feat(python): huge speedup of scalar-to-array expansion on frame init from dict

[alexander-beedie]

Spotted the existence of Series.extend_constant and reworked PR #6034 to use it; dramatically faster.

Example:

df = pl.DataFrame({
    "a": range(50_000_000),
    "b": 1234567890,
    "c": "tuvwxyz",
    "d": None,
})

# shape: (50000000, 4)
# ┌──────────┬────────────┬─────────┬──────┐
# │ a        ┆ b          ┆ c       ┆ d    │
# │ ---      ┆ ---        ┆ ---     ┆ ---  │
# │ i64      ┆ i64        ┆ str     ┆ f64  │
# ╞══════════╪════════════╪═════════╪══════╡
# │ 0        ┆ 1234567890 ┆ tuvwxyz ┆ null │
# │ 1        ┆ 1234567890 ┆ tuvwxyz ┆ null │
# │ 2        ┆ 1234567890 ┆ tuvwxyz ┆ null │
# │ 3        ┆ 1234567890 ┆ tuvwxyz ┆ null │
#  ...

Timings:

# Before: 7.1974 secs
# After:  0.8025 secs

Also:

• Fixed scalar expansion in conjunction with columns override.
• Fixed error when trying to expand scalars in the presence of struct columns.
• Some minor typing improvements, allowing removal of some # type: ignore directives.
• Additional test coverage for all of the above.

[stinodego]

Nice! I thought there had to be a better way than to multiply a list.

[alexander-beedie]

Nice! I thought there had to be a better way than to multiply a list.

I've returned to the scene of the crime... :)