Group By no longer supported in Rust WASM

[brettsaunders21]

Checks

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of Polars.

Reproducible example

df.group_by(&[col("category"), col("type")]).head(1)

Log output

only supported on 64bit arch

Issue description

With the changes to Polars to make group by use row encoding, it seems that this required 64bit support which Web Assembly doesn't. It seems like it isn't possible to use group_by anymore in WASM

Expected behavior

Carry out group_by operations when compiling polars (with default feature false, lazy true) for WASM and use it in browser with wasm-bindgen

Installed versions

0.41.2

polars = { version = "0.41.2", default-features = false, features = [ 'fmt_no_tty', 'lazy', ] }

[ritchie46]

Hmm... The usize should be changed to u64.

[daBlesr]

Hi @ritchie46, not sure if I understand, usize is architecture agnostic right? Which 'problematic' use of usize are you exactly referring to in the code?

[daBlesr]

I see that the offsets vector in RowsEncoded is casted later to i64 (after checking the types) and this causes the problem. Is your suggestion to change the type of offsets to Vec<u64>, such that they take the same number of bytes on memory?

[daBlesr]

I quickly checked the impact of the above, and the code becomes a bit less appealing using u64 as it will also require changes on encode.rs casting usize to u64. What do you think of the following, where offset<usize> (naturally) remains as-is:

let offsets = if std::mem::size_of::<usize>() % std::mem::size_of::<i64>() == 0 {
    // We can safely convert usize to i64
    bytemuck::cast_vec::<usize, i64>(offsets)
} else {
    offsets.iter().map(|&i| i as i64).collect()
};

It checks if it can simply convert, or in worst-case (for wasm) it will have some performance impact because it needs to cast the encoded rows.

[brettsaunders21]

Any update on this issue?

[daBlesr]

I created a PR which waits for a review

[daBlesr]

Hi @ritchie46, please take some time to review the PR. Let me know what you are unhappy with. Without this PR WASM does not work.

[chtotut]

Any update on this issue?
@daBlesr do you continue to work for PR or have you lost interest? (just a friendly ping)