polars_core/series/implementations/
null.rs

1use std::any::Any;
2
3use polars_error::constants::LENGTH_LIMIT_MSG;
4
5use self::compare_inner::TotalOrdInner;
6use super::*;
7use crate::chunked_array::ops::compare_inner::{IntoTotalEqInner, NonNull, TotalEqInner};
8use crate::chunked_array::ops::sort::arg_sort_multiple::arg_sort_multiple_impl;
9use crate::prelude::*;
10use crate::series::private::{PrivateSeries, PrivateSeriesNumeric};
11use crate::series::*;
12
13impl Series {
14    pub fn new_null(name: PlSmallStr, len: usize) -> Series {
15        NullChunked::new(name, len).into_series()
16    }
17}
18
19#[derive(Clone)]
20pub struct NullChunked {
21    pub(crate) name: PlSmallStr,
22    length: IdxSize,
23    // we still need chunks as many series consumers expect
24    // chunks to be there
25    chunks: Vec<ArrayRef>,
26}
27
28impl NullChunked {
29    pub(crate) fn new(name: PlSmallStr, len: usize) -> Self {
30        Self {
31            name,
32            length: len as IdxSize,
33            chunks: vec![Box::new(arrow::array::NullArray::new(
34                ArrowDataType::Null,
35                len,
36            ))],
37        }
38    }
39
40    pub fn len(&self) -> usize {
41        self.length as usize
42    }
43
44    pub fn is_empty(&self) -> bool {
45        self.length == 0
46    }
47}
48impl PrivateSeriesNumeric for NullChunked {
49    fn bit_repr(&self) -> Option<BitRepr> {
50        Some(BitRepr::U32(UInt32Chunked::full_null(
51            self.name.clone(),
52            self.len(),
53        )))
54    }
55}
56
57impl PrivateSeries for NullChunked {
58    fn compute_len(&mut self) {
59        fn inner(chunks: &[ArrayRef]) -> usize {
60            match chunks.len() {
61                // fast path
62                1 => chunks[0].len(),
63                _ => chunks.iter().fold(0, |acc, arr| acc + arr.len()),
64            }
65        }
66        self.length = IdxSize::try_from(inner(&self.chunks)).expect(LENGTH_LIMIT_MSG);
67    }
68    fn _field(&self) -> Cow<'_, Field> {
69        Cow::Owned(Field::new(self.name().clone(), DataType::Null))
70    }
71
72    #[allow(unused)]
73    fn _set_flags(&mut self, flags: StatisticsFlags) {}
74
75    fn _dtype(&self) -> &DataType {
76        &DataType::Null
77    }
78
79    #[cfg(feature = "zip_with")]
80    fn zip_with_same_type(&self, mask: &BooleanChunked, other: &Series) -> PolarsResult<Series> {
81        let len = match (self.len(), mask.len(), other.len()) {
82            (a, b, c) if a == b && b == c => a,
83            (1, a, b) | (a, 1, b) | (a, b, 1) if a == b => a,
84            (a, 1, 1) | (1, a, 1) | (1, 1, a) => a,
85            (_, 0, _) => 0,
86            _ => {
87                polars_bail!(ShapeMismatch: "shapes of `self`, `mask` and `other` are not suitable for `zip_with` operation")
88            },
89        };
90
91        Ok(Self::new(self.name().clone(), len).into_series())
92    }
93
94    fn into_total_eq_inner<'a>(&'a self) -> Box<dyn TotalEqInner + 'a> {
95        IntoTotalEqInner::into_total_eq_inner(self)
96    }
97    fn into_total_ord_inner<'a>(&'a self) -> Box<dyn TotalOrdInner + 'a> {
98        IntoTotalOrdInner::into_total_ord_inner(self)
99    }
100
101    fn subtract(&self, _rhs: &Series) -> PolarsResult<Series> {
102        null_arithmetic(self, _rhs, "subtract")
103    }
104
105    fn add_to(&self, _rhs: &Series) -> PolarsResult<Series> {
106        null_arithmetic(self, _rhs, "add_to")
107    }
108    fn multiply(&self, _rhs: &Series) -> PolarsResult<Series> {
109        null_arithmetic(self, _rhs, "multiply")
110    }
111    fn divide(&self, _rhs: &Series) -> PolarsResult<Series> {
112        null_arithmetic(self, _rhs, "divide")
113    }
114    fn remainder(&self, _rhs: &Series) -> PolarsResult<Series> {
115        null_arithmetic(self, _rhs, "remainder")
116    }
117
118    #[cfg(feature = "algorithm_group_by")]
119    fn group_tuples(&self, _multithreaded: bool, _sorted: bool) -> PolarsResult<GroupsType> {
120        Ok(if self.is_empty() {
121            GroupsType::default()
122        } else {
123            GroupsType::new_slice(vec![[0, self.length]], false, true)
124        })
125    }
126
127    #[cfg(feature = "algorithm_group_by")]
128    unsafe fn agg_list(&self, groups: &GroupsType) -> Series {
129        AggList::agg_list(self, groups)
130    }
131
132    fn _get_flags(&self) -> StatisticsFlags {
133        StatisticsFlags::empty()
134    }
135
136    fn vec_hash(
137        &self,
138        random_state: PlSeedableRandomStateQuality,
139        buf: &mut Vec<u64>,
140    ) -> PolarsResult<()> {
141        VecHash::vec_hash(self, random_state, buf)?;
142        Ok(())
143    }
144
145    fn vec_hash_combine(
146        &self,
147        build_hasher: PlSeedableRandomStateQuality,
148        hashes: &mut [u64],
149    ) -> PolarsResult<()> {
150        VecHash::vec_hash_combine(self, build_hasher, hashes)?;
151        Ok(())
152    }
153
154    fn arg_sort_multiple(
155        &self,
156        by: &[Column],
157        options: &SortMultipleOptions,
158    ) -> PolarsResult<IdxCa> {
159        let vals = (0..self.len())
160            .map(|i| (i as IdxSize, NonNull(())))
161            .collect();
162        arg_sort_multiple_impl(vals, by, options)
163    }
164}
165
166fn null_arithmetic(lhs: &NullChunked, rhs: &Series, op: &str) -> PolarsResult<Series> {
167    let output_len = match (lhs.len(), rhs.len()) {
168        (1, len_r) => len_r,
169        (len_l, 1) => len_l,
170        (len_l, len_r) if len_l == len_r => len_l,
171        _ => polars_bail!(ComputeError: "Cannot {:?} two series of different lengths.", op),
172    };
173    Ok(NullChunked::new(lhs.name().clone(), output_len).into_series())
174}
175
176impl SeriesTrait for NullChunked {
177    fn name(&self) -> &PlSmallStr {
178        &self.name
179    }
180
181    fn rename(&mut self, name: PlSmallStr) {
182        self.name = name
183    }
184
185    fn chunks(&self) -> &Vec<ArrayRef> {
186        &self.chunks
187    }
188    unsafe fn chunks_mut(&mut self) -> &mut Vec<ArrayRef> {
189        &mut self.chunks
190    }
191
192    fn chunk_lengths(&self) -> ChunkLenIter<'_> {
193        self.chunks.iter().map(|chunk| chunk.len())
194    }
195
196    fn take(&self, indices: &IdxCa) -> PolarsResult<Series> {
197        Ok(NullChunked::new(self.name.clone(), indices.len()).into_series())
198    }
199
200    unsafe fn take_unchecked(&self, indices: &IdxCa) -> Series {
201        NullChunked::new(self.name.clone(), indices.len()).into_series()
202    }
203
204    fn take_slice(&self, indices: &[IdxSize]) -> PolarsResult<Series> {
205        Ok(NullChunked::new(self.name.clone(), indices.len()).into_series())
206    }
207
208    unsafe fn take_slice_unchecked(&self, indices: &[IdxSize]) -> Series {
209        NullChunked::new(self.name.clone(), indices.len()).into_series()
210    }
211
212    fn deposit(&self, validity: &Bitmap) -> Series {
213        assert_eq!(validity.set_bits(), 0);
214        self.clone().into_series()
215    }
216
217    fn len(&self) -> usize {
218        self.length as usize
219    }
220
221    fn has_nulls(&self) -> bool {
222        !self.is_empty()
223    }
224
225    fn rechunk(&self) -> Series {
226        NullChunked::new(self.name.clone(), self.len()).into_series()
227    }
228
229    fn drop_nulls(&self) -> Series {
230        NullChunked::new(self.name.clone(), 0).into_series()
231    }
232
233    fn cast(&self, dtype: &DataType, _cast_options: CastOptions) -> PolarsResult<Series> {
234        Ok(Series::full_null(self.name.clone(), self.len(), dtype))
235    }
236
237    fn null_count(&self) -> usize {
238        self.len()
239    }
240
241    #[cfg(feature = "algorithm_group_by")]
242    fn unique(&self) -> PolarsResult<Series> {
243        let ca = NullChunked::new(self.name.clone(), self.n_unique().unwrap());
244        Ok(ca.into_series())
245    }
246
247    #[cfg(feature = "algorithm_group_by")]
248    fn n_unique(&self) -> PolarsResult<usize> {
249        let n = if self.is_empty() { 0 } else { 1 };
250        Ok(n)
251    }
252
253    #[cfg(feature = "algorithm_group_by")]
254    fn arg_unique(&self) -> PolarsResult<IdxCa> {
255        let idxs: Vec<IdxSize> = (0..self.n_unique().unwrap() as IdxSize).collect();
256        Ok(IdxCa::new(self.name().clone(), idxs))
257    }
258
259    fn unique_id(&self) -> PolarsResult<(IdxSize, Vec<IdxSize>)> {
260        if self.is_empty() {
261            Ok((0, Vec::new()))
262        } else {
263            Ok((1, vec![0; self.len()]))
264        }
265    }
266
267    fn new_from_index(&self, _index: usize, length: usize) -> Series {
268        NullChunked::new(self.name.clone(), length).into_series()
269    }
270
271    unsafe fn get_unchecked(&self, _index: usize) -> AnyValue<'_> {
272        AnyValue::Null
273    }
274
275    fn slice(&self, offset: i64, length: usize) -> Series {
276        let (chunks, len) = chunkops::slice(&self.chunks, offset, length, self.len());
277        NullChunked {
278            name: self.name.clone(),
279            length: len as IdxSize,
280            chunks,
281        }
282        .into_series()
283    }
284
285    fn split_at(&self, offset: i64) -> (Series, Series) {
286        let (l, r) = chunkops::split_at(self.chunks(), offset, self.len());
287        (
288            NullChunked {
289                name: self.name.clone(),
290                length: l.iter().map(|arr| arr.len() as IdxSize).sum(),
291                chunks: l,
292            }
293            .into_series(),
294            NullChunked {
295                name: self.name.clone(),
296                length: r.iter().map(|arr| arr.len() as IdxSize).sum(),
297                chunks: r,
298            }
299            .into_series(),
300        )
301    }
302
303    fn sort_with(&self, _options: SortOptions) -> PolarsResult<Series> {
304        Ok(self.clone().into_series())
305    }
306
307    fn arg_sort(&self, _options: SortOptions) -> IdxCa {
308        IdxCa::from_vec(self.name().clone(), (0..self.len() as IdxSize).collect())
309    }
310
311    fn is_null(&self) -> BooleanChunked {
312        BooleanChunked::full(self.name().clone(), true, self.len())
313    }
314
315    fn is_not_null(&self) -> BooleanChunked {
316        BooleanChunked::full(self.name().clone(), false, self.len())
317    }
318
319    fn reverse(&self) -> Series {
320        self.clone().into_series()
321    }
322
323    fn filter(&self, filter: &BooleanChunked) -> PolarsResult<Series> {
324        let len = if self.is_empty() {
325            // We still allow a length of `1` because it could be `lit(true)`.
326            polars_ensure!(filter.len() <= 1, ShapeMismatch: "filter's length: {} differs from that of the series: 0", filter.len());
327            0
328        } else if filter.len() == 1 {
329            return match filter.get(0) {
330                Some(true) => Ok(self.clone().into_series()),
331                None | Some(false) => Ok(NullChunked::new(self.name.clone(), 0).into_series()),
332            };
333        } else {
334            polars_ensure!(filter.len() == self.len(), ShapeMismatch: "filter's length: {} differs from that of the series: {}", filter.len(), self.len());
335            filter.sum().unwrap_or(0) as usize
336        };
337        Ok(NullChunked::new(self.name.clone(), len).into_series())
338    }
339
340    fn shift(&self, _periods: i64) -> Series {
341        self.clone().into_series()
342    }
343
344    fn sum_reduce(&self) -> PolarsResult<Scalar> {
345        Ok(Scalar::null(DataType::Null))
346    }
347
348    fn min_reduce(&self) -> PolarsResult<Scalar> {
349        Ok(Scalar::null(DataType::Null))
350    }
351
352    fn max_reduce(&self) -> PolarsResult<Scalar> {
353        Ok(Scalar::null(DataType::Null))
354    }
355
356    fn mean_reduce(&self) -> PolarsResult<Scalar> {
357        Ok(Scalar::null(DataType::Null))
358    }
359
360    fn median_reduce(&self) -> PolarsResult<Scalar> {
361        Ok(Scalar::null(DataType::Null))
362    }
363
364    fn std_reduce(&self, _ddof: u8) -> PolarsResult<Scalar> {
365        Ok(Scalar::null(DataType::Null))
366    }
367
368    fn var_reduce(&self, _ddof: u8) -> PolarsResult<Scalar> {
369        Ok(Scalar::null(DataType::Null))
370    }
371
372    fn append(&mut self, other: &Series) -> PolarsResult<()> {
373        polars_ensure!(other.dtype() == &DataType::Null, ComputeError: "expected null dtype");
374        // we don't create a new null array to keep probability of aligned chunks higher
375        self.length += other.len() as IdxSize;
376        self.chunks.extend(other.chunks().iter().cloned());
377        Ok(())
378    }
379    fn append_owned(&mut self, mut other: Series) -> PolarsResult<()> {
380        polars_ensure!(other.dtype() == &DataType::Null, ComputeError: "expected null dtype");
381        // we don't create a new null array to keep probability of aligned chunks higher
382        let other: &mut NullChunked = other._get_inner_mut().as_any_mut().downcast_mut().unwrap();
383        self.length += other.len() as IdxSize;
384        self.chunks.extend(std::mem::take(&mut other.chunks));
385        Ok(())
386    }
387
388    fn extend(&mut self, other: &Series) -> PolarsResult<()> {
389        *self = NullChunked::new(self.name.clone(), self.len() + other.len());
390        Ok(())
391    }
392
393    #[cfg(feature = "approx_unique")]
394    fn approx_n_unique(&self) -> PolarsResult<IdxSize> {
395        Ok(if self.is_empty() { 0 } else { 1 })
396    }
397
398    fn clone_inner(&self) -> Arc<dyn SeriesTrait> {
399        Arc::new(self.clone())
400    }
401
402    fn find_validity_mismatch(&self, other: &Series, idxs: &mut Vec<IdxSize>) {
403        ChunkNestingUtils::find_validity_mismatch(self, other, idxs)
404    }
405
406    fn as_any(&self) -> &dyn Any {
407        self
408    }
409
410    fn as_any_mut(&mut self) -> &mut dyn Any {
411        self
412    }
413
414    fn as_phys_any(&self) -> &dyn Any {
415        self
416    }
417
418    fn as_arc_any(self: Arc<Self>) -> Arc<dyn Any + Send + Sync> {
419        self as _
420    }
421}
422
423unsafe impl IntoSeries for NullChunked {
424    fn into_series(self) -> Series
425    where
426        Self: Sized,
427    {
428        Series(Arc::new(self))
429    }
430}
polars_core/series/implementations/null.rs

polars_core/series/implementations/
null.rs
