Bug 19642 – std.range.slide!(No.withPartial) on lengthless forward range: get empty when expecting one window

Failing example (tested in DMD64 D Compiler v2.084.0): unittest { auto r1 = recurrence!((a,n) => a[n-1] - 1)(2).until(0); assert(r1.equal([2,1])); // succeeds auto r2 = r1.slide!(No.withPartial)(2); assert(! r2.empty); // fails, should be equal to [[2,1]] } The issue appears to be in the "Standard constructor" of the Forward Range version of std.range.Slides when needsEndTracker = true. The nextSource field is initialized using drop(stepSize) and later _empty is set to true if nextSource.empty, which is incorrect behavior because if exactly n elements can be dropped, then the range should contain one window of those n elements. Suggested fix: do something similar to Slides.popFront() such as 1) change line 8430 of std.range.package.d nextSource = source.save.drop(windowSize); to nextSource = source.save; auto poppedElems = nextSource.popFrontN(windowSize); 2) change line 8458 if (nextSource.empty) to if (poppedElements < windowSize)

Another example (encountered writing nlp code manipulating ngrams): "ab cd".splitter(' ').slide!(No.withPartial)(2) Should be [["ab", "cd"]], actual is [].

Another repro: ``` assert([false, true, false] .chunkBy!(a => a) .slide!(No.withPartial)(3) .count == 1); ```

@FeepingCreature updated dlang/phobos pull request #8738 "Fix issue 16942: slide!(No.withPartial): Correctly handle first slide exhausting input range." fixing this issue: - Fix issue 19642: slide!(No.withPartial): Correctly handle first slide exhausting input range. https://github.com/dlang/phobos/pull/8738

dlang/phobos pull request #8738 "Fix issue 16942: slide!(No.withPartial): Correctly handle first slide exhausting input range." was merged into master: - d78e46c2f19b7be5b2a9dd572ed779c376051036 by Mathis Beer: Fix issue 19642: slide!(No.withPartial): Correctly handle first slide exhausting input range. https://github.com/dlang/phobos/pull/8738