ZhongxuanWang opened a new pull request, #12345:
URL: https://github.com/apache/arrow-rs/pull/12345

   # Which issue does this PR close?
   
   - Closes #12343.
   
   # Rationale for this change
   
   `parquet` lists `chrono` as a required dependency, so even a build with 
`default-features = false` and without the `arrow` feature links it, and `cargo 
audit` / `cargo deny` report RUSTSEC-2026-0543 for every crate that depends on 
`parquet`. `parquet`'s own use of `chrono` is three display helpers in 
`record::api`, which need no more than integer arithmetic.
   
   # What changes are included in this PR?
   
   - `convert_date_to_string`, `convert_timestamp_millis_to_string` and 
`convert_timestamp_micros_to_string` convert days since the epoch to a civil 
date with Howard Hinnant's `civil_from_days` (floor division, so values before 
1970 are right) and format the fields directly. The output matches chrono's 
`%Y-%m-%d`, `%Y-%m-%d %H:%M:%S%.3f %:z` and `%Y-%m-%d %H:%M:%S%.6f %:z` for 
every value chrono accepts, including chrono's `%Y` rule of a sign before years 
outside 0..=9999 (`-0001`, `+10000`).
   - `chrono` moves from `[dependencies]` to `[dev-dependencies]`, where the 
unit tests, the `arrow_reader` integration tests and the `write_parquet` 
example still use it.
   - The doc comment on `convert_date_to_string` said the date is displayed in 
the local timezone; it is UTC.
   
   With the `arrow` feature (on by default) `chrono` still comes in through 
`arrow-array`; that is #9183 and out of scope here. With 
`--no-default-features`, `cargo tree -p parquet -e normal` no longer lists 
`chrono`, and with the default features it lists it only under `arrow-array`.
   
   # Are these changes tested?
   
   Yes. The existing `test_convert_date_to_string`, 
`test_convert_timestamp_millis_to_string` and 
`test_convert_timestamp_micros_to_string` tests pin the common case. A new 
test, `test_convert_date_and_timestamp_to_string_extreme_years`, compares all 
three helpers with chrono at year 0 (including 0000-02-29), negative years, 
year 10000, the epoch, the last microsecond before it, and chrono's 
`NaiveDate::MIN` and `MAX`.
   
   Outside this PR, the new helpers were also compared with chrono 0.4.45 (the 
version in `Cargo.lock`) in a throwaway program: every one of the 191,491,529 
days chrono accepts, 400 million random millisecond and microsecond timestamps 
across chrono's range and near the epoch, and 40,006 timestamps around the ends 
of chrono's range and the epoch. All matched byte for byte. `cargo test -p 
parquet` passes with the default features and with `--all-features`.
   
   # Are there any significant user-facing changes?
   
   No public API changes. One behaviour change: a DATE, TIMESTAMP_MILLIS or 
TIMESTAMP_MICROS value outside chrono's range (more than about 262,000 years 
from year 0) used to panic in `Field`'s `Display` and `to_json_value`, through 
chrono's `unwrap`. It now formats on the same proleptic Gregorian calendar, for 
example `i64::MAX` milliseconds as `+292278994-08-17 07:12:55.807 +00:00`.
   
   # AI usage disclosure
   
   The code and tests in this PR were written with Claude Code (an AI 
assistant). They were verified by the test runs listed in "Are these changes 
tested?".
   
   🤖 Generated with [Claude Code](https://claude.com/claude-code)
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to