1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756
use indexmap::{IndexMap, IndexSet}; use instant::{Duration, Instant}; use log::*; use crate::{Analysis, EGraph, Id, Language, RecExpr, Rewrite, SearchMatches}; /** Faciliates running rewrites over an [`EGraph`]. One use for [`EGraph`]s is as the basis of a rewriting system. Since an egraph never "forgets" state when applying a [`Rewrite`], you can apply many rewrites many times quite efficiently. After the egraph is "full" (the rewrites can no longer find new equalities) or some other condition, the egraph compactly represents many, many equivalent expressions. At this point, the egraph is ready for extraction (see [`Extractor`]) which can pick the represented expression that's best according to some cost function. This technique is called [equality saturation](https://www.cs.cornell.edu/~ross/publications/eqsat/) in general. However, there can be many challenges in implementing this "outer loop" of applying rewrites, mostly revolving around which rules to run and when to stop. [`Runner`] is `egg`'s provided equality saturation engine that has reasonable defaults and implements many useful things like saturation checking, egraph size limits, and customizable rule [scheduling](trait.RewriteScheduler.html). Consider using [`Runner`] before rolling your own outer loop. Here are some of the things [`Runner`] does for you: - Saturation checking [`Runner`] checks to see if any of the rules added anything new to the [`EGraph`]. If none did, then it stops, returning [`StopReason::Saturated`](enum.StopReason.html#variant.Saturated). - Iteration limits You can set a upper limit of iterations to do in case the search doesn't stop for some other reason. If this limit is hit, it stops with [`StopReason::IterationLimit`](enum.StopReason.html#variant.IterationLimit). - [`EGraph`] size limit You can set a upper limit on the number of enodes in the egraph. If this limit is hit, it stops with [`StopReason::NodeLimit`](enum.StopReason.html#variant.NodeLimit). - Time limit You can set a time limit on the runner. If this limit is hit, it stops with [`StopReason::TimeLimit`](enum.StopReason.html#variant.TimeLimit). - Rule scheduling Some rules enable themselves, blowing up the [`EGraph`] and preventing other rewrites from running as many times. To prevent this, you can provide your own [`RewriteScheduler`] to govern when to run which rules. [`BackoffScheduler`] is the default scheduler. [`Runner`] generates [`Iteration`]s that record some data about each iteration. You can add your own data to this by implementing the [`IterationData`] trait. [`Runner`] is generic over the [`IterationData`] that it will be in the [`Iteration`]s, but by default it uses `()`. [`Runner`]: struct.Runner.html [`RewriteScheduler`]: trait.RewriteScheduler.html [`Extractor`]: struct.Extractor.html [`Rewrite`]: struct.Rewrite.html [`BackoffScheduler`]: struct.BackoffScheduler.html [`EGraph`]: struct.EGraph.html [`Iteration`]: struct.Iteration.html [`IterationData`]: trait.IterationData.html # Example ``` use egg::{*, rewrite as rw}; define_language! { enum SimpleLanguage { Num(i32), "+" = Add([Id; 2]), "*" = Mul([Id; 2]), Symbol(Symbol), } } let rules: &[Rewrite<SimpleLanguage, ()>] = &[ rw!("commute-add"; "(+ ?a ?b)" => "(+ ?b ?a)"), rw!("commute-mul"; "(* ?a ?b)" => "(* ?b ?a)"), rw!("add-0"; "(+ ?a 0)" => "?a"), rw!("mul-0"; "(* ?a 0)" => "0"), rw!("mul-1"; "(* ?a 1)" => "?a"), ]; pub struct MyIterData { smallest_so_far: usize, } type MyRunner = Runner<SimpleLanguage, (), MyIterData>; impl IterationData<SimpleLanguage, ()> for MyIterData { fn make(runner: &MyRunner) -> Self { let root = runner.roots[0]; let mut extractor = Extractor::new(&runner.egraph, AstSize); MyIterData { smallest_so_far: extractor.find_best(root).0, } } } let start = "(+ 0 (* 1 foo))".parse().unwrap(); // Runner is customizable in the builder pattern style. let runner = MyRunner::new(Default::default()) .with_iter_limit(10) .with_node_limit(10_000) .with_expr(&start) .with_scheduler(SimpleScheduler) .run(&rules); // Now we can check our iteration data to make sure that the cost only // got better over time for its in runner.iterations.windows(2) { assert!(its[0].data.smallest_so_far >= its[1].data.smallest_so_far); } println!( "Stopped after {} iterations, reason: {:?}", runner.iterations.len(), runner.stop_reason ); ``` */ pub struct Runner<L: Language, N: Analysis<L>, IterData = ()> { /// The [`EGraph`](struct.EGraph.html) used. pub egraph: EGraph<L, N>, /// Data accumulated over each [`Iteration`](struct.Iteration.html). pub iterations: Vec<Iteration<IterData>>, /// The roots of expressions added by the /// [`with_expr`](#method.with_expr) method, in insertion order. pub roots: Vec<Id>, /// Why the `Runner` stopped. This will be `None` if it hasn't /// stopped yet. pub stop_reason: Option<StopReason>, // limits iter_limit: usize, node_limit: usize, time_limit: Duration, start_time: Option<Instant>, scheduler: Box<dyn RewriteScheduler<L, N>>, } impl<L, N> Default for Runner<L, N, ()> where L: Language, N: Analysis<L> + Default, { fn default() -> Self { Runner::new(N::default()) } } /// Error returned by [`Runner`] when it stops. /// /// [`Runner`]: struct.Runner.html #[derive(Debug, Clone)] #[cfg_attr(feature = "serde-1", derive(serde::Serialize))] pub enum StopReason { /// The egraph saturated, i.e., there was an iteration where we /// didn't learn anything new from applying the rules. Saturated, /// The iteration limit was hit. The data is the iteration limit. IterationLimit(usize), /// The enode limit was hit. The data is the enode limit. NodeLimit(usize), /// The time limit was hit. The data is the time limit in seconds. TimeLimit(f64), } /// Data generated by running a [`Runner`] one iteration. /// /// If the `serde-1` feature is enabled, this implements /// [`serde::Serialize`][ser], which is useful if you want to output /// this as a JSON or some other format. /// /// [`Runner`]: struct.Runner.html /// [ser]: https://docs.rs/serde/latest/serde/trait.Serialize.html #[derive(Debug, Clone)] #[cfg_attr(feature = "serde-1", derive(serde::Serialize))] #[non_exhaustive] pub struct Iteration<IterData> { /// The number of enodes in the egraph at the start of this /// iteration. pub egraph_nodes: usize, /// The number of eclasses in the egraph at the start of this /// iteration. pub egraph_classes: usize, /// A map from rule name to number of times it was _newly_ applied /// in this iteration. pub applied: IndexMap<String, usize>, /// Seconds spent searching in this iteration. pub search_time: f64, /// Seconds spent applying rules in this iteration. pub apply_time: f64, /// Seconds spent [`rebuild`](struct.EGraph.html#method.rebuild)ing /// the egraph in this iteration. pub rebuild_time: f64, /// Total time spent in this iteration, including data generation time. pub total_time: f64, /// The user provided annotation for this iteration pub data: IterData, /// The number of rebuild iterations done after this iteration completed. pub n_rebuilds: usize, /// If the runner stopped on this iterations, this is the reason pub stop_reason: Option<StopReason>, } type RunnerResult<T> = std::result::Result<T, StopReason>; impl<L, N, IterData> Runner<L, N, IterData> where L: Language, N: Analysis<L>, IterData: IterationData<L, N>, { /// Create a new `Runner` with the given analysis and default parameters. pub fn new(analysis: N) -> Self { Self { iter_limit: 30, node_limit: 10_000, time_limit: Duration::from_secs(5), egraph: EGraph::new(analysis), roots: vec![], iterations: vec![], stop_reason: None, start_time: None, scheduler: Box::new(BackoffScheduler::default()), } } /// Sets the iteration limit. Default: 30 pub fn with_iter_limit(self, iter_limit: usize) -> Self { Self { iter_limit, ..self } } /// Sets the egraph size limit (in enodes). Default: 10,000 pub fn with_node_limit(self, node_limit: usize) -> Self { Self { node_limit, ..self } } /// Sets the runner time limit. Default: 5 seconds pub fn with_time_limit(self, time_limit: Duration) -> Self { Self { time_limit, ..self } } /// Change out the [`RewriteScheduler`] used by this [`Runner`]. /// The default one is [`BackoffScheduler`]. /// /// [`RewriteScheduler`]: trait.RewriteScheduler.html /// [`BackoffScheduler`]: struct.BackoffScheduler.html /// [`Runner`]: struct.Runner.html pub fn with_scheduler(self, scheduler: impl RewriteScheduler<L, N> + 'static) -> Self { let scheduler = Box::new(scheduler); Self { scheduler, ..self } } /// Add an expression to the egraph to be run. /// /// The eclass id of this addition will be recorded in the /// [`roots`](struct.Runner.html#structfield.roots) field, ordered by /// insertion order. pub fn with_expr(mut self, expr: &RecExpr<L>) -> Self { let id = self.egraph.add_expr(expr); self.roots.push(id); self } /// Replace the [`EGraph`](struct.EGraph.html) of this `Runner`. pub fn with_egraph(self, egraph: EGraph<L, N>) -> Self { Self { egraph, ..self } } /// Run this `Runner` until it stops. /// After this, the field /// [`stop_reason`](#structfield.stop_reason) is guaranteeed to be /// set. pub fn run(mut self, rules: &[Rewrite<L, N>]) -> Self { check_rules(rules); self.egraph.rebuild(); // TODO check that we haven't loop { if let Err(stop_reason) = self.run_one(rules) { info!("Stopping: {:?}", stop_reason); self.stop_reason = Some(stop_reason); // push on a final iteration to mark the end state self.iterations.push(Iteration { stop_reason: self.stop_reason.clone(), egraph_nodes: self.egraph.total_number_of_nodes(), egraph_classes: self.egraph.number_of_classes(), data: IterData::make(&self), applied: Default::default(), search_time: Default::default(), apply_time: Default::default(), rebuild_time: Default::default(), total_time: Default::default(), n_rebuilds: Default::default(), }); break; } } self } #[rustfmt::skip] /// Prints some information about a runners run. pub fn print_report(&self) { let search_time: f64 = self.iterations.iter().map(|i| i.search_time).sum(); let apply_time: f64 = self.iterations.iter().map(|i| i.apply_time).sum(); let rebuild_time: f64 = self.iterations.iter().map(|i| i.rebuild_time).sum(); let total_time: f64 = self.iterations.iter().map(|i| i.total_time).sum(); let iters = self.iterations.len(); let rebuilds: usize = self.iterations.iter().map(|i| i.n_rebuilds).sum(); let eg = &self.egraph; println!("Runner report"); println!("============="); println!(" Stop reason: {:?}", self.stop_reason.as_ref().unwrap()); println!(" Iterations: {}", iters); println!(" Egraph size: {} nodes, {} classes, {} memo", eg.total_number_of_nodes(), eg.number_of_classes(), eg.total_size()); println!(" Rebuilds: {}, {:.2} per iter", rebuilds, (rebuilds as f64) / (iters as f64)); println!(" Total time: {}", total_time); println!(" Search: ({:.2}) {}", search_time / total_time, search_time); println!(" Apply: ({:.2}) {}", apply_time / total_time, apply_time); println!(" Rebuild: ({:.2}) {}", rebuild_time / total_time, rebuild_time); } fn run_one(&mut self, rules: &[Rewrite<L, N>]) -> RunnerResult<()> { assert!(self.stop_reason.is_none()); info!("\nIteration {}", self.iterations.len()); self.try_start(); self.check_limits()?; let i = self.iterations.len(); let egraph_nodes = self.egraph.total_size(); let egraph_classes = self.egraph.number_of_classes(); trace!("EGraph {:?}", self.egraph.dump()); let start_time = Instant::now(); let mut matches = Vec::new(); for rule in rules { let ms = self.scheduler.search_rewrite(i, &self.egraph, rule); matches.push(ms); if self.check_limits().is_err() { // bail on searching, make sure applying doesn't do anything matches.clear(); break; } } let search_time = start_time.elapsed().as_secs_f64(); info!("Search time: {}", search_time); let apply_time = Instant::now(); let mut applied = IndexMap::new(); for (rw, ms) in rules.iter().zip(matches) { let total_matches: usize = ms.iter().map(|m| m.substs.len()).sum(); if total_matches == 0 { continue; } debug!("Applying {} {} times", rw.name(), total_matches); let actually_matched = self.scheduler.apply_rewrite(i, &mut self.egraph, rw, ms); if actually_matched > 0 { if let Some(count) = applied.get_mut(rw.name()) { *count += actually_matched; } else { applied.insert(rw.name().to_owned(), actually_matched); } debug!("Applied {} {} times", rw.name(), actually_matched); } if self.check_limits().is_err() { break; } } let apply_time = apply_time.elapsed().as_secs_f64(); info!("Apply time: {}", apply_time); let rebuild_time = Instant::now(); let n_rebuilds = self.egraph.rebuild(); let rebuild_time = rebuild_time.elapsed().as_secs_f64(); info!("Rebuild time: {}", rebuild_time); info!( "Size: n={}, e={}", self.egraph.total_size(), self.egraph.number_of_classes() ); let saturated = applied.is_empty() && self.scheduler.can_stop(i); self.iterations.push(Iteration { applied, egraph_nodes, egraph_classes, search_time, apply_time, rebuild_time, n_rebuilds, data: IterData::make(&self), total_time: start_time.elapsed().as_secs_f64(), stop_reason: None, }); if saturated { Err(StopReason::Saturated) } else { Ok(()) } } fn try_start(&mut self) { self.start_time.get_or_insert_with(Instant::now); } fn check_limits(&self) -> RunnerResult<()> { let elapsed = self.start_time.unwrap().elapsed(); if elapsed > self.time_limit { return Err(StopReason::TimeLimit(elapsed.as_secs_f64())); } let size = self.egraph.total_size(); if size > self.node_limit { return Err(StopReason::NodeLimit(size)); } if self.iterations.len() >= self.iter_limit { return Err(StopReason::IterationLimit(self.iterations.len())); } Ok(()) } } fn check_rules<L, N>(rules: &[Rewrite<L, N>]) { let mut name_counts = IndexMap::new(); for rw in rules { *name_counts.entry(rw.name()).or_default() += 1 } name_counts.retain(|_, count: &mut usize| *count > 1); if !name_counts.is_empty() { eprintln!("WARNING: Duplicated rule names may affect rule reporting and scheduling."); log::warn!("Duplicated rule names may affect rule reporting and scheduling."); for (name, &count) in name_counts.iter() { assert!(count > 1); eprintln!("Rule '{}' appears {} times", name, count); log::warn!("Rule '{}' appears {} times", name, count); } } } /** A way to customize how a [`Runner`] runs [`Rewrite`]s. This gives you a way to prevent certain [`Rewrite`]s from exploding the [`EGraph`] and dominating how much time is spent while running the [`Runner`]. [`EGraph`]: struct.EGraph.html [`Runner`]: struct.Runner.html [`Rewrite`]: struct.Rewrite.html */ #[allow(unused_variables)] pub trait RewriteScheduler<L, N> where L: Language, N: Analysis<L>, { /// Whether or not the [`Runner`](struct.Runner.html) is allowed /// to say it has saturated. /// /// This is only called when the runner is otherwise saturated. /// Default implementation just returns `true`. fn can_stop(&mut self, iteration: usize) -> bool { true } /// A hook allowing you to customize rewrite searching behavior. /// Useful to implement rule management. /// /// Default implementation just calls /// [`Rewrite::search`](struct.Rewrite.html#method.search). fn search_rewrite( &mut self, iteration: usize, egraph: &EGraph<L, N>, rewrite: &Rewrite<L, N>, ) -> Vec<SearchMatches> { rewrite.search(egraph) } /// A hook allowing you to customize rewrite application behavior. /// Useful to implement rule management. /// /// Default implementation just calls /// [`Rewrite::apply`](struct.Rewrite.html#method.apply) /// and returns number of new applications. fn apply_rewrite( &mut self, iteration: usize, egraph: &mut EGraph<L, N>, rewrite: &Rewrite<L, N>, matches: Vec<SearchMatches>, ) -> usize { rewrite.apply(egraph, &matches).len() } } /// A very simple [`RewriteScheduler`] that runs every rewrite every /// time. /// /// Using this is basically turning off rule scheduling. /// It uses the default implementation for all [`RewriteScheduler`] /// methods. /// /// This is not the default scheduler; choose it with the /// [`with_scheduler`](struct.Runner.html#method.with_scheduler) /// method. /// /// [`RewriteScheduler`]: trait.RewriteScheduler.html pub struct SimpleScheduler; impl<L, N> RewriteScheduler<L, N> for SimpleScheduler where L: Language, N: Analysis<L>, { } /// A [`RewriteScheduler`] that implements exponentional rule backoff. /// /// For each rewrite, there exists a configurable initial match limit. /// If a rewrite search yield more than this limit, then we ban this /// rule for number of iterations, double its limit, and double the time /// it will be banned next time. /// /// This seems effective at preventing explosive rules like /// associativity from taking an unfair amount of resources. /// /// [`BackoffScheduler`] is configurable in the builder-pattern style. /// /// [`RewriteScheduler`]: trait.RewriteScheduler.html /// [`BackoffScheduler`]: struct.BackoffScheduler.html pub struct BackoffScheduler { initial_match_limit: usize, ban_length: usize, stats: IndexMap<String, RuleStats>, dont_ban: IndexSet<String>, } struct RuleStats { times_applied: usize, banned_until: usize, times_banned: usize, } impl BackoffScheduler { /// Set the initial match limit after which a rule will be banned. /// Default: 1,000 pub fn with_initial_match_limit(self, initial_match_limit: usize) -> Self { Self { initial_match_limit, ..self } } /// Set the initial ban length. /// Default: 5 iterations pub fn with_ban_length(self, ban_length: usize) -> Self { Self { ban_length, ..self } } /// Never ban a particular rule. pub fn do_not_ban(mut self, name: impl Into<String>) -> Self { self.dont_ban.insert(name.into()); self } } impl Default for BackoffScheduler { fn default() -> Self { Self { dont_ban: Default::default(), stats: Default::default(), initial_match_limit: 1_000, ban_length: 5, } } } impl<L, N> RewriteScheduler<L, N> for BackoffScheduler where L: Language, N: Analysis<L>, { fn can_stop(&mut self, iteration: usize) -> bool { let n_stats = self.stats.len(); assert!(n_stats > 0); let mut banned: Vec<_> = self .stats .iter_mut() .filter(|(_, s)| s.banned_until > iteration) .collect(); if banned.is_empty() { true } else { let min_ban = banned .iter() .map(|(_, s)| s.banned_until) .min() .expect("banned cannot be empty here"); assert!(min_ban >= iteration); let delta = min_ban - iteration; let mut unbanned = vec![]; for (name, s) in &mut banned { s.banned_until -= delta; if s.banned_until == iteration { unbanned.push(name.as_str()); } } assert!(!unbanned.is_empty()); info!( "Banned {}/{}, fast-forwarded by {} to unban {}", banned.len(), n_stats, delta, unbanned.join(", "), ); false } } fn search_rewrite( &mut self, iteration: usize, egraph: &EGraph<L, N>, rewrite: &Rewrite<L, N>, ) -> Vec<SearchMatches> { if let Some(limit) = self.stats.get_mut(rewrite.name()) { if iteration < limit.banned_until { debug!( "Skipping {} ({}-{}), banned until {}...", rewrite.name(), limit.times_applied, limit.times_banned, limit.banned_until, ); return vec![]; } let matches = rewrite.search(egraph); let total_len: usize = matches.iter().map(|m| m.substs.len()).sum(); let threshold = self.initial_match_limit << limit.times_banned; if total_len > threshold { let ban_length = self.ban_length << limit.times_banned; limit.times_banned += 1; limit.banned_until = iteration + ban_length; info!( "Banning {} ({}-{}) for {} iters: {} < {}", rewrite.name(), limit.times_applied, limit.times_banned, ban_length, threshold, total_len, ); vec![] } else { limit.times_applied += 1; matches } } else { if !self.dont_ban.contains(rewrite.name()) { self.stats.insert( rewrite.name().into(), RuleStats { times_applied: 0, banned_until: 0, times_banned: 0, }, ); } rewrite.search(egraph) } } } /// Custom data to inject into the [`Iteration`]s recorded by a [`Runner`] /// /// This trait allows you to add custom data to the [`Iteration`]s /// recorded as a [`Runner`] applies rules. /// /// See the [`Runner`] docs for an example. /// /// [`Runner`] is generic over the [`IterationData`] that it will be in the /// [`Iteration`]s, but by default it uses `()`. /// /// [`Runner`]: struct.Runner.html /// [`Iteration`]: struct.Iteration.html /// [`IterationData`]: trait.IterationData.html pub trait IterationData<L, N>: Sized where L: Language, N: Analysis<L>, { /// Given the current [`Runner`](struct.Runner.html), make the /// data to be put in this [`Iteration`](struct.Iteration.html). fn make(runner: &Runner<L, N, Self>) -> Self; } impl<L, N> IterationData<L, N> for () where L: Language, N: Analysis<L>, { fn make(_: &Runner<L, N, Self>) -> Self {} }