Implement the datetime plan with fixed offsets to start by mwildehahn · Pull Request #171 · pydantic/monty

mwildehahn · 2026-02-16T17:42:16Z

Scope

This PR is intentionally kept narrow to make review tractable.

Deferred From This PR

These changes were removed from the diff against main and will land in a follow-up PR:

crates/monty-typeshed/vendor/typeshed/**
crates/monty/src/bytecode/vm/async_exec.rs
crates/monty/src/bytecode/vm/binary.rs
crates/monty/src/bytecode/vm/call.rs
crates/monty/src/bytecode/vm/compare.rs
crates/monty/src/bytecode/vm/mod.rs
crates/monty/src/bytecode/vm/scheduler.rs
crates/monty/src/types/module.rs

Dependent cleanups were also reverted to keep the branch compiling with the rollback:

crates/monty/src/repl.rs
crates/monty/src/run_progress.rs

What Those Deferred VM/Module Changes Were Trying To Do

Preserve hidden resume metadata for pending datetime.now OS calls so async future resolution can reconstruct date.today(), naive datetime.now(), and fixed-offset aware datetime.now(tz=...) correctly.
Improve CPython parity for datetime/timedelta arithmetic and comparison edge cases (overflow paths and naive/aware error handling).
Change type/module method dispatch so module-exposed builtin types (notably datetime.date / datetime.datetime) handle class methods consistently.
Centralize some comparison and unary-op edge handling inside VM paths.

Deferred Regression Coverage

The intended failing cases are captured as commented tests in:

crates/monty/test_cases/datetime__core.py

Those snippets are intentionally commented out in this PR and should be re-enabled in the follow-up PR that restores the deferred VM/module work.

codspeed-hq · 2026-02-16T17:44:43Z

Merging this PR will not alter performance

✅ 13 untouched benchmarks

_{Comparing mwildehahn:impl/datetime-fixed-offset (7ebc359) with main (4ee3fd7)}

codecov · 2026-02-16T18:34:40Z

Codecov Report

❌ Patch coverage is 63.83087% with 710 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
crates/monty/src/types/datetime.rs	62.62%	147 Missing and 44 partials ⚠️
crates/monty/src/heap.rs	34.91%	105 Missing and 5 partials ⚠️
crates/monty/src/types/date.rs	51.47%	86 Missing and 13 partials ⚠️
crates/monty/src/types/timedelta.rs	71.09%	49 Missing and 25 partials ⚠️
crates/monty/src/types/timezone.rs	70.61%	50 Missing and 12 partials ⚠️
crates/monty/src/repl.rs	0.00%	38 Missing ⚠️
crates/monty/src/object.rs	75.86%	17 Missing and 18 partials ⚠️
crates/monty/src/run_progress.rs	77.41%	31 Missing and 4 partials ⚠️
crates/monty/src/value.rs	47.61%	20 Missing and 2 partials ⚠️
crates/monty/src/heap_data.rs	0.00%	17 Missing ⚠️
... and 6 more

📢 Thoughts on this report? Let us know!

samuelcolvin · 2026-02-18T00:53:13Z

crates/monty-python/src/convert.rs

        Ok(MontyObject::String(string.extract()?))
    } else if let Ok(bytes) = obj.cast::<PyBytes>() {
        Ok(MontyObject::Bytes(bytes.extract()?))
+    } else if obj.is_instance(get_datetime_datetime(obj.py())?)? {


the cpython c-api and pyo3 have proper support python datetime objects, e.g. PyDatetime, you should cast to those here.

You might want to look at the code in pydantic for how we do this there.

samuelcolvin · 2026-02-18T00:55:37Z

crates/monty-typeshed/vendor/typeshed/stdlib/_typeshed/__init__.pyi

 #     def __new__(cls: type[Self], ...) -> Self: ...
 # In other cases, use `typing_extensions.Self`.
-Self = TypeVar('Self')
+Self = TypeVar('Self')  # noqa: Y001


why has this changed?

samuelcolvin · 2026-02-18T00:56:15Z

crates/monty-typeshed/vendor/typeshed/stdlib/builtins.pyi

please undo all these changes in ‎crates/monty-typeshed/vendor/typeshed/stdlib/

@samuelcolvin to get tests passing I had to regen typeshed. LMK if there is a better way to do this or if you still want me to remove these.

samuelcolvin · 2026-02-18T00:58:29Z

crates/monty/src/bytecode/vm/call.rs

    match (t, method_id) {
-        (Type::Dict, m) if m == StaticStrings::Fromkeys => return dict_fromkeys(args, heap, interns),
-        (Type::Bytes, m) if m == StaticStrings::Fromhex => return bytes_fromhex(args, heap, interns),
+        (Type::Dict, m) if m == StaticStrings::Fromkeys => {


we should probably move these methods into dict.rs etc. rather than have the logic here.

samuelcolvin · 2026-02-18T00:59:10Z

crates/monty/src/bytecode/vm/compare.rs

+/// Returns datetime awareness (`true` for aware, `false` for naive) for datetime values.
+///
+/// Returns `None` when the value is not a datetime reference.
+fn datetime_awareness(value: &Value, heap: &crate::heap::Heap<impl ResourceTracker>) -> Option<bool> {


I think this method is duplicated.

samuelcolvin · 2026-02-18T01:02:03Z

crates/monty/src/run.rs

+        // Pre-process OS callback results before restoring the VM to avoid borrowing
+        // conflicts with the VM's mutable borrow of `self.heap`.
+        let resume_action = match ext_result {
+            ExternalResult::Return(obj) => {
+                if let Some(transform) = self.pending_os_transform {
+                    match transform_os_return_value(obj, transform, &mut self.heap) {
+                        Ok(value) => ResumeAction::ReturnValue(value),
+                        Err(msg) => ResumeAction::Error(RunError::from(MontyException::runtime_error(format!(
+                            "invalid return type: {msg}"
+                        )))),
+                    }
+                } else {
+                    ResumeAction::ReturnObj(obj)
+                }
+            }
+            ExternalResult::Error(exc) => {
+                let err = RunError::from(exc);
+                ResumeAction::Error(err)
+            }
+            ExternalResult::Future => ResumeAction::Future,
+        };


this is ugly, we should add Datetime, Date etc. varients to MontyObject

I guess these variants should use whatever Datetime and Date etc. struct we settle on - e.g. our own Datetime type that wraps a chrono or speedate date.

samuelcolvin · 2026-02-18T01:05:12Z

crates/monty/src/types/date.rs

+
+/// Creates a date from a proleptic Gregorian ordinal value.
+pub(crate) fn from_ordinal(ordinal: i32) -> RunResult<Date> {
+    let days_since_epoch = i64::from(ordinal) - 719_163;


where is 719_163 coming from, we need more explanation

samuelcolvin · 2026-02-18T01:07:09Z

crates/monty/src/types/date.rs

there's a lot of logic here. I know I recommended speedate, but would this be simpler if we used chrono?

I initially suggested not to use chrono because of it's leap second stuff, but maybe it would be cleaner than this???

samuelcolvin · 2026-02-18T01:09:26Z

crates/monty/src/types/date.rs

+};
+
+/// `datetime.date` storage backed directly by `speedate`.
+pub(crate) type Date = speedate::Date;


you should wrapp the speedate Date (or chrono date) like

pub(crate) struct Date(speedate::Date);

Then you can implement (de)serialization directly here.

samuelcolvin · 2026-02-18T01:13:05Z

crates/monty/src/types/datetime.rs

+    if !(1..=9999).contains(&year) {
+        return Err(SimpleException::new_msg(ExcType::ValueError, format!("year {year} is out of range")).into());
+    }
+    if !(1..=12).contains(&month) {
+        return Err(SimpleException::new_msg(ExcType::ValueError, "month must be in 1..12").into());
+    }


you've missed days, but also whatever logic we use here should be shared with Date.

mwildehahn · 2026-02-18T06:33:06Z

@samuelcolvin i think i got to all the feedback -- i switched to chrono which let us drop a lot of the custom logic around date / datetime.

i'm not sure how to handle the vendored typeshed. i previously had bumped that and it brought in all the other changes. the failing test is because datetime isn't in the stdlib.

…chrono

…o types directly

…atterns

…resolution - Extracted magic numbers like 86_400, 1_000_000 into constants in timedelta, timezone, date, datetime and convert modules - Replaced invalid const 'i128::from' in timedelta constants and fixed subsec_nanos division to prevent compilation errors - Added manual Hash implementation for DateTime based on aware/naive py_eq equivalence rather than raw fields - Updated typeshed to use upstream datetime.pyi directly instead of relying on custom stub overrides and build.rs zip manipulations

- Added HeapDataMut variants for datetime types so VM generic paths don't panic on to_mut - Fixed Value::py_cmp fallback to dispatch to HeapData::py_cmp instead of returning None - Adjusted Option<Ordering> evaluation for None (Float NaN) vs TypeError (incomparable types)

…rooting - Updated heap GC mark phase to keep the lazily-allocated datetime.timezone.utc singleton alive - Updated MontyObject -> HeapData::TimeZone conversion to reuse the cached UTC singleton instead of allocating new instances, maintaining 'is' identity - Rejected explicit None passed as the name to timezone() constructors to match CPython's TypeError behavior

mwildehahn mentioned this pull request Feb 16, 2026

add support for time and datetime? #150

Open

samuelcolvin reviewed Feb 18, 2026

View reviewed changes

mwildehahn force-pushed the impl/datetime-fixed-offset branch 2 times, most recently from 421cfb7 to 5c87707 Compare February 18, 2026 05:07

mwildehahn force-pushed the impl/datetime-fixed-offset branch from f1a42f8 to 2ce1535 Compare March 3, 2026 22:34

mwildehahn added 22 commits March 3, 2026 21:33

Add detailed plan for datetime fixed-offset support

65c2d3e

Review feedback

b70773a

Review feedback

4a039a2

feat(datetime): implement core runtime types and os-callback now flow

1848a37

test(datetime): add deterministic now/today coverage across runtimes

f40a64c

chore(datetime): update python os typing and vendored datetime stubs

49d5e7b

docs(plan): log datetime phase progress and completion

af41ecc

docs(examples): add datetime planning example with schedule calculations

abc3f71

docs(examples): switch datetime planning example to fixed-offset PST

0ccc4bb

docs(examples): render datetime planning times in 12-hour PST format

1010a77

Refactor datetime core to use speedate types

332f05a

Move PLAN.md to plans/datetime.md

9a64057

Align datetime fixed-offset behavior with CPython

a769a51

Fix os snapshot roundtrip cleanup under ref-count-panic

7f11dfa

Add datetime callback validation coverage and boundary fixtures

94a5742

Add coverage tests for MontyObject datetime variants

df98337

refactor(python): use typed PyO3 datetime wrappers for conversions

2c955c6

refactor(datetime): migrate runtime date/time types from speedate to …

8a6356f

…chrono

chore(typeshed): restore vendored stdlib stubs to upstream/main

b9f87c6

refactor(vm): deduplicate datetime awareness helper

a60d4e5

refactor(datetime): drop legacy snapshot adapters and serialize chron…

c8934c6

…o types directly

fix(datetime): add explicit day range validation in from_components

c8e56bd

mwildehahn added 15 commits March 3, 2026 21:35

chore: revert pyproject.toml changes

4fb7250

refactor(datetime): reuse date validation in datetime from_components

c372274

fix(datetime): align constructor and timezone behavior with library p…

d5e1033

…atterns

Refine datetime/timezone handling and classmethod dispatch

c7a2ba8

Fix datetime os-call resume conversion and datetime hashing

f34fa48

Package custom typeshed stubs and register versions

86664e6

Fix ordering and datetime resume/repr regressions

6ee9e50

Fix exact float/longint ordering semantics

33171f5

Fix datetime tzinfo identity semantics

4971177

Add TODO notes for deferred datetime/timezone regression tests

acba67b

chore: narrow datetime PR scope and remove dead date helper

4af6764

refactor: extract datetime OS bridge metadata plumbing

fecf84f

mwildehahn force-pushed the impl/datetime-fixed-offset branch from 0ccc76d to fecf84f Compare March 4, 2026 05:38

mwildehahn added 3 commits March 3, 2026 21:39

chore: drop typeshed update script changes from datetime PR

bb0b730

chore(datetime): strip non-essential VM and iter fixture changes

79e23bb

chore: format module attr error branch for pre-commit

7ebc359

Conversation

mwildehahn commented Feb 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Scope

Deferred From This PR

What Those Deferred VM/Module Changes Were Trying To Do

Deferred Regression Coverage

Uh oh!

codspeed-hq bot commented Feb 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will not alter performance

Uh oh!

codecov bot commented Feb 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mwildehahn commented Feb 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

mwildehahn commented Feb 16, 2026 •

edited

Loading

codspeed-hq bot commented Feb 16, 2026 •

edited

Loading

codecov bot commented Feb 16, 2026 •

edited

Loading