PasDantic is an experimental runtime data validation and schema library for Delphi, inspired by Python’s Pydantic.

It allows you to declare validation rules directly on Delphi classes using attributes, then validate object instances at runtime using RTTI.

PasDantic is test-driven. The tests are the primary documentation and define the supported behaviour.

Delphi is strongly typed at compile time, but many real-world inputs are not:

• JSON from APIs
• Configuration files
• AI / LLM structured output
• User input
• External systems with weak or shifting schemas

PasDantic fills this gap by providing runtime validation that is:

• Declarative
• Explicit
• Fail-fast
• Testable

The goal is similar to Pydantic:

Define what valid data looks like, then reject everything else.

You define normal Delphi classes and annotate fields or properties with validation attributes.

type
  TUser = class
  public
    [Required]
    [Email]
    Email: string;

    [UUID]
    Id: string;
  end;

No base class is required.

You validate an instance of a model using ValidateModel:

var
  U: TUser;
  R: TValidationResult;
begin
  U := TUser.Create;
  try
    U.Email := '[email protected]';
    U.Id := '{550E8400-E29B-41D4-A716-446655440000}';

    R := ValidateModel(U);

    if not R.IsValid then
      raise Exception.Create('Invalid data');
  finally
    U.Free;
  end;
end;

Validation inspects RTTI, finds attributes, and applies the appropriate rules.

PasDantic uses attributes to describe validation constraints.

These inherit from RegexAttribute:

• Email
• UUID (supports {} and non-braced forms)
• Hostname
• etc.

Example:

type
  TExample = class
  public
    [Email]
    Email: string;
  end;

Some data types cannot be validated reliably with regex.

For example:

• IPv6 (supports ::, ::1, compressed forms, IPv4-mapped addresses)

These use parser-based validation instead of regex, while keeping the same declarative style:

type
  TNetwork = class
  public
    [IPv6]
    Address: string;
  end;

This mirrors Pydantic’s philosophy: use the right tool, not regex everywhere.

• [Required]
• [Regex]
• [Email]
• [UUID] (with or without {})
• [Hostname]

• [IPv4]

• [IPv6]

  • accepts ::
  • accepts ::1
  • accepts full and compressed forms
  • rejects malformed addresses

ValidateModel returns a TValidationResult:

R.IsValid    // Boolean
R.Errors     // Collection of validation errors

This allows you to:

• Fail immediately
• Collect and report all errors
• Integrate with UI, logging, or APIs

PasDantic is designed test-first.

If you want to know:

• How an attribute behaves
• Which edge cases are supported
• What is considered valid vs invalid

👉 Read the tests

Tests/
  Test.PasDantic.RegexAttributes.pas
  Test.PasDantic.Validator.pas
  ...

If something is covered by a test, it is supported behaviour.

• Prefer clarity over cleverness
• Avoid “magic” behaviour
• Let tests define the contract
• Avoid regex where semantics matter (e.g. IPv6)
• Be explicit rather than permissive

This is a Delphi library written for Delphi developers, not a direct port of Python code.

PasDantic currently is experimental and evolving.

• APIs may change
• Coverage is growing
• Has good test coverage so confident existing functionality works as expected
• Backwards compatibility is not guaranteed yet

That said, behaviour covered by tests is considered stable.

• Array / list validation
• Nested models
• Numeric ranges
• Default values
• Schema generation (JSON Schema / OpenAPI)
• Better error messages and paths
• LLM structured-output validation helpers

MIT License.

Inspired by Python’s Pydantic, but implemented in a way that fits Delphi’s type system, RTTI, and ecosystem.