An error against the Turbo Pascal language was encountered. This typically happens
when an illegal character is found in the source file.
The specified procedure directive is ignored by FPC programs.
When you want to use overloading using the OVERLOAD directive, then all declarations
need to have OVERLOAD specified.
Exported function names inside a specific DLL must all be different.
Exported function indexes inside a specific DLL must all be different.
It is currently not possible to include debug information in a relocatable DLL.
Stabs debug info is wrong for relocatable DLL or EXES. Use -WN if you want to debug
win32 executables.
You are declaring an object constructor with a name which is not init, and the -Ss
switch is in effect. See the switch -Ss (see page 129).
You are declaring an object destructor with a name which is not done, and the -Ss
switch is in effect. See the switch -Ss (see page 129).
You tried to compile a program with C++ style inlining, and forgot to specify the
-Si option (-Si (see page 129)). The compiler doesn’t support C++ styled inlining by
default.
Constructors must be in the ’public’ part of an object (class) declaration.
Destructors must be in the ’public’ part of an object (class) declaration.
Classes must be defined globally. They cannot be defined inside a procedure or function.
An invalid object (class) declaration was encountered, i.e. an object or class without methods
that isn’t derived from another object or class. For example:
Type o = object
a : longint;
end;
This is a note indicating that the declared object has no virtual method table.
You are calling a function with parameters that are of a different type than the declared
parameters of the function.
There is an error in the parameter list of the function or procedure – the number of
parameters is not correct.
The compiler encountered a symbol with the same name as an overloaded function, but it is
not a function it can overload.
You’re declaring overloaded functions, but with the same parameter list. Overloaded function
must have at least 1 different parameter in their declaration.
You declared a function with the same parameters but different result type or function
modifiers.
You declared the function in the interface part, or with the forward directive, but defined
it with a different parameter list.
Free Pascal allows enumeration constructions as in C. Examine the following two
declarations:
type a = (A_A,A_B,A_E:=6,A_UAS:=200);
type a = (A_A,A_B,A_E:=6,A_UAS:=4);
The second declaration would produce an error. A_UAS needs to have a value higher than A_E, i.e. at least 7.
With stores a variable locally on the stack, but this is not possible if the variable belongs to
another segment.
You are specifying the same label 2 times in a case statement.
The upper bound of a case label is less than the lower bound and this is useless.
You are trying to assign an overloaded function to a procedural variable. This is not allowed.
The length of a shortstring in Pascal is limited to 255 characters. You are trying to declare a
string with length less than 1 or greater than 255.
If you have a pointer a to an object type, then the statement new(a) will not initialize the
object (i.e. the constructor isn’t called), although space will be allocated. You should issue
the new(a,init) statement. This will allocate space, and call the constructor of the object.
You cannot use new(p) or dispose(p) if p is an untyped pointer because no size
is associated to an untyped pointer. It is accepted for compatibility in TP and
DELPHI modes, but the compiler will still warn you if it finds such a construct.
This happens when the compiler scans a procedure declaration that contains a dot, i.e.,
an object or class method, but the type in front of the dot is not a known type.
This identifier is not a method. This happens when the compiler scans a procedure
declaration that contains a dot, i.e., an object or class method, but the procedure name is not
a procedure of this type.
This identifier is not a method. This happens when the compiler scans a procedure
declaration that contains a dot, i.e., an object or class method, but the procedure name is not
a procedure of this type.
When using the -vd switch, the compiler tells you when it starts processing a procedure or
function implementation.
The compiler expects a floating point expression, and gets something else.
You are using the fail keyword outside a constructor method.
You are declaring a destructor with a parameter list. Destructor methods cannot have
parameters.
This error occurs in a situation like the following:
Type :
Tclass = Class of Tobject;
Var C : TClass;
begin
...
C.free
Free is not a class method and hence cannot be called with a class reference.
This is related to the previous error. You cannot call a method of an object from inside a
class method. The following code would produce this error:
class procedure tobject.x;
begin
free
Because free is a normal method of a class it cannot be called from a class method.
One of the labels is not of the same type as the case variable.
You can only export procedures and functions when you write a library. You cannot export
variables or constants.
A method that is declared virtual in a parent class, should be overridden in the descendant
class with the override directive. If you don’t specify the override directive, you will hide
the parent method; you will not override it.
You are trying to override a virtual method of a parent class that does not exist.
This message is no longer used, as the stored directive has been implemented.
There is an error in the read or write directives for an array property. When you declare an
array property, you can only access it with procedures and functions. The following code
would cause such an error.
tmyobject = class
i : integer;
property x [i : integer]: integer read I write i;
Fields that are declared in a protected section of an object or class declaration cannot be
accessed outside the module where the object is defined, or outside descendent object
methods.
Fields that are declared in a private section of an object or class declaration cannot be
accessed outside the module where the class is defined.
If you declare overridden methods in a class definition, they must have the same return type.
You cannot declare a function or procedure within a function or procedure that was declared
as an export procedure.
You cannot declare a procedure that is a method for an object as exported.
When calling a function declared with var parameters, the variables in the function
call must be of exactly the same type. There is no automatic type conversion.
When calling inherited methods, you are trying to call a method of a non-related class. You
can only call an inherited method of a parent class.
You are trying to use the self parameter outside an object’s method. Only methods get
passed the self parameters.
A construction like sometype.somemethod is only allowed in a method.
You are using the format : (colon) 2 times on an expression that is not a real expression.
The declaration of a set contains an error. Either one of the elements is outside the range of
the set type, or two of the elements are in fact the same.
You specified an illegal type in a new statement. The extended syntax of new needs an object
as a parameter.
When using the extended syntax of new, you must specify the constructor method of the
object you are trying to create. The procedure you specified is not a constructor.
When using the extended syntax of dispose, you must specify the destructor method of the
object you are trying to dispose of. The procedure you specified is not a destructor.
When declaring a constant record, you specified the fields in the wrong order.
A with statement needs an argument that is of the type record or class. You are using
with on an expression that is not of this type.
In Free Pascal, you can specify a return value for a function when using the exit statement.
This error occurs when you try to do this with a procedure. Procedures cannot return a
value.
You’re declaring a procedure as destructor, constructor or class operator, when the procedure
isn’t a class method.
You’re trying to use an overloaded operator when it is not overloaded for this type.
You cannot overload assignment for types that the compiler considers as equal.
The combination of operator, arguments and return type are incompatible.
You are trying to re-raise an exception where it is not allowed. You can only re-raise
exceptions in an except block.
You cannot generate an instance of a class with the extended syntax of new. The constructor
must be used for that. For the same reason, you cannot call dispose to de-allocate an
instance of a class, the destructor must be used for that.
When using the -So switch, procedure overloading is switched off. Turbo Pascal does not
support function overloading.
You are trying to overload an operator which cannot be overloaded. The following operators
can be overloaded :
+, -, *, /, =, >, <, <=, >=, is, as, in, **, :=
When overloading the = operator, the function must return a boolean value.
You are declaring a method as abstract, when it is not declared to be virtual.
You’re trying to force the compiler into doing something it cannot do yet.
You cannot derive objects, classes, cppclasses and interfaces intertwined. E.g. a class
cannot have an object as parent and vice versa.
You cannot specify more than one variable before the absolute, export, external,
weakexternal, public and cvar directives. As a result, for example the following construct
will provide this error:
Var Z : Longint;
X,Y : Longint absolute Z;
The address of an absolute directive can only point to a variable or constant. Therefore, the
following code will produce this error:
Procedure X;
var p : longint absolute x;
You cannot specify more than one variable with an initial value.
Abstract methods can only be declared, you cannot implement them. They should be
overridden by a descendant class.
You are defining an overloaded function in the implementation part of a unit, but there is no
corresponding declaration in the interface part of the unit.
If you declare objects or classes that contain virtual methods, you need to have a
constructor and destructor to initialize them. The compiler encountered an object
or class with virtual methods that doesn’t have a constructor/destructor pair.
When -vc is used, the compiler tells you when it defines macros.
When -vc is used, the compiler tells you when it undefines macros.
When -vc is used, the compiler tells you what values macros get.
When you turn on information messages (-vi), the compiler tells you what units it is
recompiling.
This tells you that the reading of the interface of the current unit has started
This tells you that the code reading of the implementation of the current unit, library or
program starts
When you request debug messages (-vd) the compiler tells you what units it recompiles for
the second time.
You want to override a property of a parent class, when there is, in fact, no such property in
the parent class.
You specified a property as Default, but the class already has a default property, and a class
can have only one default property.
Only array properties of classes can be made default properties.
You cannot have virtual constructors in objects. You can only have them in classes.
You are trying to access a default property of a class, but this class (or one of its ancestors)
doesn’t have a default property.
If you want a published section in a class definition, you must use the {$M+} switch, which
turns on generation of type information.
To be able to use an object as an ancestor object, it must be defined first. This error occurs
in the following situation:
Type ParentClas = Class;
ChildClass = Class(ParentClass)
...
end;
You cannot overload locally, i.e. inside procedures or function definitions.
This procedure directive is not allowed in the interface section of a unit. You can only use
it in the implementation section.
This procedure directive is not allowed in the implementation section of a unit. You can
only use it in the interface section.
This procedure directive cannot be part of a procedural or function type declaration.
You will get this error if a function is defined as forward twice. Or if it occurs in the
interface section, and again as a forward declaration in the implementation section.
When information messages are on, the compiler warns you when it writes the browser log
(generated with the {$Y+ } switch).
The selected assembler reader (with {$ASMMODE xxx} is not supported. The compiler can be
compiled with or without support for a particular assembler reader.
You specified a procedure directive that conflicts with other directives. For instance cdecl
and pascal are mutually exclusive.
This error happens when you declare a function or procedure with e.g. cdecl; but
omit this directive in the implementation, or vice versa. The calling convention is
part of the function declaration, and must be repeated in the function definition.
Set properties or indexed properties cannot have a default value.
The value of a default declared property must be known at compile time. The value you
specified is only known at run time. This happens e.g. if you specify a variable name as a
default value.
Only class type variables can be in a published section of a class if they are not declared as
a property.
Properties in a published section cannot be array properties. They must be moved to public
sections. Properties in a published section must be an ordinal type, a real type, strings or
sets.
Some targets need a name for the imported procedure or a cdecl specifier.
An operation on two real type values produced an overflow or a division by zero.
The upper bound of an array declaration is less than the lower bound and this is not
possible.
The size of the constant string is larger than the size you specified in string type definition.
The size of the constant string is larger than the size you specified in the Array[x..y] of
char definition.
Free Pascal supports only integer or string values as message constants.
A method declared with the message directive as message handler can take only one
parameter which must be declared as call by reference. Parameters are declared as call by
reference using the var-directive.
The Self parameter can only be passed explicitly to a method which is declared as message
handler.
Threadvars must be static or global; you cannot declare a thread local to a procedure. Local
variables are always local to a thread, because every thread has its own stack and local
variables are stored on the stack.
You cannot use direct assembler when using a binary writer. Choose an other output format
or use another assembler reader.
You are trying to load the ObjPas unit manually from a uses clause. This is not a good idea.
Use the {$MODE OBJFPC} or {$mode delphi} directives which load the unit automatically.
Override is not supported for objects, use virtual instead to override a method of a parent
object.
Some data types (e.g. ansistring) need initialization/finalization code which is implicitly
generated by the compiler. Such data types cannot be used in the variant part of a record.
Resourcestring cannot be declared local, only global or using the static directive.
An exit statement with an argument for the return value cannot be used here. This can
happen for example in try..except or try..finally blocks.
If you specify a storage symbol in a property declaration, it must be a boolean type.
You cannot use this type of symbol as storage specifier in property declaration. You can use
only methods with the result type boolean, boolean class fields or boolean constants.
A class-typed field in the published section of a class can only be a class which was compiled
in {$M+} or which is derived from such a class. Normally such a class should be derived from
TPersistent.
This error is triggered when you have a {$Calling} directive without a calling convention
specified. It also happens when declaring a procedure in a const block and you used a ; after
a procedure declaration which must be followed by a procedure directive. Correct
declarations are:
const
p : procedure;stdcall=nil;
p : procedure stdcall=nil;
The value you use to index a property must be of an ordinal type, for example an integer or
enumerated type.
The length of the procedure/function name must be at least 2 characters long. This is
because of a bug in dlltool which doesn’t parse the .def file correctly with a name of length 1.
You need to compile this file without the -WD switch on the command line.
You need to use {$MODE OBJFPC} or {$MODE DELPHI} to compile this file. Or use the
corresponding command line switch, either -Mobjfpc or -MDelphi.
Exporting of functions or procedures with a specified index is not supported on this target.
The GUID indication does not have the proper syntax. It should be of the form
{XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX}
The compiler cannot find a suitable procedure which implements the given method of an
interface. A procedure with the same name is found, but the arguments do not match.
This happens when the compiler scans a class declaration that contains interface function
name mapping code like this:
type
TMyObject = class(TObject, IDispatch)
function IUnknown.QueryInterface=MyQueryInterface;
....
and the interface before the dot is not listed in the inheritance list.
Types like qword or int64 are not allowed as array index type.
Constructor and destructor declarations are not allowed in interfaces. In the most
cases method QueryInterface of IUnknown can be used to create a new interface.
The access specifiers public, private, protected and published cannot be used in interfaces,
Objective-C protocols and categories because all methods of an interface/protocol/category
must be public.
Declarations of fields are not allowed in interfaces, helpers and Objective-C protocols and
categories. An interface/helper/protocol/category can contain only methods and properties
with method read/write specifiers.
Declaring local procedures as external is not possible. Local procedures get hidden
parameters that will make the chance of errors very high.
In Delphi mode, not all fields of a typed constant record have to be initialized, but the
compiler warns you when it detects such situations.
In all syntax modes but Delphi mode, you cannot leave some fields uninitialized in the middle
of a typed constant record.
You can leave some fields at the end of a type constant record uninitialized (The compiler
will initialize them to zero automatically). This may be the cause of subtle problems.
The varargs directive (or the “...” varargs parameter in MacPas mode) can only be used with
procedures or functions that are declared with external and one of cdecl, cppdecl,
stdcall and mwpascal. This functionality is only supported to provide a compatible
interface to C functions like printf.
You cannot declare Self as a const or var parameter, it must always be a call-by-value
parameter.
When you want to assign an interface to a constant, then the interface must have a GUID
value set.
Properties must refer to a field or method in the same class.
There are two directives in the procedure declaration that specify a calling convention. Only
the last directive will be used.
In order to initialize a method pointer with a method, the value of the Self pointer for
calling that method at run time must be known at compile time. Thus, a method pointer can
be initialized either with Nil, or with a class method that is accessed via a class type or a
class reference type.
It is not possible to specify a default value for several parameters at once. The following is
invalid:
Procedure MyProcedure (A,B : Integer = 0);
Instead, this should be declared as
Procedure MyProcedure (A : Integer = 0; B : Integer = 0);
You’re trying to force the compiler into doing something it cannot do yet.
Any array passed to a C function is passed by a pointer (i.e. by reference).
You cannot add any other argument after an array of const for cdecl functions, as the
size pushed on stack for this argument is not known.
This is an indicator that a previously declared type is being redefined as something else. This
may, or may not be, a potential source of errors.
Functions declared with the cdecl modifier do not pass an extra implicit parameter.
Openstring is not supported for functions that have the cdecl modifier.
Variables declared as threadvar cannot be initialized with a default value. The variables will
always be filled with zero at the start of a new thread.
A class method can only be specified for procedures and functions.
Some calling conventions are supported only by certain CPUs. I.e. most non-i386 ports
support only the standard ABI calling convention of the CPU.
reintroduce is not supported for objects, Objective-C classes and Objective-C protocols.
If locations for arguments are specified explicitly as it is required by some syscall conventions,
each argument must have its own location. Things like
procedure p(i,j : longint 'r1');
If one argument has an explicit argument location, all arguments of a procedure must have
one.
Syscalls specific: the specified explicit location string for this parameter cannot be parsed,
invalid, or The location specified for an argument isn’t recognized by the compiler.
The libbase for MorphOS/AmigaOS can be given only as longint, dword or any pointer
variable.
It isn’t allowed to use goto statements referencing labels outside the current procedure. The
following example shows the problem:
...
procedure p1;
label
l1;
procedure p2;
begin
goto l1; // This goto ISN'T allowed
end;
begin
p2
l1:
end;
...
Your procedure body is too long for the compiler. You should split the procedure into
multiple smaller procedures.
This can occur under many circumstances. Usually when trying to evaluate constant
expressions.
You made an expression which isn’t an integer, and the compiler expects the result to be an
integer.
One of the following is happening :
-
You’re trying to access a field of a variable that is not a record.
-
You’re indexing a variable that is not an array.
-
You’re dereferencing a variable that is not a pointer.
You are declaring a subrange, and the high limit is less than the low limit of the range.
The parameter of a exit call in macpas mode must be either the name of the current
subroutine or of a surrounding one
The type of a for loop variable must be an ordinal type. Loop variables cannot be reals or
strings. You also cannot assign values to loop variables inside the loop (Except in Delphi and
TP modes). Use a while or repeat loop instead if you need to do something like that, since
those constructs were built for that.
Declaring local variables as external is not allowed. Only global variables can reference
external variables.
The procedure is already declared with the EXTERNAL directive in an interface or forward
declaration.
The Variant type is used in the unit without any used unit using the Variants unit. The
compiler has implicitly added the Variants unit to the uses list. To remove this warning the
Variants unit needs to be added to the uses statement.
The specifier class and directive static cannot be used in interfaces because all methods of
an interface must be public.
slice can be used only for arguments accepting an open array parameter.
A DISPINTERFACE is a special type of interface which cannot have a parent class.
Dispinterface always derive from IDispatch type.
A DISPINTERFACE always needs an interface identification (a GUID).
If you declare overridden methods in a class definition, they must have the same return type.
Some versions of Delphi allow you to change the return type of interface methods, and even
to change procedures into functions, but the resulting code may crash depending on the types
used and the way the methods are called.
The dispid keyword must be followed by an ordinal constant (the dispid index).
Regardless of the size taken up by its elements, an array cannot have more than
high(ptrint) elements. Additionally, the range type must be a subrange of ptrint.
If you declare an array or record as packed in Mac Pascal mode (or as packed in any mode
with {$bitpacking on}), it will be packed at the bit level. This means it becomes impossible
to take addresses of individual array elements or record fields. The only exception to this rule
is in the case of packed arrays elements whose packed size is a multiple of 8 bits.
Only regular (and possibly in the future also open) arrays can be packed.
If you declare an array or record as packed in Mac Pascal mode (or as packed in any mode
with {$bitpacking on}), it will be packed at the bit level. For performance reasons, they
cannot be used as loop variables.
The usage of VAR, TYPE and CONST to declare new types inside an object is allowed only
inside records, objects and classes.
Only Classes, Objects, Interfaces and Records are allowed to be used as generic.
Do not use the lineinfo unit directly, Use the -gl switch which automatically adds the correct
unit for reading the selected type of debugging information. The unit that needs to be
used depends on the type of debug information used when compiling the binary.
The first time you declare a function you have to declare it completely, including all
parameters and the result type.
When specializing a generic, only non-generic types can be used as parameters.
If an object requires a VMT either because it contains a constructor or virtual methods, it’s
not allowed to create constants of it. In TP and Delphi mode this is allowed for compatibility
reasons.
It isn’t allowed to take the address of labels outside the current procedure.
Variables declared as external cannot be initialized with a default value.
Some types like file types cannot be used as function result.
To perform an operation on integers, the compiler converts both operands to their common
type, which appears to be an invalid type. To determine the common type of the operands,
the compiler takes the minimum of the minimal values of both types, and the maximum of
the maximal values of both types. The common type is then minimum..maximum.
Generics must be always specialized before being used as variable type.
When using pure assembler routines, the list with modified registers is ignored.
A property which implements an interface must be of type class or interface.
A property which implements an interface actually implements a different interface.
A property which implements an interface must have at least a read specifier.
A property which implements an interface may not have a write specifier.
A property which implements an interface may not have a stored specifier.
The interface which is implemented by a property is not an interface implemented by the
class.
The compiler parsed a floating point expression, but it is not supported.
The delegated interface is not implemented by the class given in the implements clause.
The implements keyword must be followed by an interface type.
On most targets it is not possible to change the name under which a variable is exported
inside the exports statement of a library. In that case, you have to specify the export name
at the point where the variable is declared, using the export and alias directives.
A ”weak external” symbol is a symbol which may or may not exist at (either static or
dynamic) link time. This concept may not be available (or implemented yet) on the current
cpu/OS target.
Classes and interfaces being defined forward must have the same type when being
implemented. A forward interface cannot be changed into a class.
The virtual method overrides an method that is declared with a higher visibility. This might
give unexpected results. E.g., in case the new visibility is private then a call to “inherited” in
a new child class will call the higher-visible method in a parent class and ignores the private
method.
Once a method or property has been defined in a class or object, you cannot define any fields
afterwards without starting a new visibility section (such as public, private, etc.). The
reason is that otherwise the source code can appear ambiguous to the compiler, since it
is possible to use modifiers such as default and register also as field names.
In Pascal, types are not considered to be identical simply because they are semantically
equivalent. Two variables or parameters are only considered to be of the same type if they
refer to the same type definition. As a result, it is not allowed to define new types inside
parameter lists, because then it is impossible to refer to the same type definition in the
procedure headers of the interface and implementation of a unit (both procedure headers
would define a separate type). Keep in mind that expressions such as “file of byte” or
“string[50]” also define a new type.
ABSTRACT and SEALED cannot be used together in one declaration
Sealed means that class cannot be derived. Therefore no one class is able to override an
abstract method in a sealed class.
You are declaring a method as final, when it is not declared to be virtual.
You are trying to override a virtual method of a parent class that does not exist.
It is not possible to associate multiple messages with a single method.
Only ”MoveNext” and ”Current” enumerator identifiers are supported.
”MoveNext” or ”Current” identifier must follow the enumerator modifier.
”MoveNext” enumerator pattern method must be a function with Boolean return type and
no required arguments
Class or Object can have only one enumerator MoveNext declaration.
Class or Object can have only one enumerator Current declaration.
For in loop can be used not for all types. For example it cannot be used for the enumerations
with jumps.
Objective-C messages require their Objective-C name (selector name) to be specified using
the message ‘someName:’ procedure directive. While bindings to other languages
automatically generate such names based on the identifier you use (by replacing all
underscores with colons), this is unsafe since nothing prevents an Objective-C method name
to contain actual colons.
The Objective-C language does not have any constructors or destructors. While there are
some messages with a similar purpose (such as init and dealloc), these cannot be identified
using automatic parsers and do not guarantee anything like Pascal constructors/destructors
(e.g., you have to take care of only calling “designated” inherited “constructors”). For these
reasons, we have opted to follow the standard Objective-C patterns for instance
creation/destruction.
Due to compiler implementation reasons, message names are currently limited to 255
characters.
Due to compiler implementation reasons, mangled message names (i.e., the symbol names
used in the assembler code) are currently limited to 255 characters.
If no parent class is specified for an Object Pascal class, then it automatically derives from
TObject. Objective-C classes however do not automatically derive from NSObject, because
one can have multiple root classes in Objective-C. For example, in the Cocoa framework both
NSObject and NSProxy are root classes. Therefore, you have to explicitly define a parent
class (such as NSObject) if you want to derive your Objective-C class from it.
In Object Pascal, “published” determines whether or not RTTI is generated. Since the
Objective-C runtime always needs RTTI for everything, this specified does not make sense for
Objective-C classes.
This error indicates the use of Objective-C language features without an Objective-C mode
switch active. Enable one via the -M command line switch, or the $modeswitch x directive.
It is not possible to reintroduce methods in Objective-C or Java like in Object Pascal.
Methods with the same name always map to the same virtual method entry. In order to make
this clear in the source code, the compiler always requires the override directive to be
specified when implementing overriding Objective-C or Java methods in Pascal. If the
implementation is external, this rule is relaxed because Objective-C and Java do not have
any override-style keyword (since it’s the default and only behaviour in these
languages), which makes it hard for automated header conversion tools to include it
everywhere. The type in which the inherited method is defined is explicitly mentioned,
because this may either be an objcclass or an objccategory in case of Objective-C.
An overriding Objective-C method cannot have a different message name than an inherited
method. The reason is that these message names uniquely define the message to the
Objective-C runtime, which means that giving them a different message name breaks the
“override” semantics.
Duplicating an Objective-C or Java type using type x = type y; is not yet supported. You
may be able to obtain the desired effect using type x = objcclass(y) end; resp. type x =
class(y) end; instead.
It is not possible to declare a variable as an instance of an Objective-C category
or an Object Pascal class helper. A category/class helper adds methods to the
scope of an existing class, but does not define a type by itself. An exception of this
rule is when inheriting an Object Pascal class helper from another class helper.
A category replaces an existing method in an Objective-C class, rather than that it
overrides it. Calling an inherited method from an category method will call that
method in the extended class’ parent, not in the extended class itself. The replaced
method in the original class is basically lost, and can no longer be called or referred
to. This behaviour corresponds somewhat more closely to reintroduce than to
override (although in case of reintroduce in Object Pascal, hidden methods are still
reachable via inherited). The type in which the inherited method is defined is
explicitly mentioned, because this may either be an objcclass or an objccategory.
Interface getters are called via a helper in the run time library, and hence have to use the
default calling convention for the target (register on i386 and x86_64, stdcall on other
architectures).
The data in a typed file cannot be of a reference counted type (such as ansistring or a
record containing a field that is reference counted).
You are trying to use an overloaded operator when it is not overloaded for this type.
You are trying to use an overloaded operator when it is not overloaded for this type.
When declaring a typed constant array, you provided to few elements to initialize the array
Only when a piece of code is compiled with ansistrings enabled ({$H+}), string constants
longer than 255 characters are allowed.
univ parameters are compatible with all values of the same size, but this cannot be checked
in case a parameter’s size is unknown at compile time.
You are trying to declare more than one class constructor but only one class constructor can
be declared.
You are trying to declare more than one class destructor but only one class destructor can be
declared.
You are declaring a class constructor with a parameter list. Class constructor methods cannot
have parameters.
You are declaring a class destructor with a parameter list. Class destructor methods cannot
have parameters.
Objective-Pascal constructs are not supported when {$modeswitch ObjectiveC1} is not
active.
It is not possible to use unicodechar and unicodestring constants in constant expressions that
have to be converted into an ansistring or shortstring at compile time, for example inside
typed constants. The reason is that the compiler cannot know what the actual ansi encoding
will be at run time.
Objective-C “fast enumeration” support was added in Objective-C 2.0, and hence the
appropriate modeswitch has to be activated to expose this feature. Note that Objective-C 2.0
programs require Mac OS X 10.5 or later.
Objective-C for-in loops (fast enumeration) require that the compiler can find a unit
called CocoaAll that contains definitions for the NSFastEnumerationProtocol and
NSFastEnumerationState types. If you get this error, most likely the compiler is finding and
loading an alternate CocoaAll unit.
A nested procedural variable consists of two components: the address of the procedure/function
to call (which is always known at compile time), and also a parent frame pointer (which is
never known at compile time) in case the procedural variable contains a reference to a nested
procedure/function. Therefore such typed constants can only be initialized with
global functions/procedures since these do not require a parent frame pointer.
At the moment, scanner supports recording of only one token buffer at the time (guarded by
internal error 200511173 in tscannerfile.startrecordtokens). Since generics are implemented by
recording tokens, it is not possible to have declaration of a generic (type or method) inside
another generic.
An Objective-C protocol or Java Interface must be fully defined before classes can conform to
it. This error occurs in the following situation (example for Objective-C, but the same goes
for Java interfaces):
Type MyProtocol = objcprotoocl;
ChildClass = Class(NSObject,MyProtocol)
...
end;
Destructor declarations are not allowed in records or helpers.
Class methods declarations are not allowed in records without static modifier. Records have
no inheritance and therefore non static class methods have no sense for them.
Constructor declarations with no arguments are not allowed in records or record/type
helpers.
It is required that either the result of the routine or at least one of its parameters be of the
specified type. For example class operators either take an instance of the structured type in
which they are defined, or they return one.
Type parameters may be specialized with types which (e.g. ansistring) need
initialization/finalization code which is implicitly generated by the compiler.
A section directive is not valid for variables being declared as external.
A variable placed in a custom section is always statically allocated so it must be either a
static or global variable.
Some directives and specifiers like ”virtual”, ”dynamic”, ”override” are not allowed inside
helper types in mode ObjFPC (they are ignored in mode Delphi), because they
have no meaning within helpers. Also ”abstract” isn’t allowed in either mode.
As records don’t support inheritance the use of ”inherited” is prohibited for these as well as
for record helpers (in mode ”Delphi” only).
Records with types must be defined globally. Types cannot be defined inside records which
are defined in a procedure or function or in anonymous records.
A class may delegate an interface using the ”implements” clause only to a single property.
Delegating it multiple times is a error.
Method resolution clause maps a method of an interface to a method of the current class.
Therefore the current class has to implement the interface directly. Delegation is not possible.
Method resolution is only possible for interfaces that are implemented directly, not by
delegation.
When declaring a string with a given codepage, the range of valid codepages values is limited
to 0 to 65535.
A final (class) field must be assigned a single value in the (class) constructor, and cannot be
overwritten afterwards. A final (typed) constant is read-only.
Support for final fields in non-external classes requires a full data flow analysis
implementation in FPC, which it currently still lacks.
Java interfaces define a namespace in which formal constant can be defined, but since they
define no storage it is not possible to define typed constants in them (those are more or less
the same as initialised class fields).
Java does not automatically add inherited constructors to child classes, so that they can be
hidden. For compatibility with external Java code, FPC does the same. If you require access
to the same constructors in a child class, define them in the child class and call the inherited
one from there.
The compiler sometimes internally constructs Pascal code that is subsequently injected into
the program. These messages display such code, in order to help with debugging errors in
them.
Certain language features are not supported on targets that are managed virtual machines.
The JVM does not natively support virtual constructor. Unforunately, we are
not aware of a way to emulate them in a way that makes it possible to support
calling virtual constructors for the current instance inside another constructor.
The JVM does not allow lowering the visibility of an overriding method.
nostackframe call modifier is supposed to be used in conjunction with assembler.
nostackframe call modifier used without assembler modifier might still generate local stack
needs.
Automatically generated getters/setters cannot have the same name as existing identifiers,
because this may change the behaviour of existing code.
Automatically generated property getters/setters on the JVM platform are virtual methods,
because the JVM does not support non-virtual methods that can be changed
in child classes. This means that if a child class changes an inherited property
definition, the behaviour of that property can change compared to native targets since
even if a variable is declared as the parent type, by calling the virtual method the
getter from the child will be used. This is different from the behaviour on native
targets or when not activating automatically generated setters/getters, because in
that case only the declared type of a variable influences the property behaviour.
If a property’s specified getter/setter already corresponded to the naming convention specified by
the automatic getter/setter generation setting except in terms of upper/lowercase, the
compiler will print a warning because it cannot necessarily change that other declaration
itself not can it add one using the correct case (it could conflict with the original
declaration). Manually correct the case of the getter/setter to conform to the desired coding
rules. TChild overrides
Records with constants must be defined globally. Constants cannot be defined inside
records which are defined in a procedure or function or in anonymous records.
Records with methods must be defined globally. Methods cannot be defined inside
records which are defined in a procedure or function or in anonymous records.
Records with properties must be defined globally. Properties cannot be defined inside
records which are defined in a procedure or function or in anonymous records.
Records with class members must be defined globally. Class members cannot be defined
inside records which are defined in a procedure or function or in anonymous records.
The visibility sections (protected) and (strict protected) are only useful together with
inheritance. Since records do not support that they are forbidden.
This directive is not allowed in the given context. E.g. ”static” is not allowed for instance
methods or class operators.
The use of assembler blocks/routines is not allowed inside generics.
Properties cannot be declared local, only global, using the static directive or inside structured
types.
Some platforms, such as the JVM platform, encode the parameters in the routine name in a
prescribed way, and this encoding may map different Pascal types to the same
encoded (a.k.a. “mangled”) name. This error can only be solved by removing or
changing the conflicting definitions’ parameter declarations or routine names.
A default parameter value allows you to not specify a value for this parameter when calling
the routine, and the compiler will instead pass the specified default (constant) value. As a
result, default values can only be specified for parameters that can accept constant values.
The specified pointer type modifier is ignored, because it is not supported on the current
platform. This happens, for example, when a far pointer is declared on a non-x86 platform.
A generic declared in the interface section of a unit must not reference symbols that belong
solely to the implementation section of that unit.
This tells you that the recursive reading of the uses clauses triggered already a compilation of
the current unit, so the current compilation can be aborted.
Methods introduced in a generic must be implemented for the generic. It is not possible to
implement them only for specializations.
Generic methods are not allowed in interfaces, because there is no way to specialize a suitable
implementation.
Generic methods can not be declared as virtual as there’d need to be a VMT
entry for each specialization. This is however not possible with a static VMT.
Support for dynamic packages is not implemented for the specified target OS or it is at least
not tested and thus disabled.
The HardFloat directive can only be used if an instruction set is used which supports
floating point operations.
The index specified for the compilerproc directive is not an index that’s recognized by the
compiler.
An operator overload is defined for the specified overload, but the internal overload by
the compiler takes precedence. This only happens for operators that had been
overloadable before (e.g. dynamic array + dynamic array), but aren’t anymore
due to an internal operator being defined while this behavior is controllable by a
modeswitch (in case of dynamic arrays that is the modeswitch ArrayOperators).
A threadvar section inside a class or record was started without it being prefixed by class.
This error occurs in a situation like the following:
Type
TObj = object
procedure test;
end;
begin
TObj.test;
test is not a static method and hence cannot be called through a type, but only using an instance.
If a function is declared normally in the interface or as a forward declaration, and then later
as external, the compiler must generate a stub that calls the external function. Due to code
generation limitations, this cannot be done on some platforms. Even on platforms where it is
supported, this is quite inefficient.
A custom attribute is defined, but there is no identifier to bind it to.
1 to 2
1-1
The size of enumeration values is limited to 4 bytes. As the value can be signed, the range of
valid values is limited to a signed 32 bit value (i.e. longint).
1 to 2
1-1
The size of enumeration values is limited to 4 bytes. As the value can be signed, the range of
valid values is limited to a signed 32 bit value (i.e. longint).
This error occurs if one tries to define a method for a type that is originally declared in a
different unit.
At the current location specifying a constraint is not allowed. For example in delphi
mode, a constraint might not be specified in the header of the implementation.
AmigaOS/MorphOS syscall specific: for int64/qword parameter only a single register location
is specified
AmigaOS/MorphOS syscall specific: for a parameter which is smaller than 64 bits, a register
pair is specified
AmigaOS/MorphOS syscall specific: for 64 bit register pairs, only data registers are
supported
MorphOS syscall specific: only consecutive (f.e.: d1-d2) registers are supported for 64 bit
register pairs
noreturn tells the compiler that the activation scope of the subroutine is never left. This
includes exceptions goto or any other mean. While the compiler cannot detect all such cases
some are trivial and the compiler gives an error.
Only some targets (e.g. Embedded and FreeRTOS) support the section directive.
On a certain target, not all syntax variants of the syscall directive make sense and thus those
making no sense are not supported Declarations like var i: Integer absolute i; are not
allowed
WebAssembly reference types don’t have an in-memory representation and can only be
passed by value.
Promising exports are WebAssembly-specific. They are not allowed on other platforms.
Suspending externals are WebAssembly-specific. They are not allowed on other platforms.
The base type of a set can only have 255 elements. Sets of wide characters are reduced to sets
of 1-byte characters.
As a safeguard, the system unit may only use basic string types, not the string alias which is
dependent on the mode in which a unit is compiled.
By default, c style assignment operators (+=, -=, *=, /=) are turn off. Either turn them on
by the command line parameter -Sc or in the source code by $COPERATORS ON
Certain operations like the size of an object cannot be determined at compile time, so it
cannot be e.g. part of a constant expression.