docs/compiler_faq.pod - Parrot FAQ for compiler writers
Whoa, there--you're looking at the wrong FAQ. This document is for people writing compilers that target Parrot.
To answer your question, though, Parrot should theoretically work with any C89-compliant C compiler, although some features require gcc. See the README files in the root directory for more information about building Parrot.
gcc
compile
Define a sub that takes as input a string, and returns something invokable. The easiest way to create something invokable at the moment is to use another compiler object (for example, the PIR compiler), which you can get with:
PIR
.local pmc pir_compiler pir_compiler = compreg 'PIR'
Use #line 42 "file.pir" for this. Note that this command must be in the first column; there must not be any leading space, otherwise the command is treated as a comment (due to the # character).
#line 42 "file.pir"
#
IMCC makes this easy, handling the calling conventions for you.
$P0( $P1, $P2, $P3 )
where $P0 is the function object, and $P1, $P2, and $P3 are its parameters. You can also use a function's label in place of the object:
somefunctionlabel( $P1, $P2, $P3 )
You can also get return value(s):
($P1,$P2) = $P0( $P1, $P2, $P3 )
If the function name might collide with a Parrot opcode, quote it:
i = 'new'(42)
Similar to function calls, just append . and the method name to the object. You should quote a literal method name to avoid confusion.
.
ret_val = some_obj.'some_meth'(arg)
The method name may also be a string variable representing a method name:
.local string m m = 'bold' curses_obj.m()
There are several ways to achieve this, depending on the location of the subroutine.
If the sub is in the same compilation unit use a Sub constant:
.const .Sub foo = 'foo'
The subroutine object is available in PASM too with a similar syntax:
.const .Sub P2 = 'foo' # any P register will do ... .pcc_sub foo: # denote a Sub constant
If the PIR compiler finds a 'foo' function during compiling a file, then the syntax:
foo()
gets translated to above constant declaration.
A more dynamic way is:
.local pmc foo foo = find_name 'foo'
This searches for a subroutine 'foo' in the current lexical pad, in the current namespace, in the global, and in the builtin namespace in that order. This opcode is generated, if foo() is used, but the compiler can't figure out, where the function is.
If the subroutine is in a different namespace, use the find_global opcode:
find_global
foo = find_global 'Foo', 'foo'
This searches the sub foo in the Foo namespace.
foo
Foo
Closure and Coroutine carry both a dynamic state. Therefore you need to perform two steps. First use one of the above ways to locate the Sub object. Then use the op newclosure to capture the environment.
newclosure
.local pmc coro coro = find_name 'my_coro' coro = newclosure coro
Any subroutine that contains a .yield directive is automatically created as a Coroutine PMC:
.yield
.sub my_coro # automagically a Coroutine PMC ... .yield (result) ... .end
.sub foo # do something .return bar(42) # tail call sub bar .end .sub bar # ... .end
The sub bar will return to the caller of foo.
bar
If you have a variable amounts of arguments in an array, you can pass all items of that array with the :flat directive.
:flat
ar = new 'ResizablePMCArray' push ar, "arg 1\n" push ar, "arg 2\n" ... foo(ar :flat) ...
Use a slurpy array:
.sub mysub .param pmc argv :slurpy .local int argc argc = argv ...
If you have a few fixed parameters too, you can use a slurpy array to get the rest of the arguments
.sub mysub .param pmc arg0 .param pmc arg1 .param pmc varargs :slurpy .local int num_varargs num_varargs = varargs ...
Use the :optional and :opt_flag pragmas:
:optional
:opt_flag
.sub foo .param pmc arg1 :optional .param int has_arg1 :opt_flag .param pmc arg2 :optional .param int has_arg2 :opt_flag if has_arg1 goto got_arg1 ...
Please refer to "pdds/pdd20_lexical_vars.pod" in docs for details: look for :outer.
:outer
There are two possible ways. Either use the special PIR syntax:
$P0 = global 'name_of_the_global'
or the find_global op:
find_global $P0, 'name_of_the_global'
You can retrieve the namespace hash and use the delete opcode.
delete
.sub main :main $P0 = new 'Integer' store_hll_global 'foo', $P0 store_hll_global ['Bar'], 'baz', $P0 # ... .local pmc ns, Bar_ns ns = get_hll_namespace delete ns['foo'] # delete from top level Bar_ns = ns['Bar'] # get Bar namespace delete Bar_ns['baz'] $P0 = get_hll_global ['Bar'], 'baz' print "never\n" .end
Please refer to "pdds/pdd20_lexical_vars.pod" in docs for details.
You can't. You can store a Null PMC as the value though, which will catch all further access to that variable and throw an exception.
Use find_name:
find_name
$P0 = find_name '$x' find_name $P0, 'foo' # same thing
This will find the name foo in the lexical, global, or builtin namespace, in that order, and store it in $P0.
$P0
find_lex $P0, 'foo'
or much better, if possible just use the variable defined along with the .lex definition of foo.
.lex
That is still the same:
This finds a foo variable at any outer depth starting from the top.
If your language looks up variables differently, you have to walk the 'caller' chain. See also t/dynpmc/dynlexpad.t.
Don't emit store_lex at all. Use find_lex only if the compiler doesn't know the variable. You can always just use the register that was defined in the .lex directive as an alias to that lexical, if you are in the same scope.
store_lex
find_lex
TODO
With the newclass op:
newclass
newclass $P0, 'Animal'
Each class knows which attributes its objects can have. You can add attributes to a class (not to individual objects) like so:
addattribute $P0, 'legs'
Methods are declared as functions in the class namespace with the :method keyword appended to the function declaration:
:method
.namespace [ 'Animal' ] .sub run :method print "slow and steady\n" .end
As with methods, but note the new keyword. The vtable name specified must be an existing vtable slot.
.namespace [ 'NearlyPi' ] .sub get_string :vtable .return ('three and a half') .end
Now, given an instance of NearlyPi in $P0
$S0 = $P0 say $S0 # prints 'three and a half'
You can access attributes by a short name:
$P0 = getattribute self, 'legs' assign $P0, 4 # set attribute's value
a fully qualified name:
$P0 = getattribute self, "Animal\0legs" assign $P0, 4 # set attribute's value
or by an index:
.local int offs offs = classoffset 'Animal' $I0 = offs + 0 # 1st attribute $P0 = getattribute self, $I0 $I0 = offs + 1 # 2nd attribute $P0 = getattribute self, $I0
Properties aren't inherited. If you have some additional data that don't fit into the class's hierarchy, you could use properties.
You first have to get the class PMC of the class you want to subclass. Either you use the PMC returned by the newclass op if you created the class, or use the get_class op:
get_class
get_class $P0, 'Animal'
Then you can use the subclass op to create a new class that is a subclass of this class:
subclass
subclass $P1, $P0, 'Dog'
This stores the newly created class PMC in $P1.
First, create a class without a parent class using newclass (or with only one subclass, see previous question). Then add the other parent classes to it. Please refer to the next question for an example.
If you have a class PMC (created with newclass or by subclass), you can add more parent classes to it with the addparent op:
addparent
get_class $P1, 'Dog' subclass $P2, $P1, 'SmallDog' get_class $P3, 'Pet' addparent $P2, $P3 # make "SmallDog" also a "Pet"
Just override the init vtable for that class.
newclass $P0, 'Dog' # create a class named Dog # ... .namespace ['Dog'] .sub init :vtable # ...
Or you can specify the constructor method by setting the BUILD property of the class PMC:
newclass $P0, 'Dog' # create a class named Dog new $P1, 'String' # create a string set $P1, 'initialise' # set it to the name of the constructor method setprop $P0, 'BUILD', $P1 # set the BUILD property
You can do so either with the class name:
new $P0, 'Dog'
or with the class object:
$P1 = get_class 'Dog' # find the 'Dog' class unless null $P1 goto have_dog_class printerr "Oops; can't find the 'Dog' class.\n" .return () have_dog_class: new $P0, $P1 # creates a Dog object and stores it in register $P0
The chief difference is that using a string constant will produce the specific error "Class 'Dog' not found" if that happens to be the case; the other code has to check explicitly.
During the new opcode the constructor is called.
new
You can pass only a single argument to a constructor. By convention, a hash PMC is passed to the constructor that contains the arguments as key/value pairs:
new $P0, 'Hash' set $P0['greeting'], 'hello' set $P0['size'], 1.23 find_type $I0, 'Alien' new $P1, $I0, $P0 # create an Alien object and pass # the hash to the constructor
Create an Exception object and throw it!
$P0 = new 'Exception' throw $P0
Not too hard, is it?
$P0 = new 'Exception' $P0['_message'] = 'something happened' throw $P0
Use push_eh to push an exception handler onto the stack.
push_eh
push_eh handler $P0 = new 'Exception' # or any other code ... throw $P0 # ... that might throw pop_eh exit 0
An exception handler is called with two arguments: the exception and the message. More information is available inside the exception 'object' (TBD).
handler: .get_results ($P0, $S0) print 'Exception caught:' print $S0 print "\n" exit 1
exit
Rethrow the exception if it has a severity of EXCEPT_EXIT.
EXCEPT_EXIT
handler: .include 'except_severity.pasm' .get_results ($P0, $S0) $I0 = $P0[2] if $I0 == .EXCEPT_EXIT goto unexceptional print "Exception caught!\n" exit 1 unexceptional: rethrow $P0
push_eh handler $P0 = new 'Exception' $P0['_message'] = 'something happened' throw $P0 pop_eh exit 0 handler: .local pmc exception .local string message .get_results (exception, message) print 'Exception: ' print message print "\n" exit 1
Parrot supports dynamic PMCs, loadable at runtime, to allow compiler writers to extend Parrot with additional types. For more information about writing PMCs, see "build/pmc2c.pl" in tools and "pmc.pod" in docs.
To build dynamic PMCs, add something like the following to your Makefile:
PERL = /usr/bin/perl PMCBUILD = $(PERL) /path/to/parrot/tools/build/dynpmc.pl DESTDIR = /path/to/parrot/runtime/parrot/dynext LOAD_EXT = .so PMCDIR = pmc PMCS = MyInteger MyFloat MyString MyObject PMC_FILES = MyInteger.pmc MyFloat.pmc MyString.pmc MyObject.pmc dynpmcs : $(PMC_FILES) @cd $(PMCDIR) && $(PMCBUILD) generate $(PMCS) @cd $(PMCDIR) && $(PMCBUILD) compile $(PMCS) @cd $(PMCDIR) && $(PMCBUILD) linklibs $(PMCS) @cd $(PMCDIR) && $(PMCBUILD) copy "--destination=$(DESTDIR)" $(PMCS)
Parrot supports dynamic op libraries. These allow for ops specific to one language to be used without having to place them into the Parrot core itself. For examples of dynamic op libraries, see "dynoplibs" in src.
To build dynamic op libraries, add something like the following to your makefile:
PERL = /usr/bin/perl OPSBUILD = $(PERL) /path/to/parrot/tools/build/dynops.pl DESTDIR = /path/to/parrot/runtime/parrot/dynext LOAD_EXT = .so OPSDIR = ops OPLIBS = myops OPS_FILES = myops.ops dynops : $(OPS_FILES) @cd $(OPSDIR) && $(BUILD) generate $(OPLIBS) @cd $(OPSDIR) && $(BUILD) compile $(OPLIBS) @cd $(OPSDIR) && $(BUILD) linklibs $(OPLIBS) @cd $(OPSDIR) && $(BUILD) copy "--destination=$(DESTDIR)" $(OPLIBS)
Using the NCI you can invoke functions written in C from a Parrot script. To every NCI invocation, there are two parts: the native function to be invoked, and the PIR code to do the invocation.
First the native function, to be written in C. On Windows, it is necessary to do a DLL export specification of the NCI function:
/* foo.c */ /* specify the function prototype */ #ifdef __WIN32 __declspec(dllexport) void foo(void); #else void foo(void); #endif void foo(void) { printf("Hello Parrot!\n"); }
Then, after having compiled the file as a shared library, the PIR code looks like this:
.sub main :main .local pmc lib, func # load the shared library lib = loadlib "hello" # no extension, .so or .dll is assumed # get a reference to the function from the library just # loaded, called "foo", and signature "void" (and no arguments) func = dlfunc lib, "foo", "v" # invoke func() .end
If you embedded a Parrot in your C file and you want to invoke another function in that same C file, you should pass a null string to loadlib. Do that as follows:
.local pmc lib .local string libname null libname lib = loadlib libname
Under Linux, the .c file must then be linked with the -export-dynamic option.
Create a new Env PMC and access it like a hash.
Env
.local pmc e e = new 'Env' $P0 = e['USER'] # lt
.include 'iglobals.pasm' .local pmc interp, cfg interp = getinterp cfg = interp[.IGLOBALS_CONFIG_HASH] $S0 = cfg['VERSION'] # "0.3.0"
See config_lib.pasm for all the keys in the config hash - or iterate over the config hash.
Aside from the files in your new language directory, you must modify
CREDITS MANIFEST config/gen/languages.pm config/gen/makefiles/languages.in languages/LANGUAGES.STATUS.pod
Inside your language dir, you may consider adding the following:
LICENSE MAINTAINER README STATUS
Look to existing languages for some examples.
To install Parrot::Op, copy and paste the appropriate command in to your terminal.
cpanm
cpanm Parrot::Op
CPAN shell
perl -MCPAN -e shell install Parrot::Op
For more information on module installation, please visit the detailed CPAN module installation guide.