Add constraint support (#119)

* Add constraint support

* Use custom constraint macro

* Test multiple constraints

* Avoid having to deal with macro name collisions

* Avoid requiring Legolas in caller scope

* Fix tests on Julia 1.6

* Update location comment

* Add fail-safe comment

* Set project version to 0.5.21

* Increase coverage

* Add constraints into the tour

* Add docstring for `@check`

Project.toml

@@ -1,7 +1,7 @@
 name = "Legolas"
 uuid = "741b9549-f6ed-4911-9fbf-4a1c0c97f0cd"
 authors = ["Beacon Biosignals, Inc."]
-version = "0.5.20"
+version = "0.5.21"
 
 [deps]
 Arrow = "69666777-d1a9-59fb-9406-91d4454c9d45"

docs/src/index.md

@@ -14,6 +14,7 @@ CurrentModule = Legolas
 Legolas.SchemaVersion
 Legolas.@schema
 Legolas.@version
+Legolas.@check
 Legolas.is_valid_schema_name
 Legolas.parse_identifier
 Legolas.name

examples/tour.jl

@@ -267,6 +267,53 @@ end
 # of application of constraints (i.e. the parent's are applied before the child's). Lastly, `>` aligns well with the
 # property that child schema versions have a greater number of constraints than their parents.
 
+#####
+##### Constraints
+#####
+
+# Schema authors may want to restrict the allowed values for a field without having to define a new type. The `@check`
+# macro provides this functionality in a concise way and provides user friendly error messages
+
+@schema "example.finite-positive" FinitePositive
+@version FinitePositiveV1 begin
+    a::Real
+    @check a > 0
+    @check isfinite(a)
+end
+
+# We recommend defining multiple constraints instead of combining them into one (e.g. `@check a > 0 && isfinite(a)`)
+# as this makes error message clearer:
+
+@test NamedTuple(FinitePositiveV1(; a=1)) == (a=1,)
+@test_throws Legolas.CheckConstraintError(:(a > 0)) FinitePositiveV1(; a=-1)
+@test_throws Legolas.CheckConstraintError(:(isfinite(a))) FinitePositiveV1(; a=Inf)
+
+# All `@check` constraints must be defined after the fields and any processing on the field will occur before the
+# constraints are checked:
+
+@schema "example.clamp" Clamp
+@version ClampV1 begin
+    a
+    b = clamp(b, 1, 5)
+    @check a == b
+end
+
+@test NamedTuple(ClampV1(; a=1, b=0)) == (a=1, b=1)
+
+# One use case supported by constraints is enforcing mutually exclusive use of multiple fields:
+
+@schema "example.mutually-exclusive" MutuallyExclusive
+@version MutuallyExclusiveV1 begin
+    x::Union{Real,Missing}
+    y::Union{String,Missing}
+    z::Union{Char,Missing}
+    @check !ismissing(x) ⊻ !ismissing(y) ⊻ !ismissing(z)  # `⊻` is the `xor` function
+end
+
+@test_throws Legolas.CheckConstraintError MutuallyExclusiveV1(; x=1, y="hi")
+@test_throws Legolas.CheckConstraintError MutuallyExclusiveV1(; x=1, z='a')
+@test isequal(NamedTuple(MutuallyExclusiveV1(; x=1)), (x=1, y=missing, z=missing))
+
 #####
 ##### Schema Versioning
 #####

src/Legolas.jl

@@ -5,6 +5,7 @@ using Tables, Arrow, UUIDs
 const LEGOLAS_SCHEMA_QUALIFIED_METADATA_KEY = "legolas_schema_qualified"
 
 include("lift.jl")
+include("constraints.jl")
 include("schemas.jl")
 include("tables.jl")
 include("record_merge.jl")

src/constraints.jl

@@ -0,0 +1,24 @@
+struct CheckConstraintError <: Exception
+    predicate::Expr
+end
+
+function Base.showerror(io::IO, ex::CheckConstraintError)
+    print(io, "$CheckConstraintError: $(ex.predicate)")
+    return nothing
+end
+
+"""
+    @check expr
+
+Define a constraint for a schema version (e.g. `@check x > 0`) from a boolean expression.
+The `expr` should evaulate to `true` if the constraint is met or `false` if the constraint
+is violated. Multiple constraints may be defined for a schema version. All `@check`
+constraints defined with a [`@version`](@ref) must proceed all fields defined by the schema
+version.
+
+For more details and examples, please see `Legolas.jl/examples/tour.jl`.
+"""
+macro check(expr)
+    quoted_expr = QuoteNode(expr)
+    return :($(esc(expr)) || throw(CheckConstraintError($quoted_expr)))
+end

src/schemas.jl

@@ -520,7 +520,7 @@ _schema_version_from_record_type(::Nothing) = nothing
 # includes the parent's declared field RHS statements. We cannot interpolate/incorporate these statements
 # in the child's record type definition because they may reference bindings from the parent's `@version`
 # callsite that are not available/valid at the child's `@version` callsite.
-function _generate_record_type_definitions(schema_version::SchemaVersion, record_type_symbol::Symbol)
+function _generate_record_type_definitions(schema_version::SchemaVersion, record_type_symbol::Symbol, constraints::AbstractVector)
     # generate `schema_version_type_alias_definition`
     T = Symbol(string(record_type_symbol, "SchemaVersion"))
     schema_version_type_alias_definition = :(const $T = $(Base.Meta.quot(typeof(schema_version))))
@@ -616,6 +616,7 @@ function _generate_record_type_definitions(schema_version::SchemaVersion, record
             function $R(; $(field_kwargs...))
                 $parent_record_application
                 $(field_assignments...)
+                $(constraints...)
                 return new($(keys(record_fields)...))
             end
         end
@@ -625,11 +626,13 @@ function _generate_record_type_definitions(schema_version::SchemaVersion, record
             function $R{$(type_param_names...)}(; $(field_kwargs...)) where {$(type_param_names...)}
                 $parent_record_application
                 $(parametric_field_assignments...)
+                $(constraints...)
                 return new{$(type_param_names...)}($(keys(record_fields)...))
             end
             function $R(; $(field_kwargs...))
                 $parent_record_application
                 $(field_assignments...)
+                $(constraints...)
                 return new{$((:(typeof($n)) for n in names_of_parameterized_fields)...)}($(keys(record_fields)...))
             end
         end
@@ -778,8 +781,25 @@ macro version(record_type, declared_fields_block=nothing)
 
     # parse `declared_fields_block`
     declared_field_statements = Any[]
+    declared_constraint_statements = Any[]
     if declared_fields_block isa Expr && declared_fields_block.head == :block && !isempty(declared_fields_block.args)
-        declared_field_statements = [f for f in declared_fields_block.args if !(f isa LineNumberNode)]
+        for f in declared_fields_block.args
+            if f isa LineNumberNode
+                continue
+            elseif f isa Expr && f.head === :macrocall && f.args[1] === Symbol("@check")
+                constraint_expr = Base.macroexpand(Legolas, f)
+                # Update the expression such that a failure shows the location of the user
+                # defined `@check` call. Ideally `Meta.replace_sourceloc!` would do this.
+                if f.args[2] isa LineNumberNode
+                    constraint_expr = Expr(:block, f.args[2], constraint_expr)
+                end
+                push!(declared_constraint_statements, constraint_expr)
+            elseif isempty(declared_constraint_statements)
+                push!(declared_field_statements, f)
+            else
+                return :(throw(SchemaVersionDeclarationError("all `@version` field expressions must be defined before constraints:\n", $(Base.Meta.quot(declared_fields_block)))))
+            end
+        end
     end
     declared_field_infos = DeclaredFieldInfo[]
     for stmt in declared_field_statements
@@ -800,6 +820,7 @@ macro version(record_type, declared_fields_block=nothing)
         return :(throw(SchemaVersionDeclarationError($msg)))
     end
     declared_field_names_types = Expr(:tuple, Expr(:parameters, (Expr(:kw, f.name, esc(f.type)) for f in declared_field_infos)...))
+    constraints = [Base.Meta.quot(ex) for ex in declared_constraint_statements]
 
     return quote
         if !isdefined((@__MODULE__), :__legolas_schema_name_from_prefix__)
@@ -827,7 +848,7 @@ macro version(record_type, declared_fields_block=nothing)
                 Base.@__doc__($(Base.Meta.quot(record_type)))
                 $(esc(:eval))($Legolas._generate_schema_version_definitions(schema_version, parent, $declared_field_names_types, schema_version_declaration))
                 $(esc(:eval))($Legolas._generate_validation_definitions(schema_version))
-                $(esc(:eval))($Legolas._generate_record_type_definitions(schema_version, $(Base.Meta.quot(record_type))))
+                $(esc(:eval))($Legolas._generate_record_type_definitions(schema_version, $(Base.Meta.quot(record_type)), [$(constraints...)]))
             end
         end
         nothing

test/runtests.jl

@@ -1,6 +1,7 @@
 using Compat: current_exceptions
 using Legolas, Test, DataFrames, Arrow, UUIDs
-using Legolas: SchemaVersion, @schema, @version, SchemaVersionDeclarationError, DeclaredFieldInfo
+using Legolas: @schema, @version, CheckConstraintError, SchemaVersion,
+               SchemaVersionDeclarationError, DeclaredFieldInfo
 using Accessors
 using Aqua
 
@@ -817,3 +818,76 @@ end
     @test r.i isa UInt16
     @test r.i == 1
 end
+
+#####
+##### constraints
+#####
+
+@schema "test.constraint" Constraint
+
+const CONSTRAINT_V1_EQUAL_CONSTRAINT_LINE = @__LINE__() + 4
+@version ConstraintV1 begin
+    a
+    b = clamp(b, 0, 5)
+    @check a == b
+    @check a > 0
+end
+
+@testset "constraints" begin
+    r = ConstraintV1(; a=1, b=1)
+    @test r isa ConstraintV1
+    @test r.a === 1
+    @test r.b === 1
+
+    r = ConstraintV1(; a=1, b=1.0)
+    @test r isa ConstraintV1
+    @test r.a === 1
+    @test r.b === 1.0
+
+    # In Julia 1.8+ we can test can test against "CheckConstraintError: a == b"
+    try
+        ConstraintV1(; a=1, b=2)
+        @test false  # Fail safe if the above line doesn't throw
+    catch e
+        @test e isa CheckConstraintError
+        @test e.predicate == :(a == b)
+    end
+
+    try
+        ConstraintV1(; a=0, b=0)
+        @test false  # Fail safe if the above line doesn't throw
+    catch e
+        @test e isa CheckConstraintError
+        @test e.predicate == :(a > 0)
+    end
+
+    try
+        ConstraintV1(; a=6, b=6)
+        @test false  # Fail safe if the above line doesn't throw
+    catch e
+        @test e isa CheckConstraintError
+        @test e.predicate == :(a == b)
+    end
+
+    # For exceptions that occur during processing constraints its convenient to include the
+    # location of the `@check` in the stacktrace.
+    try
+        ConstraintV1(; a=1, b=missing)  # Fails when running check `a == b`
+        @test false  # Fail safe if the above line doesn't throw
+    catch e
+        @test e isa TypeError
+
+        bt = Base.process_backtrace(catch_backtrace())
+        sf = bt[1][1]::Base.StackFrame
+        @test string(sf.file) == @__FILE__
+        @test sf.line == CONSTRAINT_V1_EQUAL_CONSTRAINT_LINE
+    end
+end
+
+@testset "constraints must be after all fields" begin
+    @test_throws SchemaVersionDeclarationError @version(ConstraintV2, begin a; @check a == 1; b end)
+end
+
+@testset "CheckConstraintError" begin
+    @test sprint(showerror, CheckConstraintError(:(1 == 2))) == "CheckConstraintError: 1 == 2"
+end