o made all (non-abstract) tdef and tsym constructors virtual
o added c*def/c*sym classref types for every (non-abstract) t*def/t*sym
class
o added cpusym unit for every architecture that derives a tcpu*def/tcpu*sym
class from the base classes, and initialises the c*def/c*sym classes with
them. This is done so that the llvm target will be able to derive from
the tcpu*def/sym classes without umpteen ifdefs, and it also means that
the WPO can devirtualise everything because the c* variables are only
initialised with one class type
o replaced all t*def/t*sym constructor calls with c*def/c*sym constructor
calls
git-svn-id: trunk@27361 -
the high level code generator. All hlcg code, except for the jvm code
generator, updated to use the new method. I didn't update jvm (it still uses
the method from cgutils), because I'm not familiar with that platform, so I
might get the types wrong.
git-svn-id: trunk@27309 -
required for the LLVM support (LLVM parameter support is not yet
included)
* always return the function return loc from a_call*, again as required
for the LLVM support
git-svn-id: branches/hlcgllvm@26992 -
defcmp.pas:
+ add new function "def_is_related" which combines the "is_related" overloads of "tobjectdef", "trecorddef" and "tstringdef" (it returns "false" for other def types which is what "tdef.is_related" did)
* compare_defs_ext & compatible_childmethod_resultdef: change call from "x.is_related" to "def_is_related(x,...)"
symtype.pas, tdef:
- remove "is_related" method
symdef.pas:
- remove "is_related" in "tobjectdef", "trecorddef" and "tstringdef"
* tobjectdef.needs_inittable: for checking whether a Corba interface somehow inherits from a IInterface don't use "is_related" anymore (we want to avoid the dependency after all), but mimic the necessary functionality of "def_is_related"
htypechk.pas, nadd.pas, ncal.pas, ncnv.pas, ngtcon.pas, nld.pas, optvirt.pas, pdecobj.pas, pdecvar.pas, pexpr.pas, pgenutil.pas:
* change call from "x.is_related" to "def_is_related(x,...)"
symtable.pas
+ use unit defcmp
* change call from "x.is_related" to "def_is_related(x,...)"
jvm/njvmcnv.pas, jvm/njvmflw.pas:
* change call from "x.is_related" to "def_is_related(x,...)"
git-svn-id: trunk@25847 -
cgobj.pas, tcg:
* g_save_registers: add the amount of used address registers to size as well
* g_save_registers: save all used address registers
* g_restore_registers: restore all stored address registers
m68k/cpubase.pas:
* rename saved_standard_address_registers to saved_address_registers
all other platform's cpubase.{inc,pas} (except alpha, ia64 and vis which are not up to date):
* add a saved_address_registers variable with one entry of RS_INVALID
At least a "make fullcycle" did complete.
git-svn-id: trunk@25664 -
(we previously internally generated assigned(unicode/ansistring) nodes,
which now cause typecheck errors because there is no special assigned
handling anymore)
git-svn-id: trunk@25557 -
that may trigger JVM bytecode verification errors if they are used
before they are initialised (this includes passing them as a "var"
parameter) + test
* sorted -CT parameters alphabetically and alligned them
git-svn-id: trunk@25387 -
a property in a child class has a higher visibility than the getter/
setter), ensure that we call the inherited method and not the method
itself (causing a stack overflow due to infinite recursion)
git-svn-id: trunk@25223 -
have to wrap an existing (g|s)etter that was marked as "abstract" (don't
mark the helper also as abstract, sicnce it contains code)
git-svn-id: trunk@23545 -
its property in case the visibility of the property is higher than
that of the getter/setter (on the JVM target when automatically
generating getters/setters)
git-svn-id: trunk@23523 -
that the type of the parameters can be determined automatically
o added compilerproc declarations for all helpers called in the compiler
via their assembler name, so we can look up the corresponding procdef
git-svn-id: trunk@23325 -
JVM target, since Java bytecode itself has no support for properties and
hence exposing properties to external Java code can only be done through
getters/setters. Use the new parameters to do so:
-CTautogetterprefix=XXX
-CTautosetterprefix=YYY
The getter/setter will get the same visibility as the property. If a
getter/setter with the same naming convention was already specified for a
property and this getter/setter is declared in the same class as the
property, then the visibility of this existing getter/setter is
modified and no new routine is generated.
Newly generated getters/setters are virtual methods, because that is
the only way in Java bytecode to allow redefining these getters/setters
in child classes. However, that also means that using these switches can
change the behvaviour of code, since normally the used property definition
is only determined by the declared type of its associated class instance,
and not by the actual instance type. The compiler will therefore warn when
such an automatically generated getter/setter is overridden by another
automatically generated getter/setter in a child class.
git-svn-id: trunk@22959 -
like field reordering (possible problems cracker classes) or using ebp as normal register (broken
stack traces from dump_stack)
+ niln is also valid in a cse domain
* parameters passed by reference shall have a complexity >1
* load nodes from outer scopes shall have a complexity >1
* better cse debugging
+ more node types added to cse
* consider parameters passed by reference in cse
* take care of cse in parameters in simple cases
git-svn-id: trunk@22050 -
used across multiple levels of inheritance such as on MIPS (and the code
has to be reusable for a_call_ref/reeg in the future anyway)
git-svn-id: trunk@21906 -
defines INTERNAL_BACKTRACE (which assumes get_frame to be internal rather
than a system unit procedure)
* changed fpc_assert() declaration for jvm to be the same as for other
targets, so the same code can be used to call it in ninl now that's
called from the first pass
git-svn-id: trunk@21903 -
result location (NR_FUNCTION_RESULT_REG is not valid on all platforms)
o this requires passing the forced function result type (if any) to this
method
o a generic, basic thlcg.a_call_name() is now available that sets the
function result location; can be called by descendants
* the availability under all circumstances of the correct function return
type enables g_call_system_proc() on the JVM platform to now determine
by itself how many stack slots are removed by the call -> do so, instead
of manually counting them (or forgetting to do so and messing up the
maximum evaluation stack height calculations)
git-svn-id: trunk@21862 -
with the enum instance corresponding to ordinal 0 in JVM constructors,
because a virtual method called by a parent constructor may already
have assigned a different value (see tests/test/jvm/tenum2.pp). This
will result in null pointer exceptions when using such fields without
first explicitly assigning a value to them though.
The old behaviour can be restored with the new -CTenumfieldinit command
line parameter
git-svn-id: trunk@21736 -
getintparaloc + adapted all call sites of getintparaloc. This
led to a number of additional, related changes:
o corrected the type information for some getintparaloc parameters
o don't allocate some intparalocs in cases they aren't used
o changed "const tvardata" parameter into "constref tvardata" for
fpc_variant_copy_overwrite to make pass-by-reference semantics
explicit
o moved a number of routines that now have to call find_system_type()
from cgobj to hlcgobj so that cgobj doesn't have to start depending
on the symtable unit
o added versions of the cpureg alloc/dealloc methods to hlcgobj that
call through to their cgobj counter parts, so we can call save/restore
the cpu registers before/after calling system helpers from hlcgobj
(not implemented in hlcgobj itself, because all basic register
allocator functionality is still part of cgobj/cgcpu)
git-svn-id: trunk@21696 -
RS_INVALID superregister (instead of sometimes RS_NO and sometimes
RS_INVALID)
* check for RS_INVALID in tcg.g_save_registers() and ignore such entries
git-svn-id: trunk@21622 -
methods that don't have to be implemented by targets using the high
level code generator, with dummy implementations that raise an internal
error (to remove warnings about instantiating classes with abstract
methods)
* let the JVM tcg descendent derive from this new thlbasecgcpu class
git-svn-id: trunk@21098 -
ansistrings are represented by nil
* fixed type conversion of constant empty ansistring/unicodestring to
pchar/pwidechar on the JVM target
git-svn-id: branches/jvmbackend@21055 -
o support for the new codepage-aware ansistrings in the jvm branch
o empty ansistrings are now always represented by a nil pointer rather than
by an empty string, because an empty string also has a code page which
can confuse code (although this will make ansistrings harder to use
in Java code)
o more string helpers code shared between the general and jvm rtl
o support for indexbyte/word in the jvm rtl (warning: first parameter
is an open array rather than an untyped parameter there, so
indexchar(pcharvar^,10,0) will be equivalent to
indexchar[pcharvar^],10,0) there, which is different from what is
intended; changing it to an untyped parameter wouldn't help though)
o default() support is not yet complete
o calling fpcres is currently broken due to limitations in
sysutils.executeprocess() regarding handling unix quoting and
the compiler using the same command lines for scripts and directly
calling external programs
o compiling the Java compiler currently requires adding ALLOW_WARNINGS=1
to the make command line
git-svn-id: branches/jvmbackend@20887 -
class before its VMT has been built, because that's the process that looks
for inherited abstract methods that are overridden -> instead use a new
synthetic method kind (tsk_callthrough_nonabstract) that will call
through to another method (in this case a constructor) if the owning
class does not contain any abstract methods, and generates an
empty routine otherwise
git-svn-id: branches/jvmbackend@20589 -
constructors (simply do nothing), because creating an abstract class
is not possible in the JVM (and while the JVM only triggers an
exception when you actually execute code that tries to create an
abstract class, the Android platform checks this using the bytecode
verifier at class verification time and therefore stumbled over the
old code)
git-svn-id: branches/jvmbackend@20547 -
on Dalvik, because those values are obtained via an "and 255" operation
which again turns them into longint values as var as the Dalvik type
system is concerned
git-svn-id: branches/jvmbackend@20527 -
to correctly handle non-pasbool types (can't just "not" their value),
and for Dalvik (a "not pasbool" became "xori 1", which turned the value
from a boolean into an integer as far as the Dalvik type checker was
concerned)
git-svn-id: branches/jvmbackend@20525 -
after an 8/16 bit operation on Dalvik (even if the value cannot have
become too big), because otherwise its type checker complains
git-svn-id: branches/jvmbackend@20520 -
their upper bound is lower than 127/32767: not necessary, and in case
of booleans not even allowed for Dalvik (because the "and" operation
turns the boolean into an integer as far as Dalvik is concerned)
git-svn-id: branches/jvmbackend@19825 -
classes, interfaces, dynamic arrays, ...) as "array of JLObject",
but instead as an array of the actual pointed-to type. The reason
for the previous behaviour was mainly that the JVM threadvar
implementation internally stores all threadvars that are
such double pointer types into an "array of JLObject". Type casting
this array to e.g. an "array of array of byte" causes an exception
even if it only contains "array of byte" instances (since the outer
array type is different). This is now solved by first taking the
element of the array and then typecasting it to the destination
type (so dynarrbyte(arr[0]) instead of pdynarrbyte(arr)^[0]).
The reason for the new (more accurate) behaviour is because the
old one caused type errors in case a double pointer field was
internally created for the nestedfpstruct support (see added
test)
git-svn-id: branches/jvmbackend@19821 -
verifier when -Cpjvmdalvik is used (including debug information). Using
-Cpjvmdalvik changes the semantics at the language-level in one case:
boolean(bytevar) will no longer return a boolean that contains the same
value as bytevar did, but will map the value to 0/1 (that also means that
such expressions cannot be passed to var-parameters in case of
-Cpjvmdalvik). Code compiled with -Cpjvmdalvik will also work fine on
the regular JVM, but it may be somewhat slower (it won't necessarily
be slower on Dalvik, because the .class -> .dex transformation
applies many optimizations itself)
git-svn-id: branches/jvmbackend@19743 -
those are also supported by the Java language (intftype.class) and required
in equivalent Pascal expressions (such as JLClass(intftype))
git-svn-id: branches/jvmbackend@19742 -
implicit pointer types into the nestedfpstruct, so that the original
parameter is properly changed when updated from inside nested routines
git-svn-id: branches/jvmbackend@19645 -
+ -CTcompactintarrayinit command line option to change the code generation
for typed array of some kind of integer-type initialization so that it
takes up less space in the bytecode (because the bytecode for all
routines, including the class initialization code that contains the
typed constant init code, is limited 64kb, large array or multiple
array constants could fairly easily bump into that limit)
git-svn-id: branches/jvmbackend@19638 -
set the external name of the underlying class type to the internal name
instead of to an (invalid) empty string
git-svn-id: branches/jvmbackend@19555 -
are used to implement procvar types, and add a constructor to the procvar
types that accept an instance implementing this interface -> much easier
and more natural to use procvar types from Java code
git-svn-id: branches/jvmbackend@19216 -
var-parameters on the JVM target for changes to the value passed
as var-parameter during the function call (because they are handled
via copy-in/copy-out, this may indicate unexpected bahviour later on).
out-parameters are checked in the same way, except if the out-parameter
is a local variable because then reading it before the call may result
in a bytecode verification error (since the variable may not yet be
initialized)
git-svn-id: branches/jvmbackend@19153 -
on the JVM target, because the generated code was invalid and I can't think
of a generic way to solve it (see added comments in njvmcal.pas)
git-svn-id: branches/jvmbackend@19055 -
o don't try to create .class files for generic types
o still generate all JVM-specific wrappers for generic types even though they
won't be written out, because when specializing all the defid's have to
match exactly
o add synthetic routine implementations after generating the specializations,
so that the synthetic routines for those specializations are also generated
(we don't specialize generic versions of the synthetic generic routines
because it's not easy or even always possible to create valid generic
versions of synthetic routines)
o Note: these are Pascal-style generics, not Java-style generics. The generic
types nor their specializations are usable from Java code (specializations
may become usable in the future)
git-svn-id: branches/jvmbackend@19047 -
to and returning them from sub routines, in order to follow the JVM
specs to the letter (not checked by the JVM bytecode verifiers, but
checked by the Android DEX verifier)
* -> also zero-extend them again at the caller side after returning
from such a function
git-svn-id: branches/jvmbackend@18919 -
required based on the source/destination sizes (like in
a_load_reg_reg(), a_load_ref_reg() and a_load_ref_ref())
git-svn-id: branches/jvmbackend@18918 -
complex parameters passed to inlined routines on the JVM target, because
it is not possible to take the address of any kind of node on the JVM
target (temp-reference nodes work for any kind of LOC_(C)REFERENCE, but
are currently only implemented for the JVM platform)
git-svn-id: branches/jvmbackend@18905 -
resultdef for internally generated method calls
* force the resulttype of methods used to build non-unicode string
constants on the JVM platform to the stringconstn's resultdef
git-svn-id: branches/jvmbackend@18832 -
as far as Java is concerned, they're now all arrays of JLObject.
When loading a value from them, we typecast the loaded value
to the appropriate type. This allows typecasting one pointer
type to another without getting verification errors (since an
"array of JLObject" is not compatible with "array of JLString")
- no longer allow dereferencing untyped pointers on the JVM
target, since that always results in invalid bytecode
git-svn-id: branches/jvmbackend@18819 -
unicodestring = java.lang.String. The reason this was the default in
the past is that this was the first string type that was implemented,
and without it being the default most code involving string operations
would fail. Now the default strings types are the same as for other
targets
+ new {$modeswitch unicodestrings} directive, that when activated
*together* with {$h+},
1) changes char into an alias for widechar
2) changes string into an alias for unicodestring
3) changes the preferred string evaluation type (in case of uncertainty)
to unicodestring
{$modeswitch unicodestrings} with {$h-} does not change anything at all
regarding the string type (it still changes the char type)
+ new uuchar unit that redefines char as widechar, and which is automatically
included by the compiler if {$modeswitch unicodestrings} is enabled
git-svn-id: branches/jvmbackend@18781 -
a generalized version of the formal var/out/constref support (this
also fixes passing string[xyz] expressions to non-formal var/out/constref
parameters)
git-svn-id: branches/jvmbackend@18774 -
also handle non-formal parameters
* do not copy in the original value in handleformalcopyoutpara() if it's
an out parameter
git-svn-id: branches/jvmbackend@18772 -
variables, class/record fields and arrays with enumtype(0) on
creation, so that using them without explicitly initializing them
doesn't cause a null-pointer exception. If enumtype(0) is not a
valid enum, they are not initialized (and since they wouldn't have
a valid value on native targets either in that case, an exception
on use is acceptable)
git-svn-id: branches/jvmbackend@18755 -
* added runerror number to JVM FpcRunTimeError exceptions
* enabled calling errorproc when a run time error occurs on the
JVM target
git-svn-id: branches/jvmbackend@18749 -
+ include mathh.inc in the system unit, implement most routines based
on java.lang.Math, and use (an unmodified!) genmath for the rest
- removed now obsolete jmathh.inc (was partial copy of mathh.inc)
- removed commented out overrides for several math routines in
njvminl (doesn't make sense to handle them inline in the compiler)
git-svn-id: branches/jvmbackend@18742 -
with wrapper calls, because the parameter nodes are already bound to the
paravarsyms of the current procdef so simply replacing the procsym and
procdef is not enough
git-svn-id: branches/jvmbackend@18739 -
types involved to determine whether or not the comparison is
unsigned (they don't always match due to bugs in the compiler,
and the sign of the comparison operation is what's used in
other code generators too)
git-svn-id: branches/jvmbackend@18734 -
(make_field_static() ) and replaced semi-duplicates of that
code with calls to this routine
* made the handling of static fields for the JVM target more
similar to that on the other targets, so that class properties
now also work there (-> updated JVM-specific code in several
places to deal with this new handling)
git-svn-id: branches/jvmbackend@18723 -
create a virtual class (factory) method that calls the constructor,
and then let the existing support for virtual class methods handle
everything
git-svn-id: branches/jvmbackend@18710 -
support those natively, so they are emulated via the procvar infrastructure
(and hence not very fast). Could probably be optimized somewhat by adding
a cache (hashmap of procvars) to the class hierarchies, maybe in the
future.
git-svn-id: branches/jvmbackend@18704 -
on the caller side whether it's a normal or class method that we are
calling, add the self-class parameter inside the invoke method if
necessary (by simply counting the number of parameters and inserting
it if we're one short)
git-svn-id: branches/jvmbackend@18699 -
their initialization can cause races between the unit they are declared
in and the class constructor they are initialized in (even if both
would be moved to the unit initialization code, a class constructor
using the set constant could run before the unit initialization code has
run)
git-svn-id: branches/jvmbackend@18698 -
o every porocedural variable type is represented by a class with one
public "invoke" method whose signature matches the signature of the
procvar
o internally, dispatching happens via java.lang.reflect.Method.invoke().
WARNING: while this allows calling private/protected or other methods
that are normally not accessible from another context, a security
manger can override this. If such a security manager is installed,
most procvars will cause security exceptions
o such dispatching also requires that all arguments are wrapped, but
that's done in the compiler-generated body of the invoke method,
so that procvars can also be called conveniently from Java code
o typecasting between a procedure of object and tmethod is supported,
as well as Delphi-style replacing of only the method pointer via
@procvar1=@procvar2.
o nested procvars are not yet supported, but most of the basic
infrastructure for them is already present
* all units/programs now get an internal __FPC_JVM_Module_Class_Alias$
type when compiled for the JVM target, which is an "external" class
that maps to the unit name. This is required to look up the
JLRMethod instances for regular functions/procedures
+ new tabstractprocdef.copyas() method that allows to create a procvar
from a procdef and vice versa
git-svn-id: branches/jvmbackend@18690 -
the functions may be called from Java code or indirectly and in that case
there's no way to guarantee that they'll be zero-extended on the caller
side
git-svn-id: branches/jvmbackend@18684 -
o primitive types are first boxed
o the parameter is passed inside an array of one class instance
o changing the parameter inside the routine (by assigning a value to it
like in Delphi.NET and different from regular Pascal code) will replace
this class instance (again boxing the value if required)
o on return, the class instance is extracted, unboxed if required, and
assigned back to the original location
o formal const parameters are handled without the extra array indirection,
since they cannot be changed
TODO: while writing tjvmcallparanode.handleformalcopyoutpara() I forgot that
calling getcopy on ttemprefnodes whose ttempcreatenode hasn't been copied
yet works fine, so that code is more complex than needed. Still have to
fix.
git-svn-id: branches/jvmbackend@18675 -
copying that over the old one into calling a dedicated fpcInitializeRec()
method that initializes the required fields. The reason is that this
initialization is performed for out-parameters, and the fpcDeepCopy()
method (used to copy one instance over another) has an out-parameter
-> infinite loop
git-svn-id: branches/jvmbackend@18674 -
o sets of enums are handled as JUEnumSet instances, others as JUBitSet
derivatives (both smallsets and varsets, to make interoperability with
Java easier)
o special handling of set constants: these have to be constructed at run
time. In case of constants in the code, create an internal constsym to
represent them. These and regular constsyms are then aliased by an
another internal staticvarsym that is used to initialise them in the
unit initialisation code.
o until they are constructed at run time, set constants are encoded as
constant Java strings (with the characters containing the set bits)
o hlcgobj conversion of tcginnode.pass_generate_code() for the genjumps
part (that's the only part of the generic code that's used by the JVM
target)
o as far as explicit typecasting support is concerned, currently the
following ones are supported (both from/to setdefs): ordinal types,
enums, any other set types (whose size is the same on native targets)
o enum setdefs also emit signatures
o overloading routines for different ordinal set types, or for different
enum set types, is not supported on the JVM target
git-svn-id: branches/jvmbackend@18662 -
pointer was in a register (has to be put into memory, because otherwise
the compiler thinks that the entire type is stored in a register instead
of only its address)
git-svn-id: branches/jvmbackend@18659 -
routine is called we don't know yet whether it has to be forced or not
for the JVM target (staticvarsyms that need initialization may still be
created during pass1 of the main program/unit code)
git-svn-id: branches/jvmbackend@18646 -
* only wrap the init/finalization code after the main program has been
completely processed (pass1, codegen), because this may influence
the required wrapping in case of the JVM target
* replace periods with slashes in the package name when constructing
the classes corresponding to the units (to execute their initialization
code) -- didn't notice this was wrong previously because due to the
incomplete wrapping, they were never being initialized until now
git-svn-id: branches/jvmbackend@18645 -
(cannot reuse the memory location on the JVM)
* optimized int_to_int conversions for certain smaller->larger types
in mem locations (can reuse the memory location on the JVM, because
stack locations are always at least 4 bytes)
git-svn-id: branches/jvmbackend@18622 -
them to shortstringclass, make sure the address operator is
"typed" so that we don't insert useless checkcast instructions
git-svn-id: branches/jvmbackend@18621 -
JDK class-style enums rather than plain ordinals like in Pascal
o for Pascal code, nothing changes, except that for the JVM target
you can always typecast any enum into a class instance (to interface
with the JDK)
o to Java programs, FPC enums look exactly like Java enum types
git-svn-id: branches/jvmbackend@18620 -
so that the result type of random dynamic array expressions doesn't suddenly
get changed into open array (causes internalerror if the size is requested),
and instead simply change the conversion type to tc_equal
git-svn-id: branches/jvmbackend@18605 -
mangled name handling ansistring rather than pshortstring based (required
for JVM target; little effect on speed, some extra memory usage)
git-svn-id: branches/jvmbackend@18597 -
o support for JVM arrays in JVM addrnodes and derefnodes (so we
can take the address of var parameters to store them in the
parentfpstruct and later dereference them)
o process loadnode.left also in tjvmloadnode.pass_generate_code() when
handling var-parameters
git-svn-id: branches/jvmbackend@18593 -
o since the JVM target has no stack/framepointer that can be passed
on to nested routines, all local variables and parameters accessed
from nested routines are grouped into a local record whose address
is passed to nested routines. The same technique is also required
for LLVM in the future
git-svn-id: branches/jvmbackend@18588 -
(non-dynamic arrays, records, shortstrings)
- removed the ability to typecast such types directly into related class
types, you have to use the @-operator first now to get a pointer to
the type
o updated the RTL and internal compiler code to properly use this
new convention
o allowed removing several special cases from
tjvmtypeconvnode.target_specific_general_typeconv(), and that
method can probably be removed completely over time
* no longer give compile time errors for pointer-related typecasts that
will fail at run time, because the checking was too complex and could
be worked around via actual pointer typecasts anyway
* removed some unnecessary checkcast operations (for shortstring/
shortstringclass)
git-svn-id: branches/jvmbackend@18574 -
* move the incstack() from a load before the potential "and" to
zero-extend, so that the maximum stack height get calculated
properly
git-svn-id: branches/jvmbackend@18565 -
