flet-validation

name	flet-validation
description	Use whenever editing validation for Python controls in sdk/python/packages/: adding/changing constrained properties, `Raises: ValueError` docstrings, `before_update()` checks, `raise ValueError`, Annotated/V rules, or __validation_rules__.

When To Use

Use this skill when you:

• add validation to a control property,
• migrate manual before_update() checks to V rules,
• update Raises sections tied to validation,
• align Python validation with Dart-side control behavior.

Do not use this skill for unrelated doc-only edits outside control validation. Do not use this skill for deprecation authoring conventions; use flet-deprecation.

Source Of Truth

• Validation runtime API lives in: sdk/python/packages/flet/src/flet/utils/validation.py
• The same runtime can validate regular dataclass models via validate(instance); this skill focuses on the control-specific authoring conventions.
• Import from public path only: from flet.utils.validation import V and when needed from flet.utils.validation import ValidationRules

Validation Decision Order

Use this order and stop at the first option that keeps logic clear:

1. Annotated[..., V.*] field rules (default).
2. __validation_rules__: ValidationRules for cross-field invariants that do not map cleanly to one field.
3. before_update() only for normalization/mutation, or for truly non-ruleable invariants.

Never duplicate the same invariant in more than one layer.

Validation Authoring Rules

1. Prefer field-level validation with Annotated[...] metadata.

   • Example (close to current slider controls):

     @dataclass
     class Sample:
         opacity: Annotated[
                Optional[Number],
                V.between(0.0, 1.0),
            ] = None
     
         value: Annotated[
             Optional[Number],
             V.ge_field("min"),
             V.le_field("max"),
         ] = None
     
         min: Annotated[
             Number,
             V.le_field("max"),
             V.le_field("value"),
         ] = 0.0
     
         max: Annotated[
             Number,
             V.ge_field("min"),
             V.ge_field("value"),
         ] = 1.0
     

2. Use class-level __validation_rules__ only for invariants that cannot be expressed cleanly with field rules.

   • Type: __validation_rules__: ValidationRules = (...)
   • Use V.ensure(...) with an explicit message only when the predicate is short and readable.
   • If V.ensure(lambda ...) becomes hard to read, prefer either:
     1. a named predicate function passed to V.ensure(...), or
     2. a clear before_update() check when it cannot be represented well with existing V.* rules.

3. Keep the before_update() override for normalization/mutation first.

   • Remove validation checks duplicated by V rules.
   • Prefer raising validation errors from rule evaluation (validate()), not from ad-hoc before_update() checks, unless the invariant is genuinely non-ruleable with current validation primitives.

4. Cross-field comparisons use field rules only.

   • Use: V.gt_field, V.ge_field, V.lt_field, V.le_field

5. None handling is inferred from type hints.

   • If a field is annotated as optional (for example Optional[T]), None is allowed.
   • If a field is not optional, None fails validation.
   • For *_field comparisons, if either side is optional and currently None, the comparison is skipped; if a non-optional side is None, validation fails.

6. Match Dart effective behavior.

Typing Style

• Follow the codebase typing style:
  • prefer Optional[T] over T | None,
  • use Union[...] when needed.

Property Docstring Style

When a property has validation, document it in that property’s docstring (google style).

1. Add a Raises: block under the property docstring.

2. Use one ValueError entry per logical rule, except between(...).

   • For V.between(a, b), use one entry: ValueError: If it is not between \a` and `b`, inclusive.`

3. Start each entry with If it ..., where 'it' refers to the property name.

4. Use canonical wording from validation helper docstrings.

   • The source of truth is sdk/python/packages/flet/src/flet/utils/validation.py.
   • Each V.* helper includes Property docstring Raises wording.
   • Keep property Raises entries as negations of the annotation rule.
   • For strict inequalities, say "strictly". For example: V.gt(x) -> ValueError: If it is not strictly greater than \x`.`
   • For sign-neutral divisibility helpers (factor_of, multiple_of), add explicit sign rules (V.gt(0) or V.lt(0)) when direction matters, and include separate Raises entries for those sign rules.

5. Mention conditional applicability when needed.

   • Example (for min/max checks against optional value):

     Raises:
         ValueError: If it is not less than or equal to [`value`][(c).],
             when [`value`][(c).] is set.
     

6. Keep examples and wording aligned with real control files.

   • Prefer concrete property names such as min, max, value, start_value, end_value, min_lines, max_lines.
   • For cross-field rules, use same-class links: [min][(c).], [max][(c).].

Cross-Referencing Conventions

Follow the cross-reference guidance in: docs-conventions.

Most common pattern to use in control property docstrings:

• same-class properties: [\prop`][(c).]`

Keep symbol labels wrapped in backticks.

Required Test Matrix

When adding/changing validation, include tests that cover:

• one valid case and one invalid case per new logical rule;
• boundary values (==, min/max edges) where applicable;
• cross-field set/unset combinations when optional values are involved;
• effective-default behavior when Dart applies defaults but Python value is omitted.

Prefer placing tests in:

• sdk/python/packages/flet/tests/test_validation.py for validation runtime;
• control-specific tests when behavior is tied to one control.

Common Pitfalls

• Keeping stale manual before_update() validations after migration.
• Missing Raises entries on validated properties.
• Combining multiple non-between rules into one ambiguous ValueError sentence.
• Forgetting Dart defaults that still apply when Python property is omitted.
• Forcing complex V.ensure(lambda ...) expressions when before_update() would be cleaner.