+2001-10-02 Miguel de Icaza <miguel@ximian.com>
+
+ * expression.cs (TypeOf::Emit): Implement.
+
+ * typemanager.cs: runtime_handle_type, new global type.
+
+ * class.cs (Property::Emit): Generate code for properties.
+
+2001-10-02 Ravi Pratap <ravi@ximian.com>
+
+ * expression.cs (Unary::ResolveOperator): Find operators on base type
+ too - we now conform exactly to the spec.
+
+ (Binary::ResolveOperator): Same here.
+
+ * class.cs (Operator::Define): Fix minor quirk in the tests.
+
+ * ../errors/cs0215.cs : Added.
+
+ * ../errors/cs0556.cs : Added.
+
+ * ../errors/cs0555.cs : Added.
+
+2001-10-01 Miguel de Icaza <miguel@ximian.com>
+
+ * cs-tokenizer.cs: Reimplemented Location to be a struct with a
+ single integer which is really efficient
+
+2001-10-01 Ravi Pratap <ravi@ximian.com>
+
+ * expression.cs (Expression::ImplicitUserConversion): Use location
+ even in the case when we are examining True operators.
+
+ * class.cs (Operator::Define): Perform extensive checks to conform
+ with the rules for operator overloading in the spec.
+
+ * expression.cs (Expression::ImplicitReferenceConversion): Implement
+ some of the other conversions mentioned in the spec.
+
+ * typemanager.cs (array_type): New static member for the System.Array built-in
+ type.
+
+ (cloneable_interface): For System.ICloneable interface.
+
+ * driver.cs (Driver::Driver): Initialize TypeManager's core types even before
+ we start resolving the tree and populating types.
+
+ * ../errors/errors.txt : Update for error numbers -7, -8, -9, -10
+
+2001-10-01 Miguel de Icaza <miguel@ximian.com>
+
+ * expression.cs (Expression::ExprClassFromMemberInfo,
+ Expression::Literalize): Create literal expressions from
+ FieldInfos which are literals.
+
+ (ConvertNumericExplicit, ImplicitNumericConversion): Fix a few
+ type casts, because they were wrong. The test suite in tests
+ caught these ones.
+
+ (ImplicitNumericConversion): ushort to ulong requires a widening
+ cast.
+
+ Int32 constant to long requires widening cast as well.
+
+ * literal.cs (LongLiteral::EmitLong): Do not generate i4 constants
+ for integers because the type on the stack is not i4.
+
+2001-09-30 Miguel de Icaza <miguel@ximian.com>
+
+ * expression.cs (report118): require location argument.
+
+ * parameter.cs: Do not dereference potential null value.
+
+ * class.cs: Catch methods that lack the `new' keyword when
+ overriding a name. Report warnings when `new' is used without
+ anything being there to override.
+
+ * modifiers.cs: Handle `NEW' as MethodAttributes.NewSlot.
+
+ * class.cs: Only add constructor to hashtable if it is non-null
+ (as now constructors can fail on define).
+
+ (TypeManager, Class, Struct): Take location arguments.
+
+ Catch field instance initialization in structs as errors.
+
+ accepting_filter: a new filter for FindMembers that is static so
+ that we dont create an instance per invocation.
+
+ (Constructor::Define): Catch errors where a struct constructor is
+ parameterless
+
+ * cs-parser.jay: Pass location information for various new
+ constructs.
+
+ * delegate.cs (Delegate): take a location argument.
+
+ * driver.cs: Do not call EmitCode if there were problesm in the
+ Definition of the types, as many Builders wont be there.
+
+ * decl.cs (Decl::Decl): Require a location argument.
+
+ * cs-tokenizer.cs: Handle properly hex constants that can not fit
+ into integers, and find the most appropiate integer for it.
+
+ * literal.cs: Implement ULongLiteral.
+
+ * rootcontext.cs: Provide better information about the location of
+ failure when CreateType fails.
+
+2001-09-29 Miguel de Icaza <miguel@ximian.com>
+
+ * rootcontext.cs (RootContext::PopulateTypes): Populates structs
+ as well.
+
+ * expression.cs (Binary::CheckShiftArguments): Add missing type
+ computation.
+ (Binary::ResolveOperator): Add type to the logical and and logical
+ or, Bitwise And/Or and Exclusive Or code paths, it was missing
+ before.
+
+ (Binary::DoNumericPromotions): In the case where either argument
+ is ulong (and most signed types combined with ulong cause an
+ error) perform implicit integer constant conversions as well.
+
+2001-09-28 Miguel de Icaza <miguel@ximian.com>
+
+ * expression.cs (UserImplicitCast): Method should always be
+ non-null.
+ (Invocation::BetterConversion): Simplified test for IntLiteral.
+
+ (Expression::ImplicitNumericConversion): Split this routine out.
+ Put the code that performs implicit constant integer conversions
+ here.
+
+ (Expression::Resolve): Become a wrapper around DoResolve so we can
+ check eclass and type being set after resolve.
+
+ (Invocation::Badness): Remove this dead function
+
+ (Binary::ResolveOperator): Do not compute the expensive argumnets
+ unless we have a union for it.
+
+ (Probe::Emit): Is needs to do an isinst and then
+ compare against null.
+
+ (::CanConvert): Added Location argument. If the Location argument
+ is null (Location.Null), then we do not report errors. This is
+ used by the `probe' mechanism of the Explicit conversion. We do
+ not want to generate an error for something that the user
+ explicitly requested to be casted. But the pipeline for an
+ explicit cast first tests for potential implicit casts.
+
+ So for now, if the Location is null, it means `Probe only' to
+ avoid adding another argument. Might have to revise this
+ strategy later.
+
+ (ClassCast): New class used to type cast objects into arbitrary
+ classes (used in Explicit Reference Conversions).
+
+ Implement `as' as well.
+
+ Reverted all the patches from Ravi below: they were broken:
+
+ * The use of `level' as a mechanism to stop recursive
+ invocations is wrong. That was there just to catch the
+ bug with a strack trace but not as a way of addressing
+ the problem.
+
+ To fix the problem we have to *understand* what is going
+ on and the interactions and come up with a plan, not
+ just get things going.
+
+ * The use of the type conversion cache that I proposed
+ last night had an open topic: How does this work across
+ protection domains. A user defined conversion might not
+ be public in the location where we are applying the
+ conversion, a different conversion might be selected
+ (ie, private A->B (better) but public B->A (worse),
+ inside A, A->B applies, but outside it, B->A will
+ apply).
+
+ * On top of that (ie, even if the above is solved),
+ conversions in a cache need to be abstract. Ie, `To
+ convert from an Int to a Short use an OpcodeCast', not
+ `To convert from an Int to a Short use the OpcodeCast on
+ the variable 5' (which is what this patch was doing).
+
+2001-09-28 Ravi Pratap <ravi@ximian.com>
+
+ * expression.cs (Invocation::ConversionExists): Re-write to use
+ the conversion cache
+
+ (Expression::ConvertImplicit): Automatic bailing out if level != 0. Also
+ cache all conversions done, not just user-defined ones.
+
+ (Invocation::BetterConversion): The real culprit. Use ConversionExists
+ to determine if a conversion exists instead of acutually trying to
+ perform the conversion. It's faster too.
+
+ (Expression::ConvertExplicit): Modify to use ConversionExists to check
+ and only then attempt the implicit conversion.
+
+2001-09-28 Ravi Pratap <ravi@ximian.com>
+
+ * expression.cs (ConvertImplicit): Use a cache for conversions
+ already found. Check level of recursion and bail out if necessary.
+
+2001-09-28 Miguel de Icaza <miguel@ximian.com>
+
+ * typemanager.cs (string_concat_string_string, string_concat_object_object):
+ Export standard methods that we expect for string operations.
+
+ * statement.cs (Block::UsageWarning): Track usage of variables and
+ report the errors for not used variables.
+
+ * expression.cs (Conditional::Resolve, ::Emit): Implement ?:
+ operator.
+
+2001-09-27 Miguel de Icaza <miguel@ximian.com>
+
+ * codegen.cs: remove unnneded code
+
+ * expression.cs: Removed BuiltinTypeAccess class
+
+ Fix the order in which implicit conversions are
+ done.
+
+ The previous fixed dropped support for boxed conversions (adding a
+ test to the test suite now)
+
+ (UserImplicitCast::CanConvert): Remove test for source being null,
+ that code is broken. We should not feed a null to begin with, if
+ we do, then we should track the bug where the problem originates
+ and not try to cover it up here.
+
+ Return a resolved expression of type UserImplicitCast on success
+ rather than true/false. Ravi: this is what I was talking about,
+ the pattern is to use a static method as a "constructor" for
+ objects.
+
+ Also, do not create arguments until the very last minute,
+ otherwise we always create the arguments even for lookups that
+ will never be performed.
+
+ (UserImplicitCast::Resolve): Eliminate, objects of type
+ UserImplicitCast are born in a fully resolved state.
+
+ * typemanager.cs (InitCoreTypes): Init also value_type
+ (System.ValueType).
+
+ * expression.cs (Cast::Resolve): First resolve the child expression.
+
+ (LValue): Add new method AddressOf to be used by
+ the `&' operator.
+
+ Change the argument of Store to take an EmitContext instead of an
+ ILGenerator, because things like FieldExpr need to be able to call
+ their children expression to generate the instance code.
+
+ (Expression::Error, Expression::Warning): Sugar functions for
+ reporting errors.
+
+ (Expression::MemberLookup): Accept a TypeContainer instead of a
+ Report as the first argument.
+
+ (Expression::ResolvePrimary): Killed. I still want to improve
+ this as currently the code is just not right.
+
+ (Expression::ResolveMemberAccess): Simplify, but it is still
+ wrong.
+
+ (Unary::Resolve): Catch errors in AddressOf operators.
+
+ (LocalVariableReference::Emit, ::Store, ::AddressOf): typecast
+ index to a byte for the short-version, or the compiler will choose
+ the wrong Emit call, which generates the wrong data.
+
+ (ParameterReference::Emit, ::Store): same.
+
+ (FieldExpr::AddressOf): Implement.
+
+ * typemanager.cs: TypeManager: made public variable instead of
+ property.
+
+ * driver.cs: document --fatal.
+
+ * report.cs (ErrorMessage, WarningMessage): new names for the old
+ Error and Warning classes.
+
+ * cs-parser.jay (member_access): Turn built-in access to types
+ into a normal simplename
+
+2001-09-27 Ravi Pratap <ravi@ximian.com>
+
+ * expression.cs (Invocation::BetterConversion): Fix to cope
+ with q being null, since this was introducing a bug.
+
+ * expression.cs (ConvertImplicit): Do built-in conversions first.
+
+2001-09-27 Ravi Pratap <ravi@ximian.com>
+
+ * expression.cs (UserImplicitCast::Resolve): Fix bug.
+
+2001-09-27 Ravi Pratap <ravi@ximian.com>
+
+ * class.cs (TypeContainer::AddConstructor): Fix a stupid bug
+ I had introduced long ago (what's new ?).
+
+ * expression.cs (UserImplicitCast::CanConvert): Static method to do
+ the work of all the checking.
+ (ConvertImplicit): Call CanConvert and only then create object if necessary.
+ (UserImplicitCast::CanConvert, ::Resolve): Re-write.
+
+ (Unary::Operator): Rename Add and Subtract to Addition and Subtraction because
+ that is the right way.
+
+ (Invocation::MakeUnionSet): Convenience function to make unions of sets for
+ overloading resolution. Use everywhere instead of cutting and pasting code.
+
+ (Binary::ResolveOperator): Use MakeUnionSet.
+
+ (UserImplicitCast::CanConvert, ::Resolve): Update to take care of the case when
+ we have to convert to bool types. Not complete yet.
+
+2001-09-27 Miguel de Icaza <miguel@ximian.com>
+
+ * typemanager.cs (TypeManager::CSharpName): support ushort.
+
+ * expression.cs (Expression::TryImplicitIntConversion): Attempts
+ to provide an expression that performsn an implicit constant int
+ conversion (section 6.1.6).
+ (Expression::ConvertImplicitRequired): Reworked to include
+ implicit constant expression conversions.
+
+ (Expression::ConvertNumericExplicit): Finished.
+
+ (Invocation::Emit): If InstanceExpression is null, then it means
+ that we perform a call on this.
+
+2001-09-26 Miguel de Icaza <miguel@ximian.com>
+
+ * expression.cs (Unary::Emit): Remove some dead code.
+ (Probe): Implement Resolve and Emit for `is'.
+ (Expression::ConvertImplicitRequired): Attempt to do constant
+ expression conversions here. Maybe should be moved to
+ ConvertImplicit, but I am not sure.
+ (Expression::ImplicitLongConstantConversionPossible,
+ Expression::ImplicitIntConstantConversionPossible): New functions
+ that tell whether is it possible to apply an implicit constant
+ expression conversion.
+
+ (ConvertNumericExplicit): Started work on explicit numeric
+ conversions.
+
+ * cs-parser.jay: Update operator constants.
+
+ * parameter.cs (Parameters::GetParameterInfo): Hook up VerifyArgs
+ (Parameters::GetSignature): Hook up VerifyArgs here.
+ (Parameters::VerifyArgs): Verifies that no two arguments have the
+ same name.
+
+ * class.cs (Operator): Update the operator names to reflect the
+ ones that the spec expects (as we are just stringizing the
+ operator names).
+
+ * expression.cs (Unary::ResolveOperator): Fix bug: Use
+ MethodInfo's ReturnType instead of LookupMethodByBuilder as the
+ previous usage did only work for our methods.
+ (Expression::ConvertImplicit): Handle decimal implicit numeric
+ conversions as well.
+ (Expression::InternalTypeConstructor): Used to invoke constructors
+ on internal types for default promotions.
+
+ (Unary::Emit): Implement special handling for the pre/post
+ increment/decrement for overloaded operators, as they need to have
+ the same semantics as the other operators.
+
+ (Binary::ResolveOperator): ditto.
+ (Invocation::ConversionExists): ditto.
+ (UserImplicitCast::Resolve): ditto.
+
+2001-09-26 Ravi Pratap <ravi@ximian.com>
+
+ * expression.cs (Unary::Emit and Binary::Emit): If we have an overloaded
+ operator, return after emitting body. Regression tests pass again !
+
+ * expression.cs (ConvertImplicit): Take TypeContainer as first argument
+ (Unary::ForceConversion, Binary::ForceConversion): Ditto.
+ (Invocation::OverloadResolve): Ditto.
+ (Invocation::BetterFunction, BetterConversion, ConversionExists): Ditto.
+
+ * everywhere : update calls to the above methods accordingly.
+
+2001-09-26 Miguel de Icaza <miguel@ximian.com>
+
+ * assign.cs (Assign): Make it inherit from ExpressionStatement.
+
+ * expression.cs (ExpressionStatement): New base class used for
+ expressions that can appear in statements, so that we can provide
+ an alternate path to generate expression that do not leave a value
+ on the stack.
+
+ (Expression::Emit, and all the derivatives): We no longer return
+ whether a value is left on the stack or not. Every expression
+ after being emitted leaves a single value on the stack.
+
+ * codegen.cs (EmitContext::EmitStatementExpression): Use the
+ facilties of ExpressionStatement if possible.
+
+ * cs-parser.jay: Update statement_expression.
+
+2001-09-25 Miguel de Icaza <miguel@ximian.com>
+
+ * driver.cs: Change the wording of message
+
+2001-09-25 Ravi Pratap <ravi@ximian.com>
+
+ * expression.cs (Binary::ResolveOperator): Had forgottten to set
+ the type of the expression to the return type of the method if
+ we have an overloaded operator match ! The regression tests pass again !
+ (Unary::ResolveOperator): Ditto.
+
+ * expression.cs (Invocation::ConversionExists): Correct the member lookup
+ to find "op_Implicit", not "implicit" ;-)
+ (UserImplicitCast): New class to take care of user-defined implicit conversions.
+ (ConvertImplicit, ForceConversion): Take TypeContainer argument
+
+ * everywhere : Correct calls to the above accordingly.
+
+ * expression.cs (UserImplicitCast::Resolve, ::Emit): Implement.
+ (ConvertImplicit): Do user-defined conversion if it exists.
+
+2001-09-24 Miguel de Icaza <miguel@ximian.com>
+
+ * assign.cs: track location.
+ (Resolve): Use implicit conversions on assignment.
+
+ * literal.cs: Oops. Not good, Emit of short access values should
+ pass (Bytes) or the wrong argument will be selected.
+
+ * expression.cs (Unary::Emit): Emit code for -expr.
+
+ (Unary::ResolveOperator): Handle `Substract' for non-constants
+ (substract from zero from the non-constants).
+ Deal with Doubles as well.
+
+ (Expression::ConvertImplicitRequired): New routine that reports an
+ error if no implicit conversion exists.
+
+ (Invocation::OverloadResolve): Store the converted implicit
+ expressions if we make them
+
+2001-09-24 Ravi Pratap <ravi@ximian.com>
+
+ * class.cs (ConstructorInitializer): Take a Location argument.
+ (ConstructorBaseInitializer): Same here.
+ (ConstructorThisInitializer): Same here.
+
+ * cs-parser.jay : Update all calls accordingly.
+
+ * expression.cs (Unary, Binary, New): Take location argument.
+ Update accordingly everywhere.
+
+ * cs-parser.jay : Update all calls to the above to take a location
+ argument.
+
+ * class.cs : Ditto.
+
+2001-09-24 Ravi Pratap <ravi@ximian.com>
+
+ * expression.cs (Invocation::BetterFunction): Take TypeContainer argument
+ (Invocation::BetterConversion): Same here
+ (Invocation::ConversionExists): Ditto.
+
+ (Invocation::ConversionExists): Implement.
+
+2001-09-22 Ravi Pratap <ravi@ximian.com>
+
+ * expression.cs (OverloadResolve): Improve some more to catch errors 1502 and 1503
+ Also take an additional TypeContainer argument.
+
+ * All over : Pass in TypeContainer as argument to OverloadResolve.
+
+ * typemanager.cs (CSharpName): Update to check for the string type and return
+ that too.
+
+ * expression.cs (Invocation::FullMethodDesc): New static method to return a string fully describing
+ a given method.
+
+2001-09-21 Ravi Pratap <ravi@ximian.com>
+
+ * expression.cs (Invocation::OverloadResolve): Re-write to conform more to the spec.
+ (Invocation::BetterFunction): Implement.
+ (Invocation::BetterConversion): Implement.
+ (Invocation::ConversionExists): Skeleton, no implementation yet.
+
+ Okay, things work fine !
+
+2001-09-21 Miguel de Icaza <miguel@ximian.com>
+
+ * typemanager.cs: declare and load enum_type, delegate_type and
+ void_type.
+
+ * expression.cs (Expression::Emit): Now emit returns a value that
+ tells whether a value is left on the stack or not. This strategy
+ might be reveted tomorrow with a mechanism that would address
+ multiple assignments.
+ (Expression::report118): Utility routine to report mismatches on
+ the ExprClass.
+
+ (Unary::Report23): Report impossible type/operator combination
+ utility function.
+
+ (Unary::IsIncrementableNumber): Whether the type can be
+ incremented or decremented with add.
+ (Unary::ResolveOperator): Also allow enumerations to be bitwise
+ complemented.
+ (Unary::ResolveOperator): Implement ++, !, ~, ++ and --.
+
+ (Invocation::Emit): Deal with new Emit convetion.
+
+ * All Expression derivatives: Updated their Emit method to return
+ whether they leave values on the stack or not.
+
+ * codegen.cs (CodeGen::EmitStatement): Pop values left on the
+ stack for expressions that are statements.
+
+2001-09-20 Miguel de Icaza <miguel@ximian.com>
+
+ * expression.cs (LValue): New interface. Must be implemented by
+ LValue objects.
+ (LocalVariableReference, ParameterReference, FieldExpr): Implement
+ LValue interface.
+
+ * assign.cs (Assign::Emit, Assign::Resolve): Use new LValue
+ interface for generating code, simplifies the code.
+
2001-09-20 Ravi Pratap <ravi@ximian.com>
* expression.cs (everywhere): Comment out return statements in ::Resolve