feat!: Update for polars 1.0.0

[etiennebacher]

No description provided.

[eitsupi]

Thanks for working on this. Some thoughts:

• compute() and collect() can be rewritten with as_polars_df() and as_tibble(), and ... can be passed to these functions.
• I recall that I had not implemented <lazyframe>$`_fetch`(). I will add it later. This is completely different from <lazyframe>$head().

[etiennebacher]

compute() and collect() can be rewritten with as_polars_df() and as_tibble(), and ... can be passed to these functions.

Thanks, but that doesn't simplify much IMO.

I recall that I had not implemented <lazyframe>$`_fetch`(). I will add it later. This is completely different from <lazyframe>$head().

This was deprecated one year ago, which is why it wasn't implemented in the rewrite. The deprecation message recommended using head + collect, hence my implementation: https://github.com/pola-rs/polars/pull/17278/files#diff-6f7ee6a0f7f99cb158f4d68d883a118b9fab6175602e8696739d900c8b0d864f

[eitsupi]

Thanks, but that doesn't simplify much IMO.

There are several additional arguments that are passed to.

hence my implementation

If so, why is it not marked as deprecated?

[etiennebacher]

If so, why is it not marked as deprecated?

It was an oversight on my side, I thought this was part of the functions exported by dplyr (like collect and compute) but it actually isn't, thanks for noticing.

[etiennebacher]

@eitsupi currently collect() checks empty dots. I'd like to allow passing arguments in the dots so that I don't have to keep the documentation up to date and can just say something like

... accepts all arguments to specify the conversion from polars to R. See polars:::?as_tibble.polars_lazy_frame() for accepted arguments.

The problem with this approach is that wrong arguments would be undetected since as_tibble.polars_lazy_frame() doesn't perform checks on dots:

library(polars)
library(tibble)

pl$LazyFrame(x = 1) |> as_tibble(foo = 2)
#> # A tibble: 1 × 1
#>       x
#>   <dbl>
#> 1     1

Should we add a check like that on as_tibble() for LazyFrame and DataFrame in polars?

---

Edit: in the meantime, I've manually added the arguments and point to ?polars:::as.data.frame.polars_lazy_frame in the docs.

[eitsupi]

@etiennebacher In my opinion, S3 methods should simply ignore empty dots. Otherwise, it becomes difficult to write common processes for different classes.
So <lazyframe>$collect() checks for the dots, but as_polars_df(<lazyframe>) simply ignores the dots.

As I wrote before, consider making compute() and collect() just wrappers around as_polars_df() and as_tibble() documenting dots with @inheritDotParams lazyframe__collect.
https://roxygen2.r-lib.org/articles/reuse.html#automatically-copy-tags

[etiennebacher]

I get an error with @inheritDotParams polars:::lazyframe__collect:

Error in FUN():
! argument "source" is missing, with no default

Not sure this works with functions accessed via :::.

In any case, I want to keep the dots check in collect(), so I won't just pass the dots around. Regarding compute(), I don't see any args in polars:::as_polars_df.polars_lazy_frame() that are not supported here so I'll just leave it like this.

Thanks for your help! I think this is ready for 1.0.0 now.

[eitsupi]

Sorry, I didn't understand the inheritance of arguments well.
As you say, lazyframe__collect is not a public function, so it cannot be inherited.

[etiennebacher]

Blocked by pola-rs/r-polars#1439, there's a substantial performance decrease on one of my test data (from 10s to 48s).

[eitsupi]

(from 10s to 48s)

Can such a large slowdown be explained by pl$col alone?
What about the results of profvis::profvis?

[etiennebacher]

Not all of it, but a pretty large part. Here's the flamegraph with 0.22.4:

and with 1.0.0:

---

(It's 36s on the picture and not 48s anymore, but the point is pl$col() is the culprit)

[eitsupi]

I see, I don't think this is a blocker anyway (it's caused by something already released upstream and there's nothing we can do about it here)

[etiennebacher]

Right, I can merge this now that the installation of polars from r-multiverse points to 1.0.0