ref(typing): Upgrade mypy to 1.20.1 by JoshFerge · Pull Request #113419 · getsentry/sentry

JoshFerge · 2026-04-20T13:28:05Z

Bumps the mypy pin to 1.20.1. Prep PRs for the bulk class-level-attribute fixes have already merged; what remains here is the pin bump, the GroupType ClassVar + __init_subclass__ migration, a handful of small typing cleanups, and targeted # type: ignore comments for cases that will get proper fixes in the follow-up PRs below.

Prep PRs (already merged)

ref(typing): Convert RegressionDetector + AttributeHandler config to ClassVar #113422 — ref(typing): RegressionDetector + AttributeHandler → ClassVar
ref(issues): Convert GroupType class config to ClassVar #113423 — ref(issues): GroupType → ClassVar + move validation to __init_subclass__
fix(typing): Annotate group_status parameters as int #113424 — fix(typing): group_status args are int, not GroupStatus
test(typing): Refetch via queryset instead of refresh_from_db after narrowing #113427 — test(typing): refetch via queryset instead of refresh_from_db after narrowing
test(typing): Misc test-side type hint fixes #113428 — test(typing): misc test-side type hint fixes

What's in this PR

Bump mypy>=1.19.1 → mypy>=1.20.1 (+ uv.lock)
Drop the now-stale kombu.* entry from [[tool.mypy.overrides]]
src/sentry/issues/grouptype.py — convert GroupType fields to ClassVar and move the category-validation check from the dead __post_init__ into __init_subclass__. Plus the corresponding test updates in tests/sentry/issues/test_grouptype.py and tests/sentry/workflow_engine/endpoints/test_validators.py.
Small typing fixes that don't change runtime behavior:
- digests/backends/base.py — widen options map to Mapping[str, Any]
- discover/compare_timeseries.py — heterogeneous mismatches dict typed as dict[int, dict[str, Any]]
- explore/endpoints/explore_saved_queries.py — drop unused # type: ignore
- tests/sentry/incidents/test_logic.py — stringify both sides of the target_identifier comparison so mypy stops flagging it as unreachable
Targeted # type: ignore on six files where the proper fix is being landed as a follow-up PR (see below)
.github/workflows/scripts/bootstrap-snuba.py — # type: ignore[arg-type] on int(workers_str) (single-file mypy can't see through sys.exit; follow-up can replace with NoReturn)

Follow-up PRs (drafts, based on this branch)

Each one removes one of the # type: ignore comments and applies the proper fix. They'll merge after this lands.

fix(typing): narrow organization_id in from_auth #113927 — fix(typing): narrow organization_id in from_auth
fix(typing): guard None options in ReconnectingMemcache #113928 — fix(typing): guard None options in ReconnectingMemcache
docs(typing): annotate save_team_assignments ignore with FIXME #113929 — docs(typing): annotate the save_team_assignments ignore with a FIXME
ref(typing): drop redundant cast in trace metric search type #113930 — ref(typing): drop redundant cast in trace metric search type
ref(typing): drop redundant cast on TraceMetric.metric_type #113932 — ref(typing): drop redundant cast on TraceMetric.metric_type
ref(typing): drop redundant casts in resolve_measurement_value #113933 — ref(typing): drop redundant casts in resolve_measurement_value

Performance

version	cold	warm
1.19.1	49.22s	17.82s
1.20.1	39.04s	5.73s

CI expectation

mypy job: clean (0 errors)
uv jobs: resolve against pypi.devinfra.sentry.io (1.20.1 has been published there)

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

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

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

`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

`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

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

github-actions · 2026-04-20T13:48:41Z

Backend Test Failures

Failures on 1b18fd5 in this run:

tests/sentry/incidents/test_logic.py::UpdateAlertRuleTriggerAction::test_pagerduty — log

[gw0] linux -- Python 3.13.1 /home/runner/work/sentry/sentry/.venv/bin/python3
tests/sentry/incidents/test_logic.py:2978: in test_pagerduty
    assert action.target_identifier == str(target_identifier)
E   AssertionError: assert 41 == '41'
E    +  where 41 = <AlertRuleTriggerAction at 0x7f0352adf3e0: id=12>.target_identifier
E    +  and   '41' = str(41)

tests/sentry/workflow_engine/endpoints/test_validators.py::TestBaseGroupTypeDetectorValidator::test_validate_type_valid — log

[gw1] linux -- Python 3.13.1 /home/runner/work/sentry/sentry/.venv/bin/python3
tests/sentry/workflow_engine/endpoints/test_validators.py:118: in test_validate_type_valid
    class TestGroupType(GroupType):
src/sentry/issues/grouptype.py:282: in __init_subclass__
    registry.add(cls)
src/sentry/issues/grouptype.py:117: in add
    raise ValueError(
E   ValueError: A group type with the type_id 1 has already been registered.

tests/sentry/workflow_engine/endpoints/test_validators.py::TestBaseGroupTypeDetectorValidator::test_validate_type_incompatible

— log

[gw0] linux -- Python 3.13.1 /home/runner/work/sentry/sentry/.venv/bin/python3
tests/sentry/workflow_engine/endpoints/test_validators.py:142: in test_validate_type_incompatible
    class TestGroupType(GroupType):
src/sentry/issues/grouptype.py:282: in __init_subclass__
    registry.add(cls)
src/sentry/issues/grouptype.py:117: in add
    raise ValueError(
E   ValueError: A group type with the type_id 1 has already been registered.

tests/sentry/notifications/notification_action/test_metric_alert_registry_handlers.py::TestBaseMetricAlertHandler::test_build_notification_context

— log

[gw0] linux -- Python 3.13.1 /home/runner/work/sentry/sentry/.venv/bin/python3
tests/sentry/notifications/notification_action/test_metric_alert_registry_handlers.py:375: in test_build_notification_context
    assert notification_context.integration_id == 1234567890
E   AssertionError: assert '1234567890' == 1234567890
E    +  where '1234567890' = NotificationContext(id=33, integration_id='1234567890', target_identifier='channel456', target_display=None, target_type=None, sentry_app_config=None, sentry_app_id=None).integration_id

…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

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

\`sentry.models.group.GroupStatus\` is a plain class whose members are ints (\`RESOLVED = 1\`, \`IGNORED = 2\`, etc.) — not an enum. Callers of the functions below all pass ints, and the in-function \`==\` comparisons (\`group_status == GroupStatus.RESOLVED\`) are int-vs-int. But the parameters were annotated as \`GroupStatus\`, i.e. \"instance of GroupStatus\", which is never what the callers pass. Under \`strict_equality = true\` the comparisons become semantically unreachable (instance-of-class never equals int). Stricter mypy reports them as \`[unreachable]\`. Change the annotations to \`int\` to match actual usage: - \`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). Safe under 1.19.1. Agent transcript: https://claudescope.sentry.dev/share/wKKwnZeeYc-V3zQbolQDCON52n93vyqkCmPrIs_kbnI

…arrowing (#113427) Tests like: \`\`\`python 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 doesn't reset the narrowed attribute type — subsequent 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\` ### Why this PR exists Prep for the mypy upgrade to 1.20.1 in #113419. Safe under the current mypy (1.19.1). Misc other test-side typing nits live separately in the companion PR.

…ClassVar (#113422) Both `RegressionDetector` (`src/sentry/statistical_detectors/detector.py`) and `AttributeHandler` (`src/sentry/rules/conditions/event_attribute.py`) 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`, etc.). The classes are never instantiated; subclasses override the fields as class-level attributes (`source = "transaction"`, `minimum_path_length = 1`). Convert the fields to `ClassVar[T]` and drop the no-op `@dataclass(frozen=True)` decorators. No runtime behavior change. ### Why this PR exists Prep for the mypy upgrade to 1.20.1 in #113419. 1.20.1 adds a new `[misc]` check, "Cannot access instance-only attribute on class object," which surfaces this pattern. Splitting the fix out so each targeted reviewer only has to look at their own area. This PR is safe under the current mypy (1.19.1). Agent transcript: https://claudescope.sentry.dev/share/8Gzqw1K3cHTGBqYv-FAytvqDyA86B3vsiZyoJkHL7DM

Small one-off type hint corrections in tests, each behavior-neutral. Split off from the `refresh_from_db()` refactor in #113427 so the patterns can be reviewed separately. - `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`: pass `integration_id` as an int in the fixture so the in-memory dataclass value matches the declared `int | None` type (no DB round-trip coerces it). - `test_logic.py`: stringify both sides of the `target_identifier` assertion in the update path. `update_alert_rule_trigger_action` leaves `target.identifier` as int in memory (unlike the create path, which wraps with `str()`), so stringifying avoids the runtime mismatch while keeping mypy happy. ### Why this PR exists Prep for the mypy upgrade to 1.20.1 in #113419. Safe under the current mypy (1.19.1). Agent transcript: https://claudescope.sentry.dev/share/OOKfaLYBcSwWWD48Y6-ebfs_iQCylldqs4WhtB_1ibU

`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]`) was really class-level access against an instance-level annotation. Convert the fields to `ClassVar[T]`, preserving the same defaults. Also convert `ReplayGroupTypeDefaults.notification_config` for the same reason. ### Small behavior change `GroupType.__post_init__` was previously dead code — nothing instantiates `GroupType`, so the category-value validation never actually ran. Move the same validation to `__init_subclass__` so it fires at subclass-definition time (earlier, and actually reachable). Updates: - `test_grouptype.py::test_category_validation` — now expects the `ValueError` at class-definition time. - `test_validators.py` — two tests were instantiating `GroupType(...)` to build a mock; since the base no longer takes fields as constructor args, convert them to subclasses of `GroupType` instead (which matches real production usage). ### Why this PR exists Prep for the mypy upgrade to 1.20.1 in #113419. 1.20.1 adds a new `[misc]` check, "Cannot access instance-only attribute on class object," and `GroupType` is the single biggest cluster of such errors in the repo. Splitting it out so the targeted reviewers only have to look at their own areas. Safe under the current mypy (1.19.1). Agent transcript: https://claudescope.sentry.dev/share/JFAQdL9BKvw8zOc8RROKUSNl-Xpuu8y4p4URoIp7DDk

…de-mypy-to-1-20-1 # Conflicts: # tests/sentry/incidents/test_logic.py # tests/sentry/models/test_dashboard.py Agent transcript: https://claudescope.sentry.dev/share/D--sRHsVvSPk6WPQmSZcYG-A3qGMCOHGYki10ltOJDI

Three new mypy 1.20 failures surfaced from master after the merge: - bootstrap-snuba.py:214 — fail() always calls sys.exit(1), so annotate it as NoReturn. Fixes the "int(workers_str)" call on a str|None arg. - columns.py:498 — cast trace_metric.metric_unit to SearchType; mypy doesn't narrow from `x in set[SearchType]` membership. - test_logic.py::test_alert_rule_owner — refetch via AlertRule.objects .get(id=...) between update_alert_rule calls so mypy drops the narrowed (user_id, team_id) types. Same pattern as #113427. Agent transcript: https://claudescope.sentry.dev/share/n_liZXmbQ0rj2hiJIxs4xsZr1MegSMvDxXqAl75y_F8

mypy 1.20.1 narrows `trace_metric.metric_unit` via the `in DURATION_TYPE` / `in SIZE_TYPE` set-membership checks, so the cast I added earlier was redundant and tripped `warn_redundant_casts`. (Locally I was still on 1.20.0 which lacked that narrowing, which is why I missed it before pushing.) Agent transcript: https://claudescope.sentry.dev/share/qVN5T3yPojgSEpPxyZeF1DC78Y1TPOBuaCbCjqQuHps

The mypy 1.20.0 false positives I worked around don't exist under 1.20.1 (which is what CI actually runs). Drop the test_logic.py queryset refetches and the bootstrap-snuba.py NoReturn annotation; full-repo mypy passes cleanly under 1.20.1 without them. Agent transcript: https://claudescope.sentry.dev/share/WDLHqJ2mEu0M75cxgnMiWXrgXC6nT52jGIlB3VRCRR8

cmanallen · 2026-04-24T16:16:32Z

+            return constants.SIZE_UNITS[cast(constants.SizeUnit, unit)] * value  # type: ignore[redundant-cast]
        elif unit in constants.DURATION_UNITS:
-            return constants.DURATION_UNITS[cast(constants.DurationUnit, unit)] * value
+            return constants.DURATION_UNITS[cast(constants.DurationUnit, unit)] * value  # type: ignore[redundant-cast]


If its redundant can the cast be removed?

yep, will do so in follow up 👍🏼 just caught in a chicken and egg as roughly 1 incompatible change with the new upgrade lands every day, so ignoring in this PR and will land fixes after the mypy upgrade PR is merged.

cmanallen · 2026-04-24T16:16:48Z

                or trace_metric.metric_unit in constants.SIZE_TYPE
            ):
-                resolved_search_type = cast(constants.SearchType, trace_metric.metric_unit)
+                resolved_search_type = cast(  # type: ignore[redundant-cast]


Can the cast be removed?

will do so in follow up 👍🏼

cmanallen · 2026-04-24T16:16:56Z

    return TraceMetric(
        metric_name=metric_name,
-        metric_type=cast(TraceMetricType, metric_type),
+        metric_type=cast(TraceMetricType, metric_type),  # type: ignore[redundant-cast]


Remove cast?

will do so in follow up 👍🏼

mypy 1.20.1 now flags passing `auth.organization_id` (typed `int | None`) to `OrganizationGlobalAccess` (expects `int`). Narrow the value locally before the call instead of carrying a `# type: ignore[arg-type]`. Follow-up to #113419 — removes one of the targeted ignores that PR introduced. Agent transcript: https://claudescope.sentry.dev/share/WG-0likblPhNpRq3ev8pLoZRMrn8oSFOE2xrVHQm5ds

`PyMemcacheCache._options` is typed `dict | None`; splatting None into `**kwargs` is a runtime TypeError. Fall back to `{}` so the arg-type ignore introduced in #113419 can be removed. Agent transcript: https://claudescope.sentry.dev/share/asYXfLRHOo-uGu5pLDRYXImp8ccvxh-gX3E_3dVUDHU

The arg-type ignore was added in #113419; document the underlying mismatch (team_roles can contain (team, None) tuples when organizations:team-roles is disabled, which doesn't match the declared signature) so the follow-up either fixes the signature or the caller. Agent transcript: https://claudescope.sentry.dev/share/c6N6Q15yfk5b4v9xtEM2JZaFwXItVVVaA_yuZo2UyDY

mypy 1.20.1 narrows `trace_metric.metric_unit` via the `in DURATION_TYPE / SIZE_TYPE` membership checks, so the `cast(constants.SearchType, ...)` is redundant under the new pin. Remove it along with the `# type: ignore[redundant-cast]` #113419 parked there. Agent transcript: https://claudescope.sentry.dev/share/IQw7RcSfBRYsfH4fqUWuAB3LmwTb1eK5eIUHbhG9Z74

mypy 1.20.1 narrows `metric_type` to the `TraceMetricType` Literal via the preceding `if metric_type not in (...)` check, so the cast is redundant under the new pin. Drop it and the unused `cast` import, and remove the `# type: ignore[redundant-cast]` #113419 added. Agent transcript: https://claudescope.sentry.dev/share/73VcpxsrWayhrckFm4nt6rNv1Ut6V2OD-g2j5UkaJxM

mypy 1.20.1 narrows `unit` via the `in SIZE_UNITS` / `in DURATION_UNITS` membership checks, making the `cast(SizeUnit/DurationUnit, unit)` calls redundant. Drop both along with the `# type: ignore[redundant-cast]` #113419 parked on each line. Agent transcript: https://claudescope.sentry.dev/share/D5BcySnDeVsi_TvpdsbF4DnHgDCKeoLRjrK-4DBQaMA

## Summary Drop the two \`cast(...)\` calls in \`BaseQueryBuilder.resolve_measurement_value\` — mypy 1.20.1 narrows \`unit\` via \`in SIZE_UNITS / DURATION_UNITS\` membership, so both casts are redundant under the new pin. Removes the pair of \`# type: ignore[redundant-cast]\` comments #113419 parked on these lines. Drafted because it depends on #113419. ## Test plan - [ ] \`.venv/bin/python -m mypy src/sentry/search/events/builder/base.py\` passes Agent transcript: https://claudescope.sentry.dev/share/9knBZJ8vsizIvYD0SEHZTCEVeYi3SKsV51XS1gS1L0k

## Summary Drop the \`cast(TraceMetricType, metric_type)\` in \`get_trace_metric_from_request\` — mypy 1.20.1 narrows \`metric_type\` via the preceding validation check, so the cast is redundant under the new pin. Removes the \`# type: ignore[redundant-cast]\` #113419 parked there. Drafted because it depends on #113419. ## Test plan - [ ] \`.venv/bin/python -m mypy src/sentry/search/eap/trace_metrics/config.py\` passes Agent transcript: https://claudescope.sentry.dev/share/EAwMhN54lGJy_R4gjE03VMzhxtMKTejQyHgo_pGSXt4

## Summary Drop the \`cast(constants.SearchType, trace_metric.metric_unit)\` in \`TraceMetricAggregateDefinition\` — mypy 1.20.1 narrows via the \`x in DURATION_TYPE / SIZE_TYPE\` membership checks, so the cast is redundant under the new pin. Removes the \`# type: ignore[redundant-cast]\` that #113419 parked there. Drafted because it depends on #113419. ## Test plan - [ ] \`.venv/bin/python -m mypy src/sentry/search/eap/columns.py\` passes Agent transcript: https://claudescope.sentry.dev/share/s6xsPeER6hqruuVvkujoI7fjuxD2y1wtbo-39lZR6yk

## Summary Remove the \`# type: ignore[arg-type]\` on \`auth.organization_id\` in \`from_auth\` by narrowing the value to \`int\` before passing it to \`OrganizationGlobalAccess\`. Follow-up to #113419. ## Test plan - [ ] \`.venv/bin/python -m mypy src/sentry/auth/access.py\` passes - [ ] Existing \`from_auth\` tests still pass Agent transcript: https://claudescope.sentry.dev/share/l4qdLaYXiQY9XtMd5ykax-e5mzkLHpsYfnHY4Jjeaoo

## Summary \`PyMemcacheCache._options\` is typed \`dict | None\` but \`**self._options\` would explode at runtime on \`None\`. Fall back to \`{}\`, which also lets us drop the \`# type: ignore[arg-type]\` that #113419 added. Agent transcript: https://claudescope.sentry.dev/share/ZoHjMlN-lDWKU2242qjXGR-0l_UDipLzt6dZKGatk3Q

## Summary Remove the \`# type: ignore[arg-type]\` on \`auth.organization_id\` in \`from_auth\` by narrowing the value to \`int\` before passing it to \`OrganizationGlobalAccess\`. Follow-up to #113419. ## Test plan - [ ] \`.venv/bin/python -m mypy src/sentry/auth/access.py\` passes - [ ] Existing \`from_auth\` tests still pass Agent transcript: https://claudescope.sentry.dev/share/l4qdLaYXiQY9XtMd5ykax-e5mzkLHpsYfnHY4Jjeaoo

## Summary \`PyMemcacheCache._options\` is typed \`dict | None\` but \`**self._options\` would explode at runtime on \`None\`. Fall back to \`{}\`, which also lets us drop the \`# type: ignore[arg-type]\` that #113419 added. Agent transcript: https://claudescope.sentry.dev/share/ZoHjMlN-lDWKU2242qjXGR-0l_UDipLzt6dZKGatk3Q

github-actions Bot added the Scope: Backend Automatically applied to PRs that change backend components label Apr 20, 2026

vercel Bot deployed to Preview April 20, 2026 13:29 View deployment

vercel Bot deployed to Preview April 20, 2026 13:35 View deployment

JoshFerge mentioned this pull request Apr 20, 2026

ref(typing): Convert RegressionDetector + AttributeHandler config to ClassVar #113422

Merged

JoshFerge mentioned this pull request Apr 20, 2026

ref(issues): Convert GroupType class config to ClassVar #113423

Merged

JoshFerge mentioned this pull request Apr 20, 2026

fix(typing): Annotate group_status parameters as int #113424

Merged

JoshFerge mentioned this pull request Apr 20, 2026

ref: Drop redundant casts in search builders #113426

Closed

JoshFerge mentioned this pull request Apr 20, 2026

test(typing): Refetch via queryset instead of refresh_from_db after narrowing #113427

Merged

JoshFerge mentioned this pull request Apr 20, 2026

test(typing): Misc test-side type hint fixes #113428

Merged

JoshFerge added 2 commits April 24, 2026 11:17

Merge remote-tracking branch 'origin/master' into joshferge/ref/upgra…

35fd6d0

…de-mypy-to-1-20-1 # Conflicts: # tests/sentry/incidents/test_logic.py # tests/sentry/models/test_dashboard.py Agent transcript: https://claudescope.sentry.dev/share/D--sRHsVvSPk6WPQmSZcYG-A3qGMCOHGYki10ltOJDI

vercel Bot deployed to Preview April 24, 2026 15:24 View deployment

vercel Bot deployed to Preview April 24, 2026 15:40 View deployment

This was referenced Apr 24, 2026

ref(typing): drop redundant cast on TraceMetric.metric_type #113932

Merged

ref(typing): drop redundant casts in resolve_measurement_value #113933

Merged

JoshFerge marked this pull request as ready for review April 24, 2026 16:13

JoshFerge requested review from a team as code owners April 24, 2026 16:13

JoshFerge requested a review from a team April 24, 2026 16:13

cmanallen approved these changes Apr 24, 2026

View reviewed changes

JoshFerge merged commit e4fbee6 into master Apr 24, 2026
80 checks passed

JoshFerge deleted the joshferge/ref/upgrade-mypy-to-1-20-1 branch April 24, 2026 16:18

armenzg assigned JoshFerge May 6, 2026

github-actions Bot locked and limited conversation to collaborators May 22, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

ref(typing): Upgrade mypy to 1.20.1#113419

ref(typing): Upgrade mypy to 1.20.1#113419
JoshFerge merged 7 commits into
masterfrom
joshferge/ref/upgrade-mypy-to-1-20-1

JoshFerge commented Apr 20, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Apr 20, 2026 •

edited

Loading

Uh oh!

cmanallen Apr 24, 2026

Uh oh!

JoshFerge Apr 24, 2026

Uh oh!

cmanallen Apr 24, 2026

Uh oh!

JoshFerge Apr 24, 2026

Uh oh!

cmanallen Apr 24, 2026

Uh oh!

JoshFerge Apr 24, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Uh oh!

Conversation

JoshFerge commented Apr 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Prep PRs (already merged)

What's in this PR

Follow-up PRs (drafts, based on this branch)

Performance

CI expectation

Uh oh!

github-actions Bot commented Apr 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Backend Test Failures

Uh oh!

cmanallen Apr 24, 2026

Choose a reason for hiding this comment

Uh oh!

JoshFerge Apr 24, 2026

Choose a reason for hiding this comment

Uh oh!

cmanallen Apr 24, 2026

Choose a reason for hiding this comment

Uh oh!

JoshFerge Apr 24, 2026

Choose a reason for hiding this comment

Uh oh!

cmanallen Apr 24, 2026

Choose a reason for hiding this comment

Uh oh!

JoshFerge Apr 24, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

JoshFerge commented Apr 20, 2026 •

edited

Loading

github-actions Bot commented Apr 20, 2026 •

edited

Loading