Fix shared state in retry (#829)

* Fix retry using shared state

* update changelog

* Update CHANGELOG.md

* Add unit test

Co-authored-by: Anders Ha <anders@andersha.com>

CHANGELOG.md

@@ -1,6 +1,8 @@
 # master
 *Please add new entries at the top.*
 
+1. Fixed issue where `SingalProducer.try(upTo:interval:count:)` shares state between invocation of `start` on the same producer.
+
 # 6.7.0
 # 6.7.0-rc1
 
@@ -42,7 +44,6 @@
 1. Bumped deployment target to iOS 9.0, per Xcode 12 warnings. (#818, kudos to @harleyjcooper)
 
 1. Fixed a few deprecation warning when the project is being built. (#819, kudos to @apps4everyone)
->>>>>>> origin/master
 
 # 6.5.0
 

Sources/SignalProducer.swift

@@ -2393,9 +2393,10 @@ extension SignalProducer {
 			return producer
 		}
 
-		var retries = count
+		return SignalProducer { observer, lifetime in
+			var retries = count
 
-		return flatMapError { error -> SignalProducer<Value, Error> in
+			lifetime += flatMapError { error -> SignalProducer<Value, Error> in
 				// The final attempt shouldn't defer the error if it fails
 				var producer = SignalProducer<Value, Error>(error: error)
 				if retries > 0 {
@@ -2408,6 +2409,8 @@ extension SignalProducer {
 				return producer
 			}
 			.retry(upTo: count)
+			.start(observer)
+		}
 	}
 
 	/// Wait for completion of `self`, *then* forward all events from

Tests/ReactiveSwiftTests/SignalProducerSpec.swift

@@ -2812,6 +2812,40 @@ class SignalProducerSpec: QuickSpec {
 					expect(errors) == [.default]
 				}
 
+				it("should not share retry counter across produced signals") {
+					let scheduler = TestScheduler()
+					var failuresReceived = 0
+
+					let original = SignalProducer<Int, TestError> { observer, _ in
+						scheduler.schedule { observer.send(error: .default) }
+					}
+					let retryDelay: DispatchTimeInterval = .seconds(1)
+					let retriable = original
+						.retry(upTo: 1, interval: retryDelay.timeInterval, on: scheduler)
+						.on(failed: { _ in failuresReceived += 1 })
+
+					// 1st produced signal
+					retriable.start()
+
+					scheduler.advance()
+					expect(failuresReceived) == 0
+
+					scheduler.advance(by: retryDelay)
+					expect(failuresReceived) == 1
+
+					// 2nd produced signal
+					retriable.start()
+
+					scheduler.advance()
+					// Shared retry counter had caused the second produced signal not to defer the failed event as expected.
+					// (https://github.com/ReactiveCocoa/ReactiveSwift/pull/829)
+					// Assert that we did not receive the 2th final failure at this point, which should always be delayed until
+					// at least `retryDelay` has elapsed.
+					expect(failuresReceived) == 1
+
+					scheduler.advance(by: retryDelay)
+					expect(failuresReceived) == 2
+				}
 			}
 
 		}