Merge #771

771: impl IntoParallelIterator for Range<char> r=cuviper a=CAD97 Closes #770 Implements both unindexed and indexed parallel iteration for `Range<char>` and `RangeInclusive<char>`. Co-authored-by: CAD97 <[email protected]>
rayon-rs · Jul 21, 2020 · 97b7e34 · 97b7e34
2 parents d3e32be + 8f37302
commit 97b7e34
Show file tree

Hide file tree

Showing 3 changed files with 196 additions and 2 deletions.
diff --git a/src/range.rs b/src/range.rs
@@ -18,6 +18,7 @@
 
 use crate::iter::plumbing::*;
 use crate::iter::*;
+use std::char;
 use std::ops::Range;
 use std::usize;
 
@@ -223,6 +224,75 @@ unindexed_range_impl! {i64, u64}
 unindexed_range_impl! {u128, u128}
 unindexed_range_impl! {i128, u128}
 
+// char is special because of the surrogate range hole
+macro_rules! convert_char {
+    ( $self:ident . $method:ident ( $( $arg:expr ),* ) ) => {{
+        let start = $self.range.start as u32;
+        let end = $self.range.end as u32;
+        if start < 0xD800 && 0xE000 < end {
+            // chain the before and after surrogate range fragments
+            (start..0xD800)
+                .into_par_iter()
+                .chain(0xE000..end)
+                .map(|codepoint| unsafe { char::from_u32_unchecked(codepoint) })
+                .$method($( $arg ),*)
+        } else {
+            // no surrogate range to worry about
+            (start..end)
+                .into_par_iter()
+                .map(|codepoint| unsafe { char::from_u32_unchecked(codepoint) })
+                .$method($( $arg ),*)
+        }
+    }};
+}
+
+impl ParallelIterator for Iter<char> {
+    type Item = char;
+
+    fn drive_unindexed<C>(self, consumer: C) -> C::Result
+    where
+        C: UnindexedConsumer<Self::Item>,
+    {
+        convert_char!(self.drive(consumer))
+    }
+
+    fn opt_len(&self) -> Option<usize> {
+        Some(self.len())
+    }
+}
+
+impl IndexedParallelIterator for Iter<char> {
+    // Split at the surrogate range first if we're allowed to
+    fn drive<C>(self, consumer: C) -> C::Result
+    where
+        C: Consumer<Self::Item>,
+    {
+        convert_char!(self.drive(consumer))
+    }
+
+    fn len(&self) -> usize {
+        // Taken from <char as Step>::steps_between
+        let start = self.range.start as u32;
+        let end = self.range.end as u32;
+        if start < end {
+            let mut count = end - start;
+            if start < 0xD800 && 0xE000 <= end {
+                count -= 0x800
+            }
+            count as usize
+        } else {
+            0
+        }
+    }
+
+    fn with_producer<CB>(self, callback: CB) -> CB::Output
+    where
+        CB: ProducerCallback<Self::Item>,
+    {
+        convert_char!(self.with_producer(callback))
+    }
+}
+
 #[test]
 fn check_range_split_at_overflow() {
     // Note, this split index overflows i8!

diff --git a/src/range_inclusive.rs b/src/range_inclusive.rs
@@ -18,6 +18,7 @@
 
 use crate::iter::plumbing::*;
 use crate::iter::*;
+use std::char;
 use std::ops::RangeInclusive;
 
 /// Parallel iterator over an inclusive range, implemented for all integer types.
@@ -48,15 +49,25 @@ pub struct Iter<T> {
 
 impl<T> Iter<T>
 where
-    RangeInclusive<T>: Clone + Iterator<Item = T> + DoubleEndedIterator,
+    RangeInclusive<T>: Eq,
+    T: Ord + Copy,
 {
     /// Returns `Some((start, end))` for `start..=end`, or `None` if it is exhausted.
     ///
     /// Note that `RangeInclusive` does not specify the bounds of an exhausted iterator,
     /// so this is a way for us to figure out what we've got.  Thankfully, all of the
     /// integer types we care about can be trivially cloned.
     fn bounds(&self) -> Option<(T, T)> {
-        Some((self.range.clone().next()?, self.range.clone().next_back()?))
+        let start = *self.range.start();
+        let end = *self.range.end();
+        if start <= end && self.range == (start..=end) {
+            // If the range is still nonempty, this is obviously true
+            // If the range is exhausted, either start > end or
+            // the range does not equal start..=end.
+            Some((start, end))
+        } else {
+            None
+        }
     }
 }
 
@@ -147,6 +158,80 @@ parallel_range_impl! {i64}
 parallel_range_impl! {u128}
 parallel_range_impl! {i128}
 
+// char is special
+macro_rules! convert_char {
+    ( $self:ident . $method:ident ( $( $arg:expr ),* ) ) => {
+        if let Some((start, end)) = $self.bounds() {
+            let start = start as u32;
+            let end = end as u32;
+            if start < 0xD800 && 0xE000 <= end {
+                // chain the before and after surrogate range fragments
+                (start..0xD800)
+                    .into_par_iter()
+                    .chain(0xE000..end + 1) // cannot use RangeInclusive, so add one to end
+                    .map(|codepoint| unsafe { char::from_u32_unchecked(codepoint) })
+                    .$method($( $arg ),*)
+            } else {
+                // no surrogate range to worry about
+                (start..end + 1) // cannot use RangeInclusive, so add one to end
+                    .into_par_iter()
+                    .map(|codepoint| unsafe { char::from_u32_unchecked(codepoint) })
+                    .$method($( $arg ),*)
+            }
+        } else {
+            empty().into_par_iter().$method($( $arg ),*)
+        }
+    };
+}
+
+impl ParallelIterator for Iter<char> {
+    type Item = char;
+
+    fn drive_unindexed<C>(self, consumer: C) -> C::Result
+    where
+        C: UnindexedConsumer<Self::Item>,
+    {
+        convert_char!(self.drive(consumer))
+    }
+
+    fn opt_len(&self) -> Option<usize> {
+        Some(self.len())
+    }
+}
+
+// Range<u32> is broken on 16 bit platforms, may as well benefit from it
+impl IndexedParallelIterator for Iter<char> {
+    // Split at the surrogate range first if we're allowed to
+    fn drive<C>(self, consumer: C) -> C::Result
+    where
+        C: Consumer<Self::Item>,
+    {
+        convert_char!(self.drive(consumer))
+    }
+
+    fn len(&self) -> usize {
+        if let Some((start, end)) = self.bounds() {
+            // Taken from <char as Step>::steps_between
+            let start = start as u32;
+            let end = end as u32;
+            let mut count = end - start;
+            if start < 0xD800 && 0xE000 <= end {
+                count -= 0x800
+            }
+            (count + 1) as usize // add one for inclusive
+        } else {
+            0
+        }
+    }
+
+    fn with_producer<CB>(self, callback: CB) -> CB::Output
+    where
+        CB: ProducerCallback<Self::Item>,
+    {
+        convert_char!(self.with_producer(callback))
+    }
+}
+
 #[test]
 #[cfg(target_pointer_width = "64")]
 fn test_u32_opt_len() {

diff --git a/tests/chars.rs b/tests/chars.rs
@@ -0,0 +1,39 @@
+use rayon::prelude::*;
+use std::char;
+
+#[test]
+fn half_open_correctness() {
+    let low = char::from_u32(0xD800 - 0x7).unwrap();
+    let high = char::from_u32(0xE000 + 0x7).unwrap();
+
+    let range = low..high;
+    let mut chars: Vec<char> = range.into_par_iter().collect();
+    chars.sort();
+
+    assert_eq!(
+        chars,
+        vec![
+            '\u{D7F9}', '\u{D7FA}', '\u{D7FB}', '\u{D7FC}', '\u{D7FD}', '\u{D7FE}', '\u{D7FF}',
+            '\u{E000}', '\u{E001}', '\u{E002}', '\u{E003}', '\u{E004}', '\u{E005}', '\u{E006}',
+        ]
+    );
+}
+
+#[test]
+fn closed_correctness() {
+    let low = char::from_u32(0xD800 - 0x7).unwrap();
+    let high = char::from_u32(0xE000 + 0x7).unwrap();
+
+    let range = low..=high;
+    let mut chars: Vec<char> = range.into_par_iter().collect();
+    chars.sort();
+
+    assert_eq!(
+        chars,
+        vec![
+            '\u{D7F9}', '\u{D7FA}', '\u{D7FB}', '\u{D7FC}', '\u{D7FD}', '\u{D7FE}', '\u{D7FF}',
+            '\u{E000}', '\u{E001}', '\u{E002}', '\u{E003}', '\u{E004}', '\u{E005}', '\u{E006}',
+            '\u{E007}',
+        ]
+    );
+}