async move {
572 /// // 2
573 /// # assert_eq!(stream.next().await.unwrap(), 2);
574 /// # }));
575 /// # }
576 /// ```
577 pub fn get(self, key: Singleton<K, L, Bounded>) -> Optional<V, L, Bounded>
578 where
579 B: IsBounded,
580 K: Hash + Eq,
581 {
582 self.make_bounded()
583 .into_keyed_stream()
584 .get(key)
585 .assume_ordering::<TotalOrder>(nondet!(/** only a single key, so totally ordered */))
586 .first()
587 }
588
589 /// Emit a keyed stream containing keys shared between the keyed singleton and the
590 /// keyed stream, where each value in the output keyed stream is a tuple of
591 /// (the keyed singleton's value, the keyed stream's value).
592 ///
593 /// # Example
594 /// ```rust
595 /// # #[cfg(feature = "deploy")] {
596 /// # use hydro_lang::prelude::*;
597 /// # use futures::StreamExt;
598 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
599 /// let tick = process.tick();
600 /// let keyed_data = process
601 /// .source_iter(q!(vec![(1, 10), (2, 20)]))
602 /// .into_keyed()
603 /// .batch(&tick, nondet!(/** test */))
604 /// .first();
605 /// let other_data = process
606 /// .source_iter(q!(vec![(1, 100), (2, 200), (1, 101)]))
607 /// .into_keyed()
608 /// .batch(&tick, nondet!(/** test */));
609 /// keyed_data.join_keyed_stream(other_data).entries().all_ticks()
610 /// # }, |mut stream| async move {
611 /// // { 1: [(10, 100), (10, 101)], 2: [(20, 200)] } in any order
612 /// # let mut results = vec![];
613 /// # for _ in 0..3 {
614 /// # results.push(stream.next().await.unwrap());
615 /// # }
616 /// # results.sort();
617 /// # assert_eq!(results, vec![(1, (10, 100)), (1, (10, 101)), (2, (20, 200))]);
618 /// # }));
619 /// # }
620 /// ```
621 pub fn join_keyed_stream<O2: Ordering, R2: Retries, V2>(
622 self,
623 keyed_stream: KeyedStream<K, V2, L, Bounded, O2, R2>,
624 ) -> KeyedStream<K, (V, V2), L, Bounded, NoOrder, R2>
625 where
626 B: IsBounded,
627 K: Eq + Hash,
628 {
629 self.make_bounded()
630 .entries()
631 .weaken_retries::<R2>()
632 .join(keyed_stream.entries())
633 .into_keyed()
634 }
635
636 /// Emit a keyed singleton containing all keys shared between two keyed singletons,
637 /// where each value in the output keyed singleton is a tuple of
638 /// (self.value, other.value).
639 ///
640 /// # Example
641 /// ```rust
642 /// # #[cfg(feature = "deploy")] {
643 /// # use hydro_lang::prelude::*;
644 /// # use futures::StreamExt;
645 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
646 /// # let tick = process.tick();
647 /// let requests = // { 1: 10, 2: 20, 3: 30 }
648 /// # process
649 /// # .source_iter(q!(vec![(1, 10), (2, 20), (3, 30)]))
650 /// # .into_keyed()
651 /// # .batch(&tick, nondet!(/** test */))
652 /// # .first();
653 /// let other = // { 1: 100, 2: 200, 4: 400 }
654 /// # process
655 /// # .source_iter(q!(vec![(1, 100), (2, 200), (4, 400)]))
656 /// # .into_keyed()
657 /// # .batch(&tick, nondet!(/** test */))
658 /// # .first();
659 /// requests.join_keyed_singleton(other)
660 /// # .entries().all_ticks()
661 /// # }, |mut stream| async move {
662 /// // { 1: (10, 100), 2: (20, 200) }
663 /// # let mut results = vec![];
664 /// # for _ in 0..2 {
665 /// # results.push(stream.next().await.unwrap());
666 /// # }
667 /// # results.sort();
668 /// # assert_eq!(results, vec![(1, (10, 100)), (2, (20, 200))]);
669 /// # }));
670 /// # }
671 /// ```
672 pub fn join_keyed_singleton<V2: Clone>(
673 self,
674 other: KeyedSingleton<K, V2, L, Bounded>,
675 ) -> KeyedSingleton<K, (V, V2), L, Bounded>
676 where
677 B: IsBounded,
678 K: Eq + Hash,
679 {
680 let result_stream = self
681 .make_bounded()
682 .entries()
683 .join(other.entries())
684 .into_keyed();
685
686 // The cast is guaranteed to succeed, since each key (in both `self` and `other`) has at most one value.
687 KeyedSingleton::new(
688 result_stream.location.clone(),
689 HydroNode::Cast {
690 inner: Box::new(result_stream.ir_node.into_inner()),
691 metadata: result_stream.location.new_node_metadata(KeyedSingleton::<
692 K,
693 (V, V2),
694 L,
695 Bounded,
696 >::collection_kind(
697 )),
698 },
699 )
700 }
701
702 /// For each value in `self`, find the matching key in `lookup`.
703 /// The output is a keyed singleton with the key from `self`, and a value
704 /// that is a tuple of (`self`'s value, Option<`lookup`'s value>).
705 /// If the key is not present in `lookup`, the option will be [`None`].
706 ///
707 /// # Example
708 /// ```rust
709 /// # #[cfg(feature = "deploy")] {
710 /// # use hydro_lang::prelude::*;
711 /// # use futures::StreamExt;
712 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
713 /// # let tick = process.tick();
714 /// let requests = // { 1: 10, 2: 20 }
715 /// # process
716 /// # .source_iter(q!(vec![(1, 10), (2, 20)]))
717 /// # .into_keyed()
718 /// # .batch(&tick, nondet!(/** test */))
719 /// # .first();
720 /// let other_data = // { 10: 100, 11: 110 }
721 /// # process
722 /// # .source_iter(q!(vec![(10, 100), (11, 110)]))
723 /// # .into_keyed()
724 /// # .batch(&tick, nondet!(/** test */))
725 /// # .first();
726 /// requests.lookup_keyed_singleton(other_data)
727 /// # .entries().all_ticks()
728 /// # }, |mut stream| async move {
729 /// // { 1: (10, Some(100)), 2: (20, None) }
730 /// # let mut results = vec![];
731 /// # for _ in 0..2 {
732 /// # results.push(stream.next().await.unwrap());
733 /// # }
734 /// # results.sort();
735 /// # assert_eq!(results, vec![(1, (10, Some(100))), (2, (20, None))]);
736 /// # }));
737 /// # }
738 /// ```
739 pub fn lookup_keyed_singleton<V2>(
740 self,
741 lookup: KeyedSingleton<V, V2, L, Bounded>,
742 ) -> KeyedSingleton<K, (V, Option<V2>), L, Bounded>
743 where
744 B: IsBounded,
745 K: Eq + Hash + Clone,
746 V: Eq + Hash + Clone,
747 V2: Clone,
748 {
749 let result_stream = self
750 .make_bounded()
751 .into_keyed_stream()
752 .lookup_keyed_stream(lookup.into_keyed_stream());
753
754 // The cast is guaranteed to succeed since both lookup and self contain at most 1 value per key
755 KeyedSingleton::new(
756 result_stream.location.clone(),
757 HydroNode::Cast {
758 inner: Box::new(result_stream.ir_node.into_inner()),
759 metadata: result_stream.location.new_node_metadata(KeyedSingleton::<
760 K,
761 (V, Option<V2>),
762 L,
763 Bounded,
764 >::collection_kind(
765 )),
766 },
767 )
768 }
769
770 /// For each value in `self`, find the matching key in `lookup`.
771 /// The output is a keyed stream with the key from `self`, and a value
772 /// that is a tuple of (`self`'s value, Option<`lookup`'s value>).
773 /// If the key is not present in `lookup`, the option will be [`None`].
774 ///
775 /// # Example
776 /// ```rust
777 /// # #[cfg(feature = "deploy")] {
778 /// # use hydro_lang::prelude::*;
779 /// # use futures::StreamExt;
780 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
781 /// # let tick = process.tick();
782 /// let requests = // { 1: 10, 2: 20 }
783 /// # process
784 /// # .source_iter(q!(vec![(1, 10), (2, 20)]))
785 /// # .into_keyed()
786 /// # .batch(&tick, nondet!(/** test */))
787 /// # .first();
788 /// let other_data = // { 10: 100, 10: 110 }
789 /// # process
790 /// # .source_iter(q!(vec![(10, 100), (10, 110)]))
791 /// # .into_keyed()
792 /// # .batch(&tick, nondet!(/** test */));
793 /// requests.lookup_keyed_stream(other_data)
794 /// # .entries().all_ticks()
795 /// # }, |mut stream| async move {
796 /// // { 1: [(10, Some(100)), (10, Some(110))], 2: (20, None) }
797 /// # let mut results = vec![];
798 /// # for _ in 0..3 {
799 /// # results.push(stream.next().await.unwrap());
800 /// # }
801 /// # results.sort();
802 /// # assert_eq!(results, vec![(1, (10, Some(100))), (1, (10, Some(110))), (2, (20, None))]);
803 /// # }));
804 /// # }
805 /// ```
806 pub fn lookup_keyed_stream<V2, O: Ordering, R: Retries>(
807 self,
808 lookup: KeyedStream<V, V2, L, Bounded, O, R>,
809 ) -> KeyedStream<K, (V, Option<V2>), L, Bounded, NoOrder, R>
810 where
811 B: IsBounded,
812 K: Eq + Hash + Clone,
813 V: Eq + Hash + Clone,
814 V2: Clone,
815 {
816 self.make_bounded()
817 .entries()
818 .weaken_retries::<R>() // TODO: Once weaken_retries() is implemented for KeyedSingleton, remove entries() and into_keyed()
819 .into_keyed()
820 .lookup_keyed_stream(lookup)
821 }
822}
823
824impl<'a, K, V, L: Location<'a>, B: KeyedSingletonBound<ValueBound = Bounded>>
825 KeyedSingleton<K, V, L, B>
826{
827 /// Flattens the keyed singleton into an unordered stream of key-value pairs.
828 ///
829 /// The value for each key must be bounded, otherwise the resulting stream elements would be
830 /// non-deterministic. As new entries are added to the keyed singleton, they will be streamed
831 /// into the output.
832 ///
833 /// # Example
834 /// ```rust
835 /// # #[cfg(feature = "deploy")] {
836 /// # use hydro_lang::prelude::*;
837 /// # use futures::StreamExt;
838 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
839 /// let keyed_singleton = // { 1: 2, 2: 4 }
840 /// # process
841 /// # .source_iter(q!(vec![(1, 2), (2, 4)]))
842 /// # .into_keyed()
843 /// # .first();
844 /// keyed_singleton.entries()
845 /// # }, |mut stream| async move {
846 /// // (1, 2), (2, 4) in any order
847 /// # let mut results = Vec::new();
848 /// # for _ in 0..2 {
849 /// # results.push(stream.next().await.unwrap());
850 /// # }
851 /// # results.sort();
852 /// # assert_eq!(results, vec![(1, 2), (2, 4)]);
853 /// # }));
854 /// # }
855 /// ```
856 pub fn entries(self) -> Stream<(K, V), L, B::UnderlyingBound, NoOrder, ExactlyOnce> {
857 self.into_keyed_stream().entries()
858 }
859
860 /// Flattens the keyed singleton into an unordered stream of just the values.
861 ///
862 /// The value for each key must be bounded, otherwise the resulting stream elements would be
863 /// non-deterministic. As new entries are added to the keyed singleton, they will be streamed
864 /// into the output.
865 ///
866 /// # Example
867 /// ```rust
868 /// # #[cfg(feature = "deploy")] {
869 /// # use hydro_lang::prelude::*;
870 /// # use futures::StreamExt;
871 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
872 /// let keyed_singleton = // { 1: 2, 2: 4 }
873 /// # process
874 /// # .source_iter(q!(vec![(1, 2), (2, 4)]))
875 /// # .into_keyed()
876 /// # .first();
877 /// keyed_singleton.values()
878 /// # }, |mut stream| async move {
879 /// // 2, 4 in any order
880 /// # let mut results = Vec::new();
881 /// # for _ in 0..2 {
882 /// # results.push(stream.next().await.unwrap());
883 /// # }
884 /// # results.sort();
885 /// # assert_eq!(results, vec![2, 4]);
886 /// # }));
887 /// # }
888 /// ```
889 pub fn values(self) -> Stream<V, L, B::UnderlyingBound, NoOrder, ExactlyOnce> {
890 let map_f = q!(|(_, v)| v)
891 .splice_fn1_ctx::<(K, V), V>(&self.location)
892 .into();
893
894 Stream::new(
895 self.location.clone(),
896 HydroNode::Map {
897 f: map_f,
898 input: Box::new(self.ir_node.into_inner()),
899 metadata: self.location.new_node_metadata(Stream::<
900 V,
901 L,
902 B::UnderlyingBound,
903 NoOrder,
904 ExactlyOnce,
905 >::collection_kind()),
906 },
907 )
908 }
909
910 /// Flattens the keyed singleton into an unordered stream of just the keys.
911 ///
912 /// The value for each key must be bounded, otherwise the removal of keys would result in
913 /// non-determinism. As new entries are added to the keyed singleton, they will be streamed
914 /// into the output.
915 ///
916 /// # Example
917 /// ```rust
918 /// # #[cfg(feature = "deploy")] {
919 /// # use hydro_lang::prelude::*;
920 /// # use futures::StreamExt;
921 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
922 /// let keyed_singleton = // { 1: 2, 2: 4 }
923 /// # process
924 /// # .source_iter(q!(vec![(1, 2), (2, 4)]))
925 /// # .into_keyed()
926 /// # .first();
927 /// keyed_singleton.keys()
928 /// # }, |mut stream| async move {
929 /// // 1, 2 in any order
930 /// # let mut results = Vec::new();
931 /// # for _ in 0..2 {
932 /// # results.push(stream.next().await.unwrap());
933 /// # }
934 /// # results.sort();
935 /// # assert_eq!(results, vec![1, 2]);
936 /// # }));
937 /// # }
938 /// ```
939 pub fn keys(self) -> Stream<K, L, B::UnderlyingBound, NoOrder, ExactlyOnce> {
940 self.entries().map(q!(|(k, _)| k))
941 }
942
943 /// Given a bounded stream of keys `K`, returns a new keyed singleton containing only the
944 /// entries whose keys are not in the provided stream.
945 ///
946 /// # Example
947 /// ```rust
948 /// # #[cfg(feature = "deploy")] {
949 /// # use hydro_lang::prelude::*;
950 /// # use futures::StreamExt;
951 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
952 /// let tick = process.tick();
953 /// let keyed_singleton = // { 1: 2, 2: 4 }
954 /// # process
955 /// # .source_iter(q!(vec![(1, 2), (2, 4)]))
956 /// # .into_keyed()
957 /// # .first()
958 /// # .batch(&tick, nondet!(/** test */));
959 /// let keys_to_remove = process
960 /// .source_iter(q!(vec![1]))
961 /// .batch(&tick, nondet!(/** test */));
962 /// keyed_singleton.filter_key_not_in(keys_to_remove)
963 /// # .entries().all_ticks()
964 /// # }, |mut stream| async move {
965 /// // { 2: 4 }
966 /// # for w in vec![(2, 4)] {
967 /// # assert_eq!(stream.next().await.unwrap(), w);
968 /// # }
969 /// # }));
970 /// # }
971 /// ```
972 pub fn filter_key_not_in<O2: Ordering, R2: Retries>(
973 self,
974 other: Stream<K, L, Bounded, O2, R2>,
975 ) -> Self
976 where
977 K: Hash + Eq,
978 {
979 check_matching_location(&self.location, &other.location);
980
981 KeyedSingleton::new(
982 self.location.clone(),
983 HydroNode::AntiJoin {
984 pos: Box::new(self.ir_node.into_inner()),
985 neg: Box::new(other.ir_node.into_inner()),
986 metadata: self.location.new_node_metadata(Self::collection_kind()),
987 },
988 )
989 }
990
991 /// An operator which allows you to "inspect" each value of a keyed singleton without
992 /// modifying it. The closure `f` is called on a reference to each value. This is
993 /// mainly useful for debugging, and should not be used to generate side-effects.
994 ///
995 /// # Example
996 /// ```rust
997 /// # #[cfg(feature = "deploy")] {
998 /// # use hydro_lang::prelude::*;
999 /// # use futures::StreamExt;
1000 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
1001 /// let keyed_singleton = // { 1: 2, 2: 4 }
1002 /// # process
1003 /// # .source_iter(q!(vec![(1, 2), (2, 4)]))
1004 /// # .into_keyed()
1005 /// # .first();
1006 /// keyed_singleton
1007 /// .inspect(q!(|v| println!("{}", v)))
1008 /// # .entries()
1009 /// # }, |mut stream| async move {
1010 /// // { 1: 2, 2: 4 }
1011 /// # for w in vec![(1, 2), (2, 4)] {
1012 /// # assert_eq!(stream.next().await.unwrap(), w);
1013 /// # }
1014 /// # }));
1015 /// # }
1016 /// ```
1017 pub fn inspect<F>(self, f: impl IntoQuotedMut<'a, F, L> + Copy) -> Self
1018 where
1019 F: Fn(&V) + 'a,
1020 {
1021 let f: ManualExpr<F, _> = ManualExpr::new(move |ctx: &L| f.splice_fn1_borrow_ctx(ctx));
1022 let inspect_f = q!({
1023 let orig = f;
1024 move |t: &(_, _)| orig(&t.1)
1025 })
1026 .splice_fn1_borrow_ctx::<(K, V), ()>(&self.location)
1027 .into();
1028
1029 KeyedSingleton::new(
1030 self.location.clone(),
1031 HydroNode::Inspect {
1032 f: inspect_f,
1033 input: Box::new(self.ir_node.into_inner()),
1034 metadata: self.location.new_node_metadata(Self::collection_kind()),
1035 },
1036 )
1037 }
1038
1039 /// An operator which allows you to "inspect" each entry of a keyed singleton without
1040 /// modifying it. The closure `f` is called on a reference to each key-value pair. This is
1041 /// mainly useful for debugging, and should not be used to generate side-effects.
1042 ///
1043 /// # Example
1044 /// ```rust
1045 /// # #[cfg(feature = "deploy")] {
1046 /// # use hydro_lang::prelude::*;
1047 /// # use futures::StreamExt;
1048 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
1049 /// let keyed_singleton = // { 1: 2, 2: 4 }
1050 /// # process
1051 /// # .source_iter(q!(vec![(1, 2), (2, 4)]))
1052 /// # .into_keyed()
1053 /// # .first();
1054 /// keyed_singleton
1055 /// .inspect_with_key(q!(|(k, v)| println!("{}: {}", k, v)))
1056 /// # .entries()
1057 /// # }, |mut stream| async move {
1058 /// // { 1: 2, 2: 4 }
1059 /// # for w in vec![(1, 2), (2, 4)] {
1060 /// # assert_eq!(stream.next().await.unwrap(), w);
1061 /// # }
1062 /// # }));
1063 /// # }
1064 /// ```
1065 pub fn inspect_with_key<F>(self, f: impl IntoQuotedMut<'a, F, L>) -> Self
1066 where
1067 F: Fn(&(K, V)) + 'a,
1068 {
1069 let inspect_f = f.splice_fn1_borrow_ctx::<(K, V), ()>(&self.location).into();
1070
1071 KeyedSingleton::new(
1072 self.location.clone(),
1073 HydroNode::Inspect {
1074 f: inspect_f,
1075 input: Box::new(self.ir_node.into_inner()),
1076 metadata: self.location.new_node_metadata(Self::collection_kind()),
1077 },
1078 )
1079 }
1080
1081 /// Gets the key-value tuple with the largest key among all entries in this [`KeyedSingleton`].
1082 ///
1083 /// Because this method requires values to be bounded, the output [`Optional`] will only be
1084 /// asynchronously updated if a new key is added that is higher than the previous max key.
1085 ///
1086 /// # Example
1087 /// ```rust
1088 /// # #[cfg(feature = "deploy")] {
1089 /// # use hydro_lang::prelude::*;
1090 /// # use futures::StreamExt;
1091 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
1092 /// let tick = process.tick();
1093 /// let keyed_singleton = // { 1: 123, 2: 456, 0: 789 }
1094 /// # Stream::<_, _>::from(process.source_iter(q!(vec![(1, 123), (2, 456), (0, 789)])))
1095 /// # .into_keyed()
1096 /// # .first();
1097 /// keyed_singleton.get_max_key()
1098 /// # .sample_eager(nondet!(/** test */))
1099 /// # }, |mut stream| async move {
1100 /// // (2, 456)
1101 /// # assert_eq!(stream.next().await.unwrap(), (2, 456));
1102 /// # }));
1103 /// # }
1104 /// ```
1105 pub fn get_max_key(self) -> Optional<(K, V), L, B::UnderlyingBound>
1106 where
1107 K: Ord,
1108 {
1109 self.entries()
1110 .assume_ordering_trusted(nondet!(
1111 /// There is only one element associated with each key, and the keys are totallly
1112 /// ordered so we will produce a deterministic value. The closure technically
1113 /// isn't commutative in the case where both passed entries have the same key
1114 /// but different values.
1115 ///
1116 /// In the future, we may want to have an `assume!(...)` statement in the UDF that
1117 /// the two inputs do not have the same key.
1118 ))
1119 .reduce(q!(
1120 move |curr, new| {
1121 if new.0 > curr.0 {
1122 *curr = new;
1123 }
1124 },
1125 idempotent = ManualProof(/* repeated elements are ignored */)
1126 ))
1127 }
1128
1129 /// Converts this keyed singleton into a [`KeyedStream`] with each group having a single
1130 /// element, the value.
1131 ///
1132 /// This is the equivalent of [`Singleton::into_stream`] but keyed.
1133 ///
1134 /// # Example
1135 /// ```rust
1136 /// # #[cfg(feature = "deploy")] {
1137 /// # use hydro_lang::prelude::*;
1138 /// # use futures::StreamExt;
1139 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
1140 /// let keyed_singleton = // { 1: 2, 2: 4 }
1141 /// # Stream::<_, _>::from(process.source_iter(q!(vec![(1, 2), (2, 4)])))
1142 /// # .into_keyed()
1143 /// # .first();
1144 /// keyed_singleton
1145 /// .clone()
1146 /// .into_keyed_stream()
1147 /// .interleave(
1148 /// keyed_singleton.into_keyed_stream()
1149 /// )
1150 /// # .entries()
1151 /// # }, |mut stream| async move {
1152 /// /// // { 1: [2, 2], 2: [4, 4] }
1153 /// # for w in vec![(1, 2), (2, 4), (1, 2), (2, 4)] {
1154 /// # assert_eq!(stream.next().await.unwrap(), w);
1155 /// # }
1156 /// # }));
1157 /// # }
1158 /// ```
1159 pub fn into_keyed_stream(
1160 self,
1161 ) -> KeyedStream<K, V, L, B::UnderlyingBound, TotalOrder, ExactlyOnce> {
1162 KeyedStream::new(
1163 self.location.clone(),
1164 HydroNode::Cast {
1165 inner: Box::new(self.ir_node.into_inner()),
1166 metadata: self.location.new_node_metadata(KeyedStream::<
1167 K,
1168 V,
1169 L,
1170 B::UnderlyingBound,
1171 TotalOrder,
1172 ExactlyOnce,
1173 >::collection_kind()),
1174 },
1175 )
1176 }
1177}
1178
1179impl<'a, K, V, L, B: KeyedSingletonBound> KeyedSingleton<K, V, L, B>
1180where
1181 L: Location<'a>,
1182{
1183 /// Shifts this keyed singleton into an atomic context, which guarantees that any downstream logic
1184 /// will all be executed synchronously before any outputs are yielded (in [`KeyedSingleton::end_atomic`]).
1185 ///
1186 /// This is useful to enforce local consistency constraints, such as ensuring that a write is
1187 /// processed before an acknowledgement is emitted. Entering an atomic section requires a [`Tick`]
1188 /// argument that declares where the keyed singleton will be atomically processed. Batching a
1189 /// keyed singleton into the _same_ [`Tick`] will preserve the synchronous execution, while
1190 /// batching into a different [`Tick`] will introduce asynchrony.
1191 pub fn atomic(self, tick: &Tick<L>) -> KeyedSingleton<K, V, Atomic<L>, B> {
1192 let out_location = Atomic { tick: tick.clone() };
1193 KeyedSingleton::new(
1194 out_location.clone(),
1195 HydroNode::BeginAtomic {
1196 inner: Box::new(self.ir_node.into_inner()),
1197 metadata: out_location
1198 .new_node_metadata(KeyedSingleton::<K, V, Atomic<L>, B>::collection_kind()),
1199 },
1200 )
1201 }
1202}
1203
1204impl<'a, K, V, L, B: KeyedSingletonBound> KeyedSingleton<K, V, Atomic<L>, B>
1205where
1206 L: Location<'a> + NoTick,
1207{
1208 /// Yields the elements of this keyed singleton back into a top-level, asynchronous execution context.
1209 /// See [`KeyedSingleton::atomic`] for more details.
1210 pub fn end_atomic(self) -> KeyedSingleton<K, V, L, B> {
1211 KeyedSingleton::new(
1212 self.location.tick.l.clone(),
1213 HydroNode::EndAtomic {
1214 inner: Box::new(self.ir_node.into_inner()),
1215 metadata: self
1216 .location
1217 .tick
1218 .l
1219 .new_node_metadata(KeyedSingleton::<K, V, L, B>::collection_kind()),
1220 },
1221 )
1222 }
1223}
1224
1225impl<'a, K, V, L: Location<'a>> KeyedSingleton<K, V, Tick<L>, Bounded> {
1226 /// Shifts the state in `self` to the **next tick**, so that the returned keyed singleton at
1227 /// tick `T` always has the entries of `self` at tick `T - 1`.
1228 ///
1229 /// At tick `0`, the output has no entries, since there is no previous tick.
1230 ///
1231 /// This operator enables stateful iterative processing with ticks, by sending data from one
1232 /// tick to the next. For example, you can use it to compare state across consecutive batches.
1233 ///
1234 /// # Example
1235 /// ```rust
1236 /// # #[cfg(feature = "deploy")] {
1237 /// # use hydro_lang::prelude::*;
1238 /// # use futures::StreamExt;
1239 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
1240 /// let tick = process.tick();
1241 /// # // ticks are lazy by default, forces the second tick to run
1242 /// # tick.spin_batch(q!(1)).all_ticks().for_each(q!(|_| {}));
1243 /// # let batch_first_tick = process
1244 /// # .source_iter(q!(vec![(1, 2), (2, 3)]))
1245 /// # .batch(&tick, nondet!(/** test */))
1246 /// # .into_keyed();
1247 /// # let batch_second_tick = process
1248 /// # .source_iter(q!(vec![(2, 4), (3, 5)]))
1249 /// # .batch(&tick, nondet!(/** test */))
1250 /// # .into_keyed()
1251 /// # .defer_tick(); // appears on the second tick
1252 /// let input_batch = // first tick: { 1: 2, 2: 3 }, second tick: { 2: 4, 3: 5 }
1253 /// # batch_first_tick.chain(batch_second_tick).first();
1254 /// input_batch.clone().filter_key_not_in(
1255 /// input_batch.defer_tick().keys() // keys present in the previous tick
1256 /// )
1257 /// # .entries().all_ticks()
1258 /// # }, |mut stream| async move {
1259 /// // { 1: 2, 2: 3 } (first tick), { 3: 5 } (second tick)
1260 /// # for w in vec![(1, 2), (2, 3), (3, 5)] {
1261 /// # assert_eq!(stream.next().await.unwrap(), w);
1262 /// # }
1263 /// # }));
1264 /// # }
1265 /// ```
1266 pub fn defer_tick(self) -> KeyedSingleton<K, V, Tick<L>, Bounded> {
1267 KeyedSingleton::new(
1268 self.location.clone(),
1269 HydroNode::DeferTick {
1270 input: Box::new(self.ir_node.into_inner()),
1271 metadata: self
1272 .location
1273 .new_node_metadata(KeyedSingleton::<K, V, Tick<L>, Bounded>::collection_kind()),
1274 },
1275 )
1276 }
1277}
1278
1279impl<'a, K, V, L, B: KeyedSingletonBound<ValueBound = Unbounded>> KeyedSingleton<K, V, L, B>
1280where
1281 L: Location<'a>,
1282{
1283 /// Returns a keyed singleton with a snapshot of each key-value entry at a non-deterministic
1284 /// point in time.
1285 ///
1286 /// # Non-Determinism
1287 /// Because this picks a snapshot of each entry, which is continuously changing, each output has a
1288 /// non-deterministic set of entries since each snapshot can be at an arbitrary point in time.
1289 pub fn snapshot(
1290 self,
1291 tick: &Tick<L>,
1292 _nondet: NonDet,
1293 ) -> KeyedSingleton<K, V, Tick<L>, Bounded> {
1294 assert_eq!(Location::id(tick.outer()), Location::id(&self.location));
1295 KeyedSingleton::new(
1296 tick.clone(),
1297 HydroNode::Batch {
1298 inner: Box::new(self.ir_node.into_inner()),
1299 metadata: tick
1300 .new_node_metadata(KeyedSingleton::<K, V, Tick<L>, Bounded>::collection_kind()),
1301 },
1302 )
1303 }
1304}
1305
1306impl<'a, K, V, L, B: KeyedSingletonBound<ValueBound = Unbounded>> KeyedSingleton<K, V, Atomic<L>, B>
1307where
1308 L: Location<'a> + NoTick,
1309{
1310 /// Returns a keyed singleton with a snapshot of each key-value entry, consistent with the
1311 /// state of the keyed singleton being atomically processed.
1312 ///
1313 /// # Non-Determinism
1314 /// Because this picks a snapshot of each entry, which is continuously changing, each output has a
1315 /// non-deterministic set of entries since each snapshot can be at an arbitrary point in time.
1316 pub fn snapshot_atomic(self, _nondet: NonDet) -> KeyedSingleton<K, V, Tick<L>, Bounded> {
1317 KeyedSingleton::new(
1318 self.location.clone().tick,
1319 HydroNode::Batch {
1320 inner: Box::new(self.ir_node.into_inner()),
1321 metadata: self.location.tick.new_node_metadata(KeyedSingleton::<
1322 K,
1323 V,
1324 Tick<L>,
1325 Bounded,
1326 >::collection_kind(
1327 )),
1328 },
1329 )
1330 }
1331}
1332
1333impl<'a, K, V, L, B: KeyedSingletonBound<ValueBound = Bounded>> KeyedSingleton<K, V, L, B>
1334where
1335 L: Location<'a>,
1336{
1337 /// Creates a keyed singleton containing only the key-value pairs where the value satisfies a predicate `f`.
1338 ///
1339 /// The closure `f` receives a reference `&V` to each value and returns a boolean. If the predicate
1340 /// returns `true`, the key-value pair is included in the output. If it returns `false`, the pair
1341 /// is filtered out.
1342 ///
1343 /// The closure `f` receives a reference `&V` rather than an owned value `V` because filtering does
1344 /// not modify or take ownership of the values. If you need to modify the values while filtering
1345 /// use [`KeyedSingleton::filter_map`] instead.
1346 ///
1347 /// # Example
1348 /// ```rust
1349 /// # #[cfg(feature = "deploy")] {
1350 /// # use hydro_lang::prelude::*;
1351 /// # use futures::StreamExt;
1352 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
1353 /// let keyed_singleton = // { 1: 2, 2: 4, 3: 1 }
1354 /// # process
1355 /// # .source_iter(q!(vec![(1, 2), (2, 4), (3, 1)]))
1356 /// # .into_keyed()
1357 /// # .first();
1358 /// keyed_singleton.filter(q!(|&v| v > 1))
1359 /// # .entries()
1360 /// # }, |mut stream| async move {
1361 /// // { 1: 2, 2: 4 }
1362 /// # let mut results = Vec::new();
1363 /// # for _ in 0..2 {
1364 /// # results.push(stream.next().await.unwrap());
1365 /// # }
1366 /// # results.sort();
1367 /// # assert_eq!(results, vec![(1, 2), (2, 4)]);
1368 /// # }));
1369 /// # }
1370 /// ```
1371 pub fn filter<F>(self, f: impl IntoQuotedMut<'a, F, L> + Copy) -> KeyedSingleton<K, V, L, B>
1372 where
1373 F: Fn(&V) -> bool + 'a,
1374 {
1375 let f: ManualExpr<F, _> = ManualExpr::new(move |ctx: &L| f.splice_fn1_borrow_ctx(ctx));
1376 let filter_f = q!({
1377 let orig = f;
1378 move |t: &(_, _)| orig(&t.1)
1379 })
1380 .splice_fn1_borrow_ctx::<(K, V), bool>(&self.location)
1381 .into();
1382
1383 KeyedSingleton::new(
1384 self.location.clone(),
1385 HydroNode::Filter {
1386 f: filter_f,
1387 input: Box::new(self.ir_node.into_inner()),
1388 metadata: self
1389 .location
1390 .new_node_metadata(KeyedSingleton::<K, V, L, B>::collection_kind()),
1391 },
1392 )
1393 }
1394
1395 /// An operator that both filters and maps values. It yields only the key-value pairs where
1396 /// the supplied closure `f` returns `Some(value)`.
1397 ///
1398 /// The closure `f` receives each value `V` and returns `Option<U>`. If the closure returns
1399 /// `Some(new_value)`, the key-value pair `(key, new_value)` is included in the output.
1400 /// If it returns `None`, the key-value pair is filtered out.
1401 ///
1402 /// # Example
1403 /// ```rust
1404 /// # #[cfg(feature = "deploy")] {
1405 /// # use hydro_lang::prelude::*;
1406 /// # use futures::StreamExt;
1407 /// # tokio_test::block_on(hydro_lang::test_util::stream_transform_test(|process| {
1408 /// let keyed_singleton = // { 1: "42", 2: "hello", 3: "100" }
1409 /// # process
1410 /// # .source_iter(q!(vec![(1, "42"), (2, "hello"), (3, "100")]))
1411 /// # .into_keyed()
1412 /// # .first();
1413 /// keyed_singleton.filter_map(q!(|s| s.parse::<i32>().ok()))
1414 /// # .entries()
1415 /// # }, |mut stream| async move {
1416 /// // { 1: 42, 3: 100 }
1417 /// # let mut results = Vec::new();
1418 /// # for _ in 0..2 {
1419 /// # results.push(stream.next().await.unwrap());
1420 /// # }
1421 /// # results.sort();
1422 /// # assert_eq!(results, vec![(1, 42), (3, 100)]);
1423 /// # }));
1424 /// # }
1425 /// ```
1426 pub fn filter_map<F, U>(
1427 self,
1428 f: impl IntoQuotedMut<'a, F, L> + Copy,
1429 ) -> KeyedSingleton<K, U, L, B>
1430 where
1431 F: Fn(V) -> Option<U> + 'a,
1432 {
1433 let f: ManualExpr<F, _> = ManualExpr::new(move |ctx: &L| f.splice_fn1_ctx(ctx));
1434 let filter_map_f = q!({
1435 let orig = f;
1436 move |(k, v)| orig(v).map(|o| (k, o))
1437 })
1438 .splice_fn1_ctx::<(K, V), Option<(K, U)>>(&self.location)
1439 .into();
1440
1441 KeyedSingleton::new(
1442 self.location.clone(),
1443 HydroNode::FilterMap {
1444 f: filter_map_f,
1445 input: Box::new(self.ir_node.into_inner()),
1446 metadata: self
1447 .location
1448 .new_node_metadata(KeyedSingleton::<K, U, L, B>::collection_kind()),
1449 },
1450 )
1451 }
1452
1453 /// Returns a keyed singleton with entries consisting of _new_ key-value pairs that have
1454 /// arrived since the previous batch was released.
1455 ///
1456 /// Currently, there is no `all_ticks` dual on [`KeyedSingleton`], instead you may want to use
1457 /// [`KeyedSingleton::into_keyed_stream`] then yield with [`KeyedStream::all_ticks`].
1458 ///
1459 /// # Non-Determinism
1460 /// Because this picks a batch of asynchronously added entries, each output keyed singleton
1461 /// has a non-deterministic set of key-value pairs.
1462 pub fn batch(self, tick: &Tick<L>, nondet: NonDet) -> KeyedSingleton<K, V, Tick<L>, Bounded>
1463 where
1464 L: NoTick,
1465 {
1466 self.atomic(tick).batch_atomic(nondet)
1467 }
1468}
1469
1470impl<'a, K, V, L, B: KeyedSingletonBound<ValueBound = Bounded>> KeyedSingleton<K, V, Atomic<L>, B>
1471where
1472 L: Location<'a> + NoTick,
1473{
1474 /// Returns a keyed singleton with entries consisting of _new_ key-value pairs that are being
1475 /// atomically processed.
1476 ///
1477 /// Currently, there is no dual to asynchronously yield back outside the tick, instead you
1478 /// should use [`KeyedSingleton::into_keyed_stream`] and yield a [`KeyedStream`].
1479 ///
1480 /// # Non-Determinism
1481 /// Because this picks a batch of asynchronously added entries, each output keyed singleton
1482 /// has a non-deterministic set of key-value pairs.
1483 pub fn batch_atomic(self, nondet: NonDet) -> KeyedSingleton<K, V, Tick<L>, Bounded> {
1484 let _ = nondet;
1485 KeyedSingleton::new(
1486 self.location.clone().tick,
1487 HydroNode::Batch {
1488 inner: Box::new(self.ir_node.into_inner()),
1489 metadata: self.location.tick.new_node_metadata(KeyedSingleton::<
1490 K,
1491 V,
1492 Tick<L>,
1493 Bounded,
1494 >::collection_kind(
1495 )),
1496 },
1497 )
1498 }
1499}
1500
1501#[cfg(test)]
1502mod tests {
1503 #[cfg(feature = "deploy")]
1504 use futures::{SinkExt, StreamExt};
1505 #[cfg(feature = "deploy")]
1506 use hydro_deploy::Deployment;
1507 #[cfg(any(feature = "deploy", feature = "sim"))]
1508 use stageleft::q;
1509
1510 #[cfg(any(feature = "deploy", feature = "sim"))]
1511 use crate::compile::builder::FlowBuilder;
1512 #[cfg(any(feature = "deploy", feature = "sim"))]
1513 use crate::location::Location;
1514 #[cfg(any(feature = "deploy", feature = "sim"))]
1515 use crate::nondet::nondet;
1516
1517 #[cfg(feature = "deploy")]
1518 #[tokio::test]
1519 async fn key_count_bounded_value() {
1520 let mut deployment = Deployment::new();
1521
1522 let mut flow = FlowBuilder::new();
1523 let node = flow.process::<()>();
1524 let external = flow.external::<()>();
1525
1526 let (input_port, input) = node.source_external_bincode(&external);
1527 let out = input
1528 .into_keyed()
1529 .first()
1530 .key_count()
1531 .sample_eager(nondet!(/** test */))
1532 .send_bincode_external(&external);
1533
1534 let nodes = flow
1535 .with_process(&node, deployment.Localhost())
1536 .with_external(&external, deployment.Localhost())
1537 .deploy(&mut deployment);
1538
1539 deployment.deploy().await.unwrap();
1540
1541 let mut external_in = nodes.connect(input_port).await;
1542 let mut external_out = nodes.connect(out).await;
1543
1544 deployment.start().await.unwrap();
1545
1546 assert_eq!(external_out.next().await.unwrap(), 0);
1547
1548 external_in.send((1, 1)).await.unwrap();
1549 assert_eq!(external_out.next().await.unwrap(), 1);
1550
1551 external_in.send((2, 2)).await.unwrap();
1552 assert_eq!(external_out.next().await.unwrap(), 2);
1553 }
1554
1555 #[cfg(feature = "deploy")]
1556 #[tokio::test]
1557 async fn key_count_unbounded_value() {
1558 let mut deployment = Deployment::new();
1559
1560 let mut flow = FlowBuilder::new();
1561 let node = flow.process::<()>();
1562 let external = flow.external::<()>();
1563
1564 let (input_port, input) = node.source_external_bincode(&external);
1565 let out = input
1566 .into_keyed()
1567 .fold(q!(|| 0), q!(|acc, _| *acc += 1))
1568 .key_count()
1569 .sample_eager(nondet!(/** test */))
1570 .send_bincode_external(&external);
1571
1572 let nodes = flow
1573 .with_process(&node, deployment.Localhost())
1574 .with_external(&external, deployment.Localhost())
1575 .deploy(&mut deployment);
1576
1577 deployment.deploy().await.unwrap();
1578
1579 let mut external_in = nodes.connect(input_port).await;
1580 let mut external_out = nodes.connect(out).await;
1581
1582 deployment.start().await.unwrap();
1583
1584 assert_eq!(external_out.next().await.unwrap(), 0);
1585
1586 external_in.send((1, 1)).await.unwrap();
1587 assert_eq!(external_out.next().await.unwrap(), 1);
1588
1589 external_in.send((1, 2)).await.unwrap();
1590 assert_eq!(external_out.next().await.unwrap(), 1);
1591
1592 external_in.send((2, 2)).await.unwrap();
1593 assert_eq!(external_out.next().await.unwrap(), 2);
1594
1595 external_in.send((1, 1)).await.unwrap();
1596 assert_eq!(external_out.next().await.unwrap(), 2);
1597
1598 external_in.send((3, 1)).await.unwrap();
1599 assert_eq!(external_out.next().await.unwrap(), 3);
1600 }
1601
1602 #[cfg(feature = "deploy")]
1603 #[tokio::test]
1604 async fn into_singleton_bounded_value() {
1605 let mut deployment = Deployment::new();
1606
1607 let mut flow = FlowBuilder::new();
1608 let node = flow.process::<()>();
1609 let external = flow.external::<()>();
1610
1611 let (input_port, input) = node.source_external_bincode(&external);
1612 let out = input
1613 .into_keyed()
1614 .first()
1615 .into_singleton()
1616 .sample_eager(nondet!(/** test */))
1617 .send_bincode_external(&external);
1618
1619 let nodes = flow
1620 .with_process(&node, deployment.Localhost())
1621 .with_external(&external, deployment.Localhost())
1622 .deploy(&mut deployment);
1623
1624 deployment.deploy().await.unwrap();
1625
1626 let mut external_in = nodes.connect(input_port).await;
1627 let mut external_out = nodes.connect(out).await;
1628
1629 deployment.start().await.unwrap();
1630
1631 assert_eq!(
1632 external_out.next().await.unwrap(),
1633 std::collections::HashMap::new()
1634 );
1635
1636 external_in.send((1, 1)).await.unwrap();
1637 assert_eq!(
1638 external_out.next().await.unwrap(),
1639 vec![(1, 1)].into_iter().collect()
1640 );
1641
1642 external_in.send((2, 2)).await.unwrap();
1643 assert_eq!(
1644 external_out.next().await.unwrap(),
1645 vec![(1, 1), (2, 2)].into_iter().collect()
1646 );
1647 }
1648
1649 #[cfg(feature = "deploy")]
1650 #[tokio::test]
1651 async fn into_singleton_unbounded_value() {
1652 let mut deployment = Deployment::new();
1653
1654 let mut flow = FlowBuilder::new();
1655 let node = flow.process::<()>();
1656 let external = flow.external::<()>();
1657
1658 let (input_port, input) = node.source_external_bincode(&external);
1659 let out = input
1660 .into_keyed()
1661 .fold(q!(|| 0), q!(|acc, _| *acc += 1))
1662 .into_singleton()
1663 .sample_eager(nondet!(/** test */))
1664 .send_bincode_external(&external);
1665
1666 let nodes = flow
1667 .with_process(&node, deployment.Localhost())
1668 .with_external(&external, deployment.Localhost())
1669 .deploy(&mut deployment);
1670
1671 deployment.deploy().await.unwrap();
1672
1673 let mut external_in = nodes.connect(input_port).await;
1674 let mut external_out = nodes.connect(out).await;
1675
1676 deployment.start().await.unwrap();
1677
1678 assert_eq!(
1679 external_out.next().await.unwrap(),
1680 std::collections::HashMap::new()
1681 );
1682
1683 external_in.send((1, 1)).await.unwrap();
1684 assert_eq!(
1685 external_out.next().await.unwrap(),
1686 vec![(1, 1)].into_iter().collect()
1687 );
1688
1689 external_in.send((1, 2)).await.unwrap();
1690 assert_eq!(
1691 external_out.next().await.unwrap(),
1692 vec![(1, 2)].into_iter().collect()
1693 );
1694
1695 external_in.send((2, 2)).await.unwrap();
1696 assert_eq!(
1697 external_out.next().await.unwrap(),
1698 vec![(1, 2), (2, 1)].into_iter().collect()
1699 );
1700
1701 external_in.send((1, 1)).await.unwrap();
1702 assert_eq!(
1703 external_out.next().await.unwrap(),
1704 vec![(1, 3), (2, 1)].into_iter().collect()
1705 );
1706
1707 external_in.send((3, 1)).await.unwrap();
1708 assert_eq!(
1709 external_out.next().await.unwrap(),
1710 vec![(1, 3), (2, 1), (3, 1)].into_iter().collect()
1711 );
1712 }
1713
1714 #[cfg(feature = "sim")]
1715 #[test]
1716 fn sim_unbounded_singleton_snapshot() {
1717 let mut flow = FlowBuilder::new();
1718 let node = flow.process::<()>();
1719
1720 let (input_port, input) = node.sim_input();
1721 let output = input
1722 .into_keyed()
1723 .fold(q!(|| 0), q!(|acc, _| *acc += 1))
1724 .snapshot(&node.tick(), nondet!(/** test */))
1725 .entries()
1726 .all_ticks()
1727 .sim_output();
1728
1729 let count = flow.sim().exhaustive(async || {
1730 input_port.send((1, 123));
1731 input_port.send((1, 456));
1732 input_port.send((2, 123));
1733
1734 let all = output.collect_sorted::<Vec<_>>().await;
1735 assert_eq!(all.last().unwrap(), &(2, 1));
1736 });
1737
1738 assert_eq!(count, 8);
1739 }
1740
1741 #[cfg(feature = "deploy")]
1742 #[tokio::test]
1743 async fn join_keyed_stream() {
1744 let mut deployment = Deployment::new();
1745
1746 let mut flow = FlowBuilder::new();
1747 let node = flow.process::<()>();
1748 let external = flow.external::<()>();
1749
1750 let tick = node.tick();
1751 let keyed_data = node
1752 .source_iter(q!(vec![(1, 10), (2, 20)]))
1753 .into_keyed()
1754 .batch(&tick, nondet!(/** test */))
1755 .first();
1756 let requests = node
1757 .source_iter(q!(vec![(1, 100), (2, 200), (3, 300)]))
1758 .into_keyed()
1759 .batch(&tick, nondet!(/** test */));
1760
1761 let out = keyed_data
1762 .join_keyed_stream(requests)
1763 .entries()
1764 .all_ticks()
1765 .send_bincode_external(&external);
1766
1767 let nodes = flow
1768 .with_process(&node, deployment.Localhost())
1769 .with_external(&external, deployment.Localhost())
1770 .deploy(&mut deployment);
1771
1772 deployment.deploy().await.unwrap();
1773
1774 let mut external_out = nodes.connect(out).await;
1775
1776 deployment.start().await.unwrap();
1777
1778 let mut results = vec![];
1779 for _ in 0..2 {
1780 results.push(external_out.next().await.unwrap());
1781 }
1782 results.sort();
1783
1784 assert_eq!(results, vec![(1, (10, 100)), (2, (20, 200))]);
1785 }
1786}