feat: Implement join on struct dtype

[lukemanley]

Join on Struct dtype via row encoding.

closes #15500
closes #10721

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 79.28%. Comparing base (036ee1e) to head (b27663a).
Report is 4 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #21093      +/-   ##
==========================================
+ Coverage   79.20%   79.28%   +0.08%     
==========================================
  Files        1589     1589              
  Lines      226488   226527      +39     
  Branches     2594     2594              
==========================================
+ Hits       179399   179613     +214     
+ Misses      46493    46318     -175     
  Partials      596      596              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[coastalwhite]

Nice, I think we can use no_order=true for the row encoding. That allows a few optimizations that are not allowed normally.

[lukemanley]

Nice, I think we can use no_order=true for the row encoding. That allows a few optimizations that are not allowed normally.

Thanks. Let me know if this is what you had in mind.

Also, the how="full" join is failing for new_streaming. It seems to be failing this check:

polars/crates/polars-core/src/frame/mod.rs

Line 520 in 036ee1e

assert_eq!(col.len(), height);

It looks like col is a scalar column in this case. Any suggestions on that?

[ritchie46]

Also, the how="full" join is failing for new_streaming. It seems to be failing this check:

Hmm, I think we should take a look at the backtrace for that one. It might be a bug.

[lukemanley]

Hmm, I think we should take a look at the backtrace for that one. It might be a bug.

I'm a bit stuck on this one. The following succeeds about 50% of the time and fails the other 50%:

lhs = pl.LazyFrame(
    {
        "a": [{"x": 1}, {"x": 2}],
        "b": [1, 2],
    }
)
rhs = pl.LazyFrame(
    {
        "a": [{"x": 1}],
        "c": [1],
    }
)
lhs.join(rhs, on=["a"], how="inner").collect(new_streaming=True)