sicher 0.2.0

Validating data

New is_type(value, type): returns TRUE/FALSE and never throws for invalid data.
New parse_as(value, type): returns the parsed value or throws.
New safe_parse(value, type): returns a sicher_parse_result with success, data, issues (a data frame with path, message, expected, got and code) and error.
New validate_json(json, type): parses JSON with jsonlite and checks the result in one step (safe = TRUE returns a safe_parse() result).
New to_json_schema(type): exports a type as a JSON Schema (draft 2020-12) list, or as a JSON string with json = TRUE.

Errors now report every problem instead of stopping at the first one, and each problem includes its full location, for example todos[[2]]$title or users$id. ListOf() errors now say which element failed.
Missing fields and unexpected fields in list and data frame types are reported once per field, e.g. Type error in 'person$age': Missing required field (expected numeric).
Type errors are now conditions of class sicher_type_error with an issues element, so they can be caught with tryCatch(..., sicher_type_error = function(e) e$issues).
Using a type factory without calling it (e.g. Factor instead of Factor()) now gives a hint.

Tuple(...): fixed-length lists with a type per position.
Record(keys, values): named lists or named vectors with arbitrary keys (Record<K, V>, z.record()).
DiscriminatedUnion(key, ...): tagged unions where the key field picks the member to check, so errors point at the right fields (z.discriminatedUnion()).
Lazy(fn): recursive types such as trees (z.lazy()).
InstanceOf(class): objects inheriting from an S3, S4 or R6 class (z.instanceof()).
Brand(type, brand) and brand(value, type): nominal types, so a UserId cannot be used where an OrderId is expected.
FunctionOf(params, .return): function signatures. Checked functions are wrapped with typed_function(), so their later calls are checked too.
Factor(levels), Date, DateTime, Environment and MatrixOf(type, nrow, ncol) for common R objects. MatrixOf is named so it does not clash with the Matrix package.
Positive, NonNegative and Whole numeric types.
Intersection operator &. Two list types (or two data frame types) are merged into one type with the fields of both; any other combination must satisfy both sides.

Refine(type, predicate, message): adds a custom rule with a custom error message on top of a type.
MinLength(type, n), MaxLength(type, n) and Unique(type). Lengths are counted with length() (nrow() for data frames), like NonEmpty() and [n].
Nullable(type): like Optional(), but inside list and data frame types the field must be present (field: T | null versus field?: T).
Default(type, value): replaces NULL with a default and fills in missing list fields.
Coerce(type, using = NULL): converts values before checking them, for example "42" to 42. A conversion that would introduce NA is reported as an error.

Pick(), Omit(), Partial(), Required() and KeyOf() derive new schemas from list and data frame types.
create_list_type() and create_dataframe_type() gain an extra argument: "error" (default, unchanged behaviour), "allow" to keep unknown fields, or "drop" to remove them.
extend() now also works on data frame types and keeps the base type’s extra setting.

Typed variables and typed_function() arguments and return values now store the parsed value. This only makes a difference for types that transform values (Default(), Coerce(), extra = "drop", FunctionOf()); for every other type the stored value is exactly the assigned one.
New type_of(x) returns the declared type of a typed variable (NULL for regular variables), so one variable’s type can be reused for another: y %:% type_of(x) %<-% value.
New untype(x) turns a typed variable back into a regular variable.
New sicher.mode = "warn" option: type violations are reported as warnings and the assigned value is kept, which helps when adding sicher to existing code. Readonly() is still enforced.

Scalar() now also accepts union types.
infer_type() now infers Factor(), Date and DateTime instead of Any.
Union type names are built from their members, so unions that contain intersections or other unions print correctly.
Readonly types now have a print method (<type: readonly<string> >).
The package now declares R (>= 4.0.0) (needed for type_of()) and imports methods (used by InstanceOf() for S4 classes). jsonlite is suggested for validate_json() and to_json_schema(json = TRUE).
The hex logo was removed from the README and the website.

Adding the Literal type check
Adding the extend method to extend a preexisting list schema.
Adding NonNA() modifier — rejects any value that contains at least one NA.
Adding Between(min, max) — closed-interval numeric range constraint.
Adding Matches(pattern) — PCRE regex-constrained string type.
Adding NonEmpty() modifier — rejects zero-length vectors, lists, and zero-row data frames.