Draft
Conversation
Bump the mypy pin to 1.20.1 and fix the bulk of new type-check failures. 1.20.1 adds a stricter check for accessing instance-only attributes on a class object. Several class-config patterns in the codebase (RegressionDetector, GroupType, AttributeHandler, ReplayGroupTypeDefaults) declared configuration fields as dataclass instance attributes but only ever accessed them via the class. Convert those to ClassVar so access through the class is valid and drop the no-op @DataClass(frozen=True) decorators. The GroupType category validation that previously lived in __post_init__ (dead code — no one instantiated GroupType) moves into __init_subclass__ so it still runs at subclass-definition time. Also addresses: - Incorrect GroupStatus annotation where int is passed in practice (fetch_alert_threshold, fetch_resolve_threshold, from_workflow_engine_models, build_alert_context) - Redundant casts in search/eap and search/events builders - Unused kombu.* mypy override - Unused type: ignore comments - Mapping type fix in digests/backends/base.py - None-narrowing fix in auth/access.from_auth - None guard in reconnectingmemcache Cuts the full-repo mypy failure count from 134 -> 15. Agent transcript: https://claudescope.sentry.dev/share/mi26dmBkpSB5zU45vLb9vayuq61XpEbOEAaKoC-2AGQ
github-actions
bot
added
the
Scope: Backend
Fixes the 15 remaining mypy 1.20 test failures so the full-repo run is clean. Patterns fixed: - Replace `x.refresh_from_db()` with a fresh `Model.objects.get(id=x.id)` assignment in tests where mypy retained narrowing from an earlier assertion (e.g. `assert widget.discover_widget_split is None`). refresh_from_db mutates in place so mypy never re-widens; an assignment does. - Guard or cast `Mapping | None` / `object` before `**kwargs` splats. - Assert `is not None` before dereferencing an Optional parsed MRI. - Type a heterogeneous debug_meta case list as `list[dict[str, Any]]`. - Annotate exchange_token test result as `dict[str, Any]` — the declared return type is `dict[str, str]` but real OAuth responses include ints. - Compare notification_context.integration_id against an int, not a str. Agent transcript: https://claudescope.sentry.dev/share/_Bezr4OWYESAQM3P9KhWPgffIOzp6lSEgTG_mq5brNM
JoshFerge
added a commit
that referenced
this pull request
Apr 20, 2026
Both `RegressionDetector` and `AttributeHandler` declared their configuration fields as dataclass instance attributes (via `@dataclass(frozen=True)`), but every usage accesses them through the class (`cls.source`, `cls.minimum_path_length`). The classes are never instantiated; subclasses override the fields as class- level attributes. Convert the fields to `ClassVar[T]` so access through the class is valid and drop the no-op `@dataclass(frozen=True)` decorators. Prep work for the mypy 1.20 upgrade (#113419), which adds a new `misc` error for 'Cannot access instance-only attribute on class object'. Agent transcript: https://claudescope.sentry.dev/share/JZ-p3R-cM7eIdQz7AaZNFl38wjaYVyV7YpBDM4yLTyU
JoshFerge
added a commit
that referenced
this pull request
Apr 20, 2026
`GroupType` is never instantiated in production: subclasses declare their configuration as class-level attributes (`type_id = 1001`, `slug = "foo"`, etc.) and the registry stores the classes themselves. Despite that, the base declared fields as `@dataclass(frozen=True)` instance attributes, so every access like `group_type.type_id` (where `group_type: type[GroupType]`) is really class-level access. Convert the fields to `ClassVar[T]`, preserving the same defaults. Also convert the `ReplayGroupTypeDefaults` mixin's `notification_config` for the same reason. Behavior change: `GroupType.__post_init__` was previously dead code (nothing instantiates `GroupType`), so category validation never actually ran. Move the same validation to `__init_subclass__` so it fires at subclass-definition time. Update the two tests that were constructing `GroupType(...)` (not a real production usage pattern) to define subclasses and the category-validation test to expect the failure at class-definition time. Prep work for the mypy 1.20 upgrade (#113419) — safe under 1.19.1. Agent transcript: https://claudescope.sentry.dev/share/8kuOPo_F7g0JelnKiHguToNZr0_y4AKNCOGNvydSZac
JoshFerge
added a commit
that referenced
this pull request
Apr 20, 2026
`sentry.models.group.GroupStatus` is a plain class whose members are ints (`RESOLVED = 1`, `IGNORED = 2`, etc.), not an enum. Callers of these functions all pass ints, and the in-function `==` comparisons against `GroupStatus.RESOLVED` are int-vs-int. But the parameters were annotated as `GroupStatus`, i.e. 'instance of GroupStatus'. Under `strict_equality = true` the comparison becomes semantically unreachable (instance-of-class never equals int), which is surfaced as `[unreachable]` errors by stricter mypy. Change the annotations to `int` to match how the values are actually used: - `fetch_alert_threshold` - `fetch_resolve_threshold` - `AlertContext.from_workflow_engine_models` - `MetricAlertNotificationMessageBuilder.build_alert_context` No runtime behavior change. Prep for the mypy 1.20 upgrade (#113419). Agent transcript: https://claudescope.sentry.dev/share/zuptd8j6fTczb1hxTiv3VRqXHCYcX_7vZGRag0vi_DY
JoshFerge
added a commit
that referenced
this pull request
Apr 20, 2026
Three `cast(...)` calls that mypy 1.20.1 reports as redundant, and which are in fact redundant under 1.19.1 too (mypy already infers the narrowed type from the surrounding `if unit in constants.SIZE_UNITS` etc. checks). - `src/sentry/search/eap/columns.py`: drop `cast(constants.SearchType, ...)` inside a branch already guarded by membership in DURATION_TYPE/SIZE_TYPE. - `src/sentry/search/events/builder/base.py`: drop `cast(constants.SizeUnit, unit)` and `cast(constants.DurationUnit, unit)` inside the corresponding membership- checked branches. - `src/sentry/search/eap/trace_metrics/config.py`: drop `cast(TraceMetricType, metric_type)` where `metric_type` is already narrowed by `if ... not in ALLOWED_METRIC_TYPES` early-return. Prep for the mypy 1.20 upgrade (#113419). Agent transcript: https://claudescope.sentry.dev/share/rGGex5V3dtTetZqF-KULYaNyoOlBBLtcZWOK-0xI95U
JoshFerge
added a commit
that referenced
this pull request
Apr 20, 2026
Collected test-only type cleanups. All are preexisting latent issues that
stricter mypy (1.20.1) surfaces; these fixes are behavior-neutral under
the current checker too.
Two main patterns:
1. `x.refresh_from_db()` after an `assert x.field is None` (or similar) —
refresh_from_db mutates in place, so mypy keeps the narrowing; later
assertions against non-None values are therefore unreachable. Replace
with `x = Model.objects.get(id=x.id)`, which is semantically equivalent
(fresh row from db) and resets the type. Applies to:
- tests/sentry/models/test_dashboard.py
- tests/snuba/api/endpoints/test_organization_events_mep.py
- tests/snuba/api/endpoints/test_organization_events_stats_mep.py
- tests/sentry/monitors/logic/test_mark_ok.py
- tests/sentry/grouping/seer_similarity/test_training_mode.py
2. Misc annotations/guards:
- tests/sentry/snuba/metrics/test_naming_layer.py: assert parsed MRI is
not None before dereferencing.
- tests/sentry/integrations/vsts/test_provider.py: annotate
`exchange_token` test result as `dict[str, Any]`. The declared return
type is `dict[str, str]` but real OAuth responses include ints
(`expires_in`); the annotation is a preexisting lie, but changing the
api's signature is out of scope here.
- tests/sentry/issues/test_occurrence_consumer.py: type a heterogeneous
debug_meta case list as `list[dict[str, Any]]`.
- tests/sentry/sentry_apps/api/endpoints/test_sentry_apps.py: guard
`Mapping | None` before `**data`.
- tests/sentry/notifications/notification_action/test_metric_alert_registry_handlers.py:
compare `notification_context.integration_id` against an int (the
dataclass field type) rather than a str.
- tests/sentry/incidents/test_logic.py: compare `target_identifier`
against `str(target_identifier)` since the model field is `str`.
Prep for the mypy 1.20 upgrade (#113419). Safe under 1.19.1.
Agent transcript: https://claudescope.sentry.dev/share/0y-hoZTElhR6ktlkfhGtYwf6fxuky_-5F7Fk4q7smC8
Contributor
Backend Test FailuresFailures on
|
JoshFerge
added a commit
that referenced
this pull request
Apr 20, 2026
…arrowing
Tests like:
assert widget.discover_widget_split is None # narrows to None
...
widget.refresh_from_db() # mypy keeps the narrowing
assert widget.discover_widget_split == TRANSACTION_LIKE # mypy: unreachable
`refresh_from_db()` mutates the object in place, so mypy does not reset the
narrowed attribute type — later assertions comparing against non-None values
become semantically unreachable under stricter checkers. Replace with
`x = Model.objects.get(id=x.id)`, which is semantically equivalent (fresh
row from the same table) and an assignment resets the inferred type.
Applied in:
- tests/sentry/models/test_dashboard.py
- tests/snuba/api/endpoints/test_organization_events_mep.py
- tests/snuba/api/endpoints/test_organization_events_stats_mep.py
- tests/sentry/monitors/logic/test_mark_ok.py
- tests/sentry/grouping/seer_similarity/test_training_mode.py
Prep for the mypy 1.20 upgrade (#113419). Safe under 1.19.1.
Agent transcript: https://claudescope.sentry.dev/share/hYewe-LTcVPdbrD9iGU6eeyfLBiHv5fKzkiUjiPAepc
JoshFerge
added a commit
that referenced
this pull request
Apr 20, 2026
Small one-off type hint corrections in tests. Each is behavior-neutral. - test_naming_layer.py: `assert parsed_mri is not None` before dereferencing `.mri_string`. - test_provider.py: annotate the `exchange_token` test result as `dict[str, Any]`. The declared return type is `dict[str, str]` but real OAuth responses include ints (`expires_in`); this is a preexisting lie in the production signature, so the test locally widens until that gets fixed separately. - test_occurrence_consumer.py: annotate a heterogeneous debug_meta case list as `list[dict[str, Any]]`. - test_sentry_apps.py: guard `Mapping | None` before `**data`. - test_metric_alert_registry_handlers.py: compare `notification_context.integration_id` against an int (the dataclass field type) rather than a str. - test_logic.py: compare `target_identifier` against `str(target_identifier)` since the model field is typed `str`. Prep for the mypy 1.20 upgrade (#113419). Safe under 1.19.1. Agent transcript: https://claudescope.sentry.dev/share/zQXk8MwFd_O336VWRbM_DtJHLs12UetlIk6Q3HiZFeY
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Bumps the mypy pin to 1.20.1 and cleans up the remaining new type-check failures that don't belong in a standalone prep PR.
Prep PRs (merge these first, with the old mypy still active)
All six are draft / 1.19.1-safe / can be reviewed independently:
What's left for this PR after the prep merges
(`explore_saved_queries.py`, `test_logic.py`)
where 1.20 catches a preexisting bug: when `organizations:team-roles`
is disabled, the code passes `list[tuple[Team, None]]` to a function
expecting `list[tuple[Team, str]]`
Performance
CI expectation after prep PRs land