This allows performance debugging to be turned on by setting:
```nushell
$env.POLARS_PLUGIN_PERF = "true"
```
Furthermore, this improves the other plugin debugging by allowing the
env variable for debugging to be set at any time versus having to be
available when nushell is launched:
```nushell
$env.POLARS_PLUGIN_DEBUG = "true"
```
This plugin introduces a `perf` function that will output timing
results. This works very similar to the perf function available in
nu_utils::utils::perf. This version prints everything to std error to
not break the plugin stream and uses the engine interface to see if the
env variable is configured.
This pull requests uses this `perf` function when:
* opening csv files as dataframes
* opening json lines files as dataframes
This will hopefully help provide some more fine grained information on
how long it takes polars to open different dataframes. The `perf` can
also be utilized later for other dataframes use cases.
Per discussion on discord dataframes channel with @maxim-uvarov and pyz.
When converting a dataframe to an nushell value via `polars into-nu`,
the index column should not be added by default and should only be added
when specifying `--index`
As reported by @maxim-uvarov and pyz in the dataframes discord channel:
```nushell
[[a b]; [1 1] [1 2] [2 1] [2 2] [3 1] [3 2]] | polars into-df | polars with-column ((polars col a) / (polars col b)) --name c
× Type mismatch.
╭─[entry #45:1:102]
1 │ [[a b]; [1 1] [1 2] [2 1] [2 2] [3 1] [3 2]] | polars into-df | polars with-column ((polars col a) / (polars col b)) --name c
· ───────┬──────
· ╰── Right hand side not a dataframe expression
╰────
```
This pull request corrects the type casting on the right hand side and
allows more than just polars literal expressions.
# Description
@maxim-uvarov did a ton of research and work with the dply-rs author and
ritchie from polars and found out that the allocator matters on macos
and it seems to be what was messing up the performance of polars plugin.
ritchie suggested to use jemalloc but i switched it to mimalloc to match
nushell and it seems to run better.
## Before (default allocator)
note - using 1..10 vs 1..100 since it takes so long. also notice how
high the `max` timings are compared to mimalloc below.
```nushell
❯ 1..10 | each {timeit {polars open Data7602DescendingYearOrder.csv | polars group-by year | polars agg (polars col geo_count | polars sum) | polars collect | null}} | | {mean: ($in | math avg), min: ($in | math min), max: ($in | math max), stddev: ($in | into int | into float | math stddev | into int | $'($in)ns' | into duration)}
╭────────┬─────────────────────────╮
│ mean │ 4sec 999ms 605µs 995ns │
│ min │ 983ms 627µs 42ns │
│ max │ 13sec 398ms 135µs 791ns │
│ stddev │ 3sec 476ms 479µs 939ns │
╰────────┴─────────────────────────╯
❯ use std bench
❯ bench { polars open Data7602DescendingYearOrder.csv | polars group-by year | polars agg (polars col geo_count | polars sum) | polars collect | null } -n 10
╭───────┬────────────────────────╮
│ mean │ 6sec 220ms 783µs 983ns │
│ min │ 1sec 184ms 997µs 708ns │
│ max │ 18sec 882ms 81µs 708ns │
│ std │ 5sec 350ms 375µs 697ns │
│ times │ [list 10 items] │
╰───────┴────────────────────────╯
```
## After (using mimalloc)
```nushell
❯ 1..100 | each {timeit {polars open Data7602DescendingYearOrder.csv | polars group-by year | polars agg (polars col geo_count | polars sum) | polars collect | null}} | | {mean: ($in | math avg), min: ($in | math min), max: ($in | math max), stddev: ($in | into int | into float | math stddev | into int | $'($in)ns' | into duration)}
╭────────┬───────────────────╮
│ mean │ 103ms 728µs 902ns │
│ min │ 97ms 107µs 42ns │
│ max │ 149ms 430µs 84ns │
│ stddev │ 5ms 690µs 664ns │
╰────────┴───────────────────╯
❯ use std bench
❯ bench { polars open Data7602DescendingYearOrder.csv | polars group-by year | polars agg (polars col geo_count | polars sum) | polars collect | null } -n 100
╭───────┬───────────────────╮
│ mean │ 103ms 620µs 195ns │
│ min │ 97ms 541µs 166ns │
│ max │ 130ms 262µs 166ns │
│ std │ 4ms 948µs 654ns │
│ times │ [list 100 items] │
╰───────┴───────────────────╯
```
## After (using jemalloc - just for comparison)
```nushell
❯ 1..100 | each {timeit {polars open Data7602DescendingYearOrder.csv | polars group-by year | polars agg (polars col geo_count | polars sum) | polars collect | null}} | | {mean: ($in | math avg), min: ($in | math min), max: ($in | math max), stddev: ($in | into int | into float | math stddev | into int | $'($in)ns' | into duration)}
╭────────┬───────────────────╮
│ mean │ 113ms 939µs 777ns │
│ min │ 108ms 337µs 333ns │
│ max │ 166ms 467µs 458ns │
│ stddev │ 6ms 175µs 618ns │
╰────────┴───────────────────╯
❯ use std bench
❯ bench { polars open Data7602DescendingYearOrder.csv | polars group-by year | polars agg (polars col geo_count | polars sum) | polars collect | null } -n 100
╭───────┬───────────────────╮
│ mean │ 114ms 363µs 530ns │
│ min │ 108ms 804µs 833ns │
│ max │ 143ms 521µs 459ns │
│ std │ 5ms 88µs 56ns │
│ times │ [list 100 items] │
╰───────┴───────────────────╯
```
## After (using parquet + mimalloc)
```nushell
❯ 1..100 | each {timeit {polars open data.parquet | polars group-by year | polars agg (polars col geo_count | polars sum) | polars collect | null}} | | {mean: ($in | math avg), min: ($in | math min), max: ($in | math max), stddev: ($in | into int | into float | math stddev | into int | $'($in)ns' | into duration)}
╭────────┬──────────────────╮
│ mean │ 34ms 255µs 492ns │
│ min │ 31ms 787µs 250ns │
│ max │ 76ms 408µs 416ns │
│ stddev │ 4ms 472µs 916ns │
╰────────┴──────────────────╯
❯ use std bench
❯ bench { polars open data.parquet | polars group-by year | polars agg (polars col geo_count | polars sum) | polars collect | null } -n 100
╭───────┬──────────────────╮
│ mean │ 34ms 897µs 562ns │
│ min │ 31ms 518µs 542ns │
│ max │ 65ms 943µs 625ns │
│ std │ 3ms 450µs 741ns │
│ times │ [list 100 items] │
╰───────┴──────────────────╯
```
# User-Facing Changes
<!-- List of all changes that impact the user experience here. This
helps us keep track of breaking changes. -->
# Tests + Formatting
<!--
Don't forget to add tests that cover your changes.
Make sure you've run and fixed any issues with these commands:
- `cargo fmt --all -- --check` to check standard code formatting (`cargo
fmt --all` applies these changes)
- `cargo clippy --workspace -- -D warnings -D clippy::unwrap_used` to
check that you're using the standard code style
- `cargo test --workspace` to check that all tests pass (on Windows make
sure to [enable developer
mode](https://learn.microsoft.com/en-us/windows/apps/get-started/developer-mode-features-and-debugging))
- `cargo run -- -c "use toolkit.nu; toolkit test stdlib"` to run the
tests for the standard library
> **Note**
> from `nushell` you can also use the `toolkit` as follows
> ```bash
> use toolkit.nu # or use an `env_change` hook to activate it
automatically
> toolkit check pr
> ```
-->
# After Submitting
<!-- If your PR had any user-facing changes, update [the
documentation](https://github.com/nushell/nushell.github.io) after the
PR is merged, if necessary. This will help us keep the docs up to date.
-->
This reverts commit 68adc4657f.
# Description
Reverts the lazyframe refactor (#12669) for the next release, since
there are still a few lingering issues. This temporarily solves #12863
and #12828. After the release, the lazyframes can be added back and
cleaned up.
# Description
Removes the old `nu-cmd-dataframe` crate in favor of the polars plugin.
As such, this PR also removes the `dataframe` feature, related CI, and
full releases of nushell.
# Description
This PR adds a few functions to `Span` for merging spans together:
- `Span::append`: merges two spans that are known to be in order.
- `Span::concat`: returns a span that encompasses all the spans in a
slice. The spans must be in order.
- `Span::merge`: merges two spans (no order necessary).
- `Span::merge_many`: merges an iterator of spans into a single span (no
order necessary).
These are meant to replace the free-standing `nu_protocol::span`
function.
The spans in a `LiteCommand` (the `parts`) should always be in order
based on the lite parser and lexer. So, the parser code sees the most
usage of `Span::append` and `Span::concat` where the order is known. In
other code areas, `Span::merge` and `Span::merge_many` are used since
the order between spans is often not known.
# Description
This PR introduces a `ByteStream` type which is a `Read`-able stream of
bytes. Internally, it has an enum over three different byte stream
sources:
```rust
pub enum ByteStreamSource {
Read(Box<dyn Read + Send + 'static>),
File(File),
Child(ChildProcess),
}
```
This is in comparison to the current `RawStream` type, which is an
`Iterator<Item = Vec<u8>>` and has to allocate for each read chunk.
Currently, `PipelineData::ExternalStream` serves a weird dual role where
it is either external command output or a wrapper around `RawStream`.
`ByteStream` makes this distinction more clear (via `ByteStreamSource`)
and replaces `PipelineData::ExternalStream` in this PR:
```rust
pub enum PipelineData {
Empty,
Value(Value, Option<PipelineMetadata>),
ListStream(ListStream, Option<PipelineMetadata>),
ByteStream(ByteStream, Option<PipelineMetadata>),
}
```
The PR is relatively large, but a decent amount of it is just repetitive
changes.
This PR fixes#7017, fixes#10763, and fixes#12369.
This PR also improves performance when piping external commands. Nushell
should, in most cases, have competitive pipeline throughput compared to,
e.g., bash.
| Command | Before (MB/s) | After (MB/s) | Bash (MB/s) |
| -------------------------------------------------- | -------------:|
------------:| -----------:|
| `throughput \| rg 'x'` | 3059 | 3744 | 3739 |
| `throughput \| nu --testbin relay o> /dev/null` | 3508 | 8087 | 8136 |
# User-Facing Changes
- This is a breaking change for the plugin communication protocol,
because the `ExternalStreamInfo` was replaced with `ByteStreamInfo`.
Plugins now only have to deal with a single input stream, as opposed to
the previous three streams: stdout, stderr, and exit code.
- The output of `describe` has been changed for external/byte streams.
- Temporary breaking change: `bytes starts-with` no longer works with
byte streams. This is to keep the PR smaller, and `bytes ends-with`
already does not work on byte streams.
- If a process core dumped, then instead of having a `Value::Error` in
the `exit_code` column of the output returned from `complete`, it now is
a `Value::Int` with the negation of the signal number.
# After Submitting
- Update docs and book as necessary
- Release notes (e.g., plugin protocol changes)
- Adapt/convert commands to work with byte streams (high priority is
`str length`, `bytes starts-with`, and maybe `bytes ends-with`).
- Refactor the `tee` code, Devyn has already done some work on this.
---------
Co-authored-by: Devyn Cairns <devyn.cairns@gmail.com>
Fix for #12730
All of the code expected a list of floats, but the syntax shape expected
a table. Resolved by changing the syntax shape to list of floats.
cc: @maxim-uvarov
This moves to predominantly supporting only lazy dataframes for most
operations. It removes a lot of the type conversion between lazy and
eager dataframes based on what was inputted into the command.
For the most part the changes will mean:
* You will need to run `polars collect` after performing operations
* The into-lazy command has been removed as it is redundant.
* When opening files a lazy frame will be outputted by default if the
reader supports lazy frames
A list of individual command changes can be found
[here](https://hackmd.io/@nucore/Bk-3V-hW0)
---------
Co-authored-by: Ian Manske <ian.manske@pm.me>
# Description
I added some more tests to our mighty `polars` ~~, yet I don't know how
to add expected results in some of them. I would like to ask for help.~~
~~My experiments are in the last commit: [polars:
experiments](f7e5e72019).
Without those experiments `cargo test` goes well.~~
UPD. I moved out my unsuccessful test experiments into a separate
[branch](https://github.com/maxim-uvarov/nushell/blob/polars-tests-broken2/).
So, this branch seems ready for a merge.
@ayax79, maybe you'll find time for me please? It's not urgent for sure.
P.S. I'm very new to git. Please feel free to give me any suggestions on
how I should use it better
# Description
I would like to help with `polars` plugin development and add tests to
all the `polars` command's existing params.
Since I have never written any lines of Rust, even though the task of
creating tests is relatively simple, I would like to ask for feedback to
ensure I did everything correctly here.
# Description
Continuing from #12568, this PR further reduces the size of `Expr` from
64 to 40 bytes. It also reduces `Expression` from 128 to 96 bytes and
`Type` from 32 to 24 bytes.
This was accomplished by:
- for `Expr` with multiple fields (e.g., `Expr::Thing(A, B, C)`),
merging the fields into new AST struct types and then boxing this struct
(e.g. `Expr::Thing(Box<ABC>)`).
- replacing `Vec<T>` with `Box<[T]>` in multiple places. `Expr`s and
`Expression`s should rarely be mutated, if at all, so this optimization
makes sense.
By reducing the size of these types, I didn't notice a large performance
improvement (at least compared to #12568). But this PR does reduce the
memory usage of nushell. My config is somewhat light so I only noticed a
difference of 1.4MiB (38.9MiB vs 37.5MiB).
---------
Co-authored-by: Stefan Holderbach <sholderbach@users.noreply.github.com>
# Description
This pull request provides three new commands:
`polars store-ls` - moved from `polars ls`. It provides the list of all
object stored in the plugin cache
`polars store-rm` - deletes a cached object
`polars store-get` - gets an object from the cache.
The addition of `polars store-get` required adding a reference_count to
cached entries. `polars get` is the only command that will increment
this value. `polars rm` will remove the value despite it's count. Calls
to PolarsPlugin::custom_value_dropped will decrement the value.
The prefix store- was chosen due to there already being a `polars cache`
command. These commands were not made sub-commands as there isn't a way
to display help for sub commands in plugins (e.g. `polars store`
displaying help) and I felt the store- seemed fine anyways.
The output of `polars store-ls` now shows the reference count for each
object.
# User-Facing Changes
polars ls has now moved to polars store-ls
---------
Co-authored-by: Jack Wright <jack.wright@disqo.com>
# Description
The polars dtype command is largerly redundant since the introduction of
the schema command. The schema command also has the added benefit that
it's output can be used as a parameter to other schema commands:
```nushell
[[a b]; [5 6] [5 7]] | polars into-df -s ($df | polars schema
```
# User-Facing Changes
`polars dtypes` has been removed. Users should use `polars schema`
instead.
Co-authored-by: Jack Wright <jack.wright@disqo.com>
I had previously changed NuLazyFrame::collect to set the NuDataFrame's
from_lazy field to false to prevent conversion back to a lazy frame. It
appears there are cases where this should happen. Instead, I am only
setting from_lazy=false inside the `polars collect` command.
[Related discord
message](https://discord.com/channels/601130461678272522/1227612017171501136/1230600465159421993)
Co-authored-by: Jack Wright <jack.wright@disqo.com>
# Description
This is just some cleanup. I moved to_pipeline_data and to_cache_value
to the CustomValueSupport trait, where I should've put them to begin
with.
Co-authored-by: Jack Wright <jack.wright@disqo.com>
# Description
@maxim-uvarov discovered the following error:
```
> [[a b]; [6 2] [1 4] [4 1]] | polars into-lazy | polars sort-by a | polars unique --subset [a]
Error: × Error using as series
╭─[entry #1:1:68]
1 │ [[a b]; [6 2] [1 4] [4 1]] | polars into-lazy | polars sort-by a | polars unique --subset [a]
· ──────┬──────
· ╰── dataframe has more than one column
╰────
```
During investigation, I discovered the root cause was that the lazy frame was incorrectly converted back to a eager dataframe. In order to keep this from happening, I explicitly set that the dataframe did not come from an eager frame. This causes the conversion logic to not attempt to convert the dataframe later in the pipeline.
---------
Co-authored-by: Jack Wright <jack.wright@disqo.com>
# Description
This adds a `SharedCow` type as a transparent copy-on-write pointer that
clones to unique on mutate.
As an initial test, the `Record` within `Value::Record` is shared.
There are some pretty big wins for performance. I'll post benchmark
results in a comment. The biggest winner is nested access, as that would
have cloned the records for each cell path follow before and it doesn't
have to anymore.
The reusability of the `SharedCow` type is nice and I think it could be
used to clean up the previous work I did with `Arc` in `EngineState`.
It's meant to be a mostly transparent clone-on-write that just clones on
`.to_mut()` or `.into_owned()` if there are actually multiple
references, but avoids cloning if the reference is unique.
# User-Facing Changes
- `Value::Record` field is a different type (plugin authors)
# Tests + Formatting
- 🟢 `toolkit fmt`
- 🟢 `toolkit clippy`
- 🟢 `toolkit test`
- 🟢 `toolkit test stdlib`
# After Submitting
- [ ] use for `EngineState`
- [ ] use for `Value::List`
# Description
From @maxim-uvarov's
[post](https://discord.com/channels/601130461678272522/1227612017171501136/1228656319704203375).
When calling `to-lazy` back to back in a pipeline, an error should not
occur:
```
> [[a b]; [6 2] [1 4] [4 1]] | polars into-lazy | polars into-lazy
Error: nu:🐚:cant_convert
× Can't convert to NuDataFrame.
╭─[entry #1:1:30]
1 │ [[a b]; [6 2] [1 4] [4 1]] | polars into-lazy | polars into-lazy
· ────────┬───────
· ╰── can't convert NuLazyFrameCustomValue to NuDataFrame
╰────
```
This pull request ensures that custom value's of NuLazyFrameCustomValue are properly converted when passed in.
Co-authored-by: Jack Wright <jack.wright@disqo.com>
# Description
@maxim-uvarov discovered an issue with the current implementation. When
executing [[index a]; [1 1]] | polars into-df, a plugin_failed_to_decode
error occurs. This happens because a Record is created with two columns
named "index" as an index column is added during conversion. This pull
request addresses the problem by not adding an index column if there is
already a column named "index" in the dataframe.
---------
Co-authored-by: Jack Wright <jack.wright@disqo.com>
# Description
All polars commands that output a file were not handling relative paths
correctly.
A command like
``` [[a b]; [6 2] [1 4] [4 1]] | polars into-df | polars to-parquet foo.json```
was outputting the foo.json to the directory of the plugin executable.
This pull request pulls in nu-path and using it for resolving the file paths.
Related discussion
https://discord.com/channels/601130461678272522/1227612017171501136/1227889870358183966
# User-Facing Changes
None
# Tests + Formatting
Done, added tests for each of the polars to-* commands.
---------
Co-authored-by: Jack Wright <jack.wright@disqo.com>
# Description
`polars ls` is already different that `dfr ls`. Currently it just shows
the cache key, columns, rows, and type. I have added:
- creation time
- size
- span contents
- span start and end
<img width="1471" alt="Screenshot 2024-04-10 at 17 27 06"
src="https://github.com/nushell/nushell/assets/56345/545918b7-7c96-4c25-bc01-b9e2b659a408">
# Tests + Formatting
Done
Co-authored-by: Jack Wright <jack.wright@disqo.com>
