Model time zone: which day CURRENT_DATE and the date of an instant mean

未关闭
#710 6 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
5/5
预计耗时
一周以上
新手友好度
25/100
Issue 类型
功能
描述清晰度
描述清楚
活跃度
活跃

调研方向

Start with core-spec/relational_semantics.md §10.2 and trace how the model schema and expression semantics define timezone-sensitive evaluation. Review core-spec/spec.md, core-spec/expression_language.md, and the proposed fixture and cases for deterministic compliance coverage. Done means the model zone behavior, unsupported-zone errors, compatibility without time_zone, and test expectations are specified consistently; the issue author already has a branch and offers to open a PR.

由索引模型根据 Issue 内容生成。

描述

needs-triage

An Ossie model can declare a field as DateTimeTz, an instant (core-spec/spec.md, Data types). Turning an instant into a date or a clock reading needs a time zone, and so does CURRENT_DATE. The spec names no zone for either. Layer 1 promises that "expressions in an Ossie model evaluate identically on any conforming engine" (core-spec/spec_layers.md). Section 10.2 of core-spec/relational_semantics.md lists timezone-sensitive functions among the dialect differences the query interface inherits. For a field such as CAST(created_at AS DATE) the two statements pull apart: the expression is Layer 1, and its result depends on the zone of the engine.

The model author cannot name a zone in portable SQL. The expression language has no AT TIME ZONE and no timestamp literal with a zone (core-spec/expression_language.md, Date/Time Construction).

The instant 2026-11-06 03:00:00 UTC, which is 22:00 on November 5 in New York, as a date (checked 2026-11-10):

Engine Zone it uses Expression Date
DuckDB 2.0.0 session TimeZone = UTC CAST(TIMESTAMPTZ '2026-11-06 03:00:00+00' AS DATE) 2026-11-06
DuckDB 2.0.0 session TimeZone = America/New_York same 2026-11-05
ClickHouse 26.9.5.2, clickhouse local server zone, TZ=UTC toDate(ts), ts a DateTime column with no zone parameter 2026-11-06
ClickHouse 26.9.5.2, clickhouse local server zone, TZ=America/New_York same 2026-11-05

By its documentation, BigQuery computes CURRENT_DATE in UTC unless the call names a zone. A Databricks SQL TIMESTAMP is an instant. Its documentation says the session time zone is applied whenever the value or a part of it is extracted.

A store in New York counts its days in its own zone, as its reports do. "Orders today" is the filter orders.order_day = CURRENT_DATE, where order_day is CAST(created_at AS DATE). On an engine in UTC, both sides are already the next day from 19:00 New York time (20:00 in summer). From then on the filter returns only the orders placed since 19:00. For a store east of UTC, after local midnight the filter also returns part of the previous day.

The zone of a store's dates is a property of its data, the same for every query against the model. That separates it from query-time parameters (#648), which change between queries. The problem exists in a model before any query is written, in a field like order_day above. A Layer 3 filter is "one Layer 1 boolean expression" (core-spec/wide_table_semantics.md, §3.4), so a model zone also fixes CURRENT_DATE in Layer 3 filters.

Proposal: an optional time_zone on the semantic model, an IANA name such as America/New_York. When it is set, every expression evaluated against the model uses that zone:

  • CURRENT_DATE and CURRENT_TIME are counted in the model zone. CURRENT_TIMESTAMP is the current instant.
  • An instant is read as its wall-clock time in the model zone when converted to Date, DateTime or Time. The same holds in YEAR to SECOND, in EXTRACT and DATE_PART with any part, and in DATE_TRUNC. DATEDIFF and DATEADD follow it with a part of day or coarser.
  • In the other direction, a DateTime or Date value compared with an instant is read in the model zone. A TIMESTAMP_NTZ literal is such a value.
  • A wall-clock time that a daylight saving change skips or repeats maps to an implementation-defined instant.
  • An engine that cannot evaluate an expression in the model zone rejects the query with an error. It does not fall back to another zone. An engine rejects a model whose zone name it does not recognize.

Whether a value is an instant depends on its type in the engine, for example a timestamp with time zone. A timestamp column without a zone holds wall-clock values, read as local time in the model zone. Some sources store UTC readings in such a column. A way to declare that, for example datatype: DateTimeTz on the field, could follow separately.

Without time_zone, the zone stays implementation-defined, as today, so no existing model changes meaning. The change adds a sentence to Section 10.2 that a model zone fixes the timezone-sensitive functions.

One zone per model covers a model of one store. A model that combines stores in several zones would need a zone per dataset or per row. A dataset-level time_zone could come later without changing what the model-level one means.

The query compliance data (SCOTT) has only DATE columns, and CURRENT_* cannot be pinned to a fixed value. Deterministic cases need a fixture table with a DateTimeTz column. Its instants fall on both sides of a day boundary in the model zone, and one falls near a daylight saving change. Over it, with time_zone set: CAST(... AS DATE), DATE_TRUNC('month', ...), EXTRACT(HOUR FROM ...), DATEDIFF(day, ...), and a TIMESTAMP_NTZ literal compared with the column. CURRENT_DATE can be tested only through a relation such as CURRENT_DATE = CAST(CURRENT_TIMESTAMP AS DATE).

Two tools already keep a zone setting outside the core. The OrionBelt model in ossie-converters has settings.defaultTimezone, an IANA name used as a fallback zone for naive timestamps (orionbelt/src/ossie_orionbelt/schemas/obml-schema.json). dactopus-ossie-clickhouse reads the model zone from its vendor entry in the model's extensions. The #522 thread calls behavior defined only through extensions non-portable, so a model zone looks like a candidate for the core.

I have a branch with the schema field, the spec text and tests, and can open a PR if this direction looks right.

主要语言
Python
星标
2.7k
派生
353
平均合并
3 天 19 小时
30 天内合并 PR
86

环境准备

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

apache/ossie 的其他 Issue

查看 apache/ossie 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。