Performance improvements

[joe-clickhouse]

Summary

• Optimized DateTime and DateTime64 native read hot paths, especially for naive UTC and UTC-equivalent timezones, by moving more conversion work onto direct epoch-arithmetic paths and dedicated Cython helpers.
• Added a dedicated Cython DateTime64 naive-UTC decode loop and routed UTC-equivalent timezone reads through the same Cython C-API construction path used for naive UTC.
• Cython DateTime and DateTime64 reads now construct datetime objects directly via the CPython datetime C API, bypassing the intermediate component tuple and the Python-level datetime(...) constructor.
• Fixed Cython DateTime conversion bugs and a pre-existing leap-year bug in the epoch decoder where centuries divisible by 400 (2400, 2800, 3200, ...) were treated as non-leap. Expanded epoch/date arithmetic test coverage.
• Optimized fixed-width numeric INSERT serialization by routing native writes through a shared write_native_col helper.
• Added a fast memcpy path for compatible 1-D C-contiguous NumPy arrays in the Cython native write path.
• Optimized Map INSERT serialization with a shared build_map_columns helper and a faster Cython implementation for flattening keys/values plus offsets.
• Reduced Map read materialization overhead by avoiding an intermediate pair-tuple allocation.
• Optimized Decimal and BigDecimal reads by replacing per-row string construction with direct Decimal(int).scaleb(-scale).

Checklist

Delete items not relevant to your PR:

• Unit and integration tests covering the common scenarios were added
• A human-readable description of the changes was provided to include in CHANGELOG

[genzgd]

I don't know if we can use this: https://www.benjoffe.com/fast-date-64, but I keeping meaning to take a look.

[joe-clickhouse]

Thanks @genzgd! Great find. I'll definitely take a look.

[Copilot]

Pull request overview

This PR focuses on speeding up hot-path native reads/writes (especially temporal types and Map/numeric inserts) by shifting more work into Cython and using direct epoch arithmetic and buffer-level fast paths, while also fixing a leap-year bug in epoch decoding.

Changes:

• Optimizes DateTime/DateTime64 decoding (naive UTC + UTC-equivalent tz) via epoch arithmetic and CPython datetime C-API construction, adding supporting helpers and tests.
• Adds faster native write helpers (write_native_col, build_map_columns) including a NumPy memcpy fast path for compatible arrays.
• Reduces Map read/write materialization overhead and improves Decimal/BigDecimal read performance by avoiding per-row string building.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
tests/unit_tests/test_driver/test_epoch_arithmetic.py	Adds unit tests for epoch arithmetic helpers and UTC-equivalent timezone helpers.
clickhouse_connect/driverc/dataconv.pyx	Adds epoch-to-datetime arithmetic helpers, DateTime64 decode loops, NumPy memcpy native writes, and Map flattening helper; fixes leap-year logic.
clickhouse_connect/driverc/buffer.pyx	Optimizes varint string-length parsing and adds a UTF-8 decode fast path in string reads.
clickhouse_connect/driver/tzutil.py	Reworks utcfromtimestamp and adds arithmetic-based helpers for UTC-equivalent tz-aware datetimes and explicit microseconds.
clickhouse_connect/driver/dataconv.py	Routes UTC-equivalent timezone reads through new arithmetic helper and adds pure-Python fallbacks for new C helpers.
clickhouse_connect/datatypes/temporal.py	Routes DateTime64 reads through shared conversion helpers (UTC-equivalent vs non-UTC).
clickhouse_connect/datatypes/numeric.py	Routes native fixed-width numeric inserts through write_native_col and optimizes Decimal/BigDecimal reads.
clickhouse_connect/datatypes/container.py	Uses new build_map_columns helper and reduces Map read allocations.
CHANGELOG.md	Documents the performance improvements and bug fix in UNRELEASED notes.

[Copilot]

Pull request overview

Copilot reviewed 9 out of 9 changed files in this pull request and generated 4 comments.

[Copilot]

Pull request overview

Copilot reviewed 10 out of 10 changed files in this pull request and generated 1 comment.

[joe-clickhouse]

Also, I've looked into the https://www.benjoffe.com/fast-date-64 stuff and it's super impressive. In the clickhouse-connect code though, after the current improvements in the branch, profiling shows the dt64 read hot path bottleneck is pretty much entirely now in the datetime_new pyobject construction. Will definitely keep this around for future ref though. Thanks!