I'm writing a Matlab class that implements custom concatenation in order to present a clean, expressive interface to the user. Unlike the built-in's, the result of the overloaded horzcat() --- for example --- should be a scalar object set according to class' semantics. However, I'm not sure anymore about which rules apply to overloaded functions...
A class to illustrate the issue would be:
classdef A < handle
properties
k;
end;
methods
function obj = A(varargin)
if nargin == 0
obj.k = 0;
else
obj.k = varargin{1};
end;
end;
function obj = horzcat(obj1, varargin)
fprintf('1st arg is %s.\n', class(obj1));
if nargin > 1
fprintf('2nd arg is %s.\n', class(varargin{1}));
end;
obj = obj1; % Not important
end;
end;
end
Now, when testing this class with a small script:
% Please don't run this is you have valuable
% data in your base workspace.
clear classes;
clc;
a = A('dummy');
b = [-1, a]; % Should fail
the output is:
1st arg is double.
2nd arg is A.
This is surprising, because first argument is a double, i.e. the built-in horzcat() should have been called, subsequently throwing an exception about type mismatch or something. To my understanding, the type of the first argument decides which of the overloaded functions will be called (OK, we restrict ourselves to non-static methods). The code was tested on R2011b/Linux and R2012a/Windows, with the same results.
So, this is my question: what are the actual rules of calling overloaded methods? The Matlab documentation does not explain the above result, and I'd like to avoid developing code based on my illusions about how the language works.
Thank you in advance for your input.
I think MATLAB's documentation is pretty clear about it:
From here:
MATLAB uses the dominant argument to determine which version of a
function to call. If the dominant argument is an object, then MATLAB
calls the method defined by the object's class, if there is one.
And from here more about dominant argument:
The dominant argument in a method's argument list determines which
version of the method or function that the MATLAB runtime calls.
Dominance is determined by the relative precedences of the classes of
the arguments. In general, user-defined classes take precedence over
built-in MATLAB classes. Therefore, the left most argument determines
which method to call. However, user-defined classes can specify the
relative dominance of specific classes.
Related
I'm new to Lua and trying to understand the concept of OOP in Lua. To do so, I've tried creating an object and creating methods and "private variables". My issue is when I try to use "setters" or "getters", it's indicating that my tables are returning NIL which means I'm either having a scoping issue or something else I can't figure out.
The kicker is I'm using an example from an online Lua coding tutorial, and when I run the tutorial it works flawlessly. However, when I run mine, I get NIL or nothing outputs whenever I try to "get" or return a value from one of the member functions.
I'm using a couple of different environments:
ZeroBrain
Sublime Text
Lua for Windows
Do you know why my code is not returning populated tables?
newPlayer = function(n, h, a, r)
player = {}
n = n or ""
h = h or 100
a = a or 100
r = r or 0
function player:getPlayerName()
return n
end
function player:getPlayerHealth()
return h
end
function player:getPlayerArmor()
return a
end
function player:getPlayerRank()
return r
end
function player:setPlayerName(arg)
n = arg
end
function player:setPlayerHealth(arg)
h = arg
end
function player:setPlayerArmor(arg)
a = arg
end
function player:setPlayerRank(arg)
r = arg
end
function player:connect(arg)
print(string.format(" %s joined" , arg))
end
return player
end
player1 = newPlayer("John", 100, 100, 1000)
player1.getPlayerName()
Your code does not contain "populated tables" to return.
Your newPlayer function does create a table, and it does return it. It creates a number of functions within that table. But that's all newPlayer does: creates a table and puts some functions in it.
The data accessed by those functions is not part of the table. n, h, a, and r (BTW, please use better variable names) are all local variables. Your inner functions will access the specific stack containing those variables, but the variables themselves will not be magically associated with the table.
Your principle problem is almost certainly with the setters. And it comes from a combination of this:
function player:setPlayerName(arg)
with this:
player1.getPlayerName()
When you create a function using a : character between a table name and the function's name, you are using syntactic sugar for a function which implicitly takes as its first argument a value called self. As the name suggests, this is supposed to represent the object which this function is being called upon. So your function creation code is equivalent to:
function player.setPlayerName(self, arg)
Since you create all of your functions with :, all of your functions take at least one parameter.
The : syntax can also be used when calling such functions. If you did player1:getPlayerName(), this would cause the table you accessed to find the getPlayerName function to be used as the first argument in the function call. So that line would be equivalent to player1.getPlayerName(player1).
Obviously, these two syntaxes are mirrors of one another: functions created with : take a parameter that is expected to refer to the table it is being called on, and functions called with : will be given the table which was accessed to get that function.
But... your code didn't stick to the symmetry. You created the functions with :, but you call them with .
Now, you get functions are able to get away with this because... well, none of your values are actually part of the table. So your get functions just return the local value that they adopted from their creating context.
The set functions pose a problem. See, they take a parameter. But because the function was declared with :, they really take two parameters, the first being the implicit self.
Now, : syntax is just syntactic sugar; it's just a convenient way to do what you could have done yourself. So it is in theory OK to call a function with . even if you created it with :. But if you do so, you must pass the table as the first parameter. Though your code doesn't show it, I strongly suspect you didn't do that.
If you called player1.setPlayerName("foo"), what will happen is that the implicit self parameter will get the value "foo", and the arg parameter will be nil. And you will assign that nil value to the n local variable. So subsequent calls to player1.getPlayerName() will return nil.
Basically, what's going on here is that you're combining two different ways of creating objects in Lua. You stored your private data in a way that external code cannot access (ie: local upvalues), but that data is now no longer part of the table itself. Which means that, although you dutifully create those functions with : syntax to indicate that they take a self table, they never actually use that table. And because they never use the table, it's a lot harder to figure out what's going wrong.
Basically, the key here is to be symmetrical. If you create a function with :, then you should either call it with : or make sure to pass it the object table as the first parameter.
Broadly speaking, the standard way to create private members is by convention, not by forbidding it. That is, you agree not to mess with any members of a table other than those with certain names. Python convention is to pretend that names starting with _ don't exist, and Lua programs sometimes use that.
Upvalues are an interesting solution for private variables, but they do come with problems. If you want to invent a member variable, you have to do it in a centralized place rather than wherever you might need one. Even if the variable is optional, you have to create a named local at the top of the function.
TLDR of Nicol's answer, see my answer to another question:
function player:setPlayerArmor(arg)
a = arg
end
The : syntax is syntactic sugar. It creates an implicit 'self' argument when declared, and when used. If you declare it one way and use it another, the arguments won't be what you're expecting. Say your player has 100 health. Look at this result:
player1.setPlayerHealth(55, 66)
print(player1.getPlayerHealth())
-- will display '66', not '55' because `:` declares implicit 'self' argument
This displays 66 because the setPlayerHealth function has an implicit 'self' parameter because it was declared with :. If you instead called it
with the ::
player1:setPlayerHealth(55, 66)
print(player1:getPlayerHealth())
-- will display '55' because `:` passes player1 as self
function player:setHealth1(arg)
-- implicit 'self' argument refers to player1 when called on player1
end
-- is the same as
function player.setHealth2(self, arg)
-- with `.` notation, you need to add the 'self' argument explicitly
end
player1.setHealth1(31) -- self argument will be 31 and arg will be nil
player1.setHealth2(32) -- self argument will be 32 and arg will be nil
player1:setHealth1(33) -- self argument will be player1 and arg will be 33
player1:setHealth2(34) -- self argument will be player1 and arg will be 34
I know that everything is an object and you send messages to objects in Smalltalk to do almost everything.
Now how can we implement an object (memory representation and basic operations) to represent a primitive data type? For example how + for integers is implemented?
I looked at the source code for Smalltalk and found this in Smallint.st. Can someone explain this piece of code?
+ arg [
"Sum the receiver and arg and answer another Number"
<category: 'built ins'>
<primitive: VMpr_SmallInteger_plus>
^self generality == arg generality
ifFalse: [self retrySumCoercing: arg]
ifTrue: [(LargeInteger fromInteger: self) + (LargeInteger fromInteger: arg)]
]
Here is the link of above code: https://github.com/gnu-smalltalk/smalltalk/blob/62dab58e5231909c7286f1e61e26c9f503b2b3df/kernel/SmallInt.st
Conceptually speaking primitive methods are pieces of behavior (routines) implemented by the Virtual Machine (VM), not by regular Smalltalk code.
When the Smalltalk compiler finds the statement <primitive: ...> it interprets this as an special type of method whose argument (in your case VMpr_SmallInteger_plus) indicates the integer index of the target routine within the VM.
In this sense a primitive is a global routine not bound to the MethodDictionary of any particular class. The primitive logic is intended for a receiver and arguments of certain classes and that's why it must check that the receiver and the arguments (if any) conform its requirements. If not, the primitive fails and in that case the control flows to the Smalltalk code that follows the <primitive: ...> statement. Otherwise the primitive succeeds and the Smalltalk code below is not executed. Note also that the compiler will not allow for any Smalltalk code other than temporary declaration occurring above the <primitive:...> sentence.
In your example, if the argument arg is not of the expected class (presumably a SmallInteger) the routine gives up trying to sum it to the receiver and delegates the resolution of the operation to the Smalltalk code.
If the argument happens to be a SmallInteger, the primitive will compute the result (using the routine held in the VM) and answer with it.
I haven't seen the code of this primitive but it could also happen that the primitive fails if the result of the sum does not fit in a SmallInteger, in which case both the receiver and the argument would be cast to LargeIntegers and the addition would take place in the #+ method of the appropriate class (LargePositiveInteger or LargeNegativeInteger).
The other branch of the Smalltalk code allows for the implementation of a polymorphic sum between a SmallInteger and any other type of object. For instance this part of the Smalltalk code would take place if you evaluate 3 + 4.0 because in this case the argument is a Float. Something similar happens if you evaluate 3 + (4 / 3), etc.
I keep reading (for example in the Closure Compiler and other compilers) that inlining of functions isn't always safe. Could you please provide an example when inlining of functions shouldn't be done?
In many languages, inlining a function will have no observable semantic effects, although it is likely to affect the compiled size and execution time of the program. However, that is not true in languages in which the call stack and/or local variable bindings are visible.
As a simple example, in Javscript the local variable arguments always refers to an array-like object containing the arguments to the current function call. Clearly, if the function in which it occurred were inlined, its semantics would change. An inliner would have to either refuse to inline a function whose body references arguments or it would have to modify the code in a way which preserved the semantics, possibly by creating another local variable with a different name and substituting the reference.
Another example would be the (non-recommended) use of eval. Name lookup in the string passed to eval is done within the scope of the function which calls eval. For example:
inner = function(s) { var x = 4; return eval(s); }
outer = function(s) { var x = 3; return inner(s); }
outer("x+1")
Here the value returned by outer is 5. If inner were inlined, which would require renaming its local variable x to avoid name conflict, the value returned would be 4. (If both inner and outer were inlined, the value would probably be something else again.)
In general, it's going to be very difficult to inline a function which calls eval because there is no easy way to know the contents of the argument to eval.
This question already has answers here:
Difference between . and : in Lua
(3 answers)
Closed 8 years ago.
I'm still playing around with lua modules and I've found the following "interesting" issue that occurs depending on how you create your methods / functions inside a module.
Note the following code in a file called test_suite.lua:
local mtests = {} -- public interface
function mtests:create_widget(arg1)
print(arg1)
-- does something
assert(condition)
print("TEST PASSED")
end
return mtests
Using the above code, arg1 is always nil, no matter what I pass in when calling create_widget(). However, if I change the definition of the function to look like this:
function mtests.create_widget(arg1) -- notice the period instead of colon
print(arg1)
-- does something
assert(condition)
print("TEST PASSED")
end
then, the system displays arg1 properly.
This is how I call the method:
execute_test.lua
local x = require "test_suite"
x.create_widget(widgetname)
Can you tell me what the difference is? I've been reading: http://lua-users.org/wiki/ModuleDefinition
But I haven't come across anything that explains this to me.
Thanks.
All a colon does in a function declaration is add an implicit self argument. It's just a bit of syntactic sugar.
So if you're calling this with (assuming you assign the mtests table to foo), foo.create_widget(bar), then bar is actually assigned to self, and arg1 is left unassigned, and hence nil.
foo = {}
function foo:bar(arg)
print(self)
print(arg)
end
Calling it as foo.bar("Hello") prints this:
Hello
nil
However, calling it as foo:bar("Hello") or foo.bar(foo, "Hello") gives you this:
table: 0xDEADBEEF (some hex identifier)
Hello
It's basically the difference between static and member methods in a language like Java, C#, C++, etc.
Using : is more or less like using a this or self reference, and your object (table) does not have a arg1 defined on it (as something like a member). On the other way, using . is just like defining a function or method that is part of the table (maybe a static view if you wish) and then it uses the arg1 that was defined on it.
. defines a static method / member, a static lib, Which means you can't create a new object of it. static methods / libs are just for having some customized functions like printing or download files from the web, clearing memory and...
: Is used for object members, members that are not static. These members change something in an object, for example clearing a specified textbox, deleting an object and...
Metamethod functions(Functions that have :) can be made in lua tables or C/C++ Bindings. a metamethod function is equal to something like this on a non-static object:
function meta:Print()
self:Remove()
end
function meta.Print(self)
self:Remove()
end
Also, with . you can get a number/value that doesn't require any call from a non-static or static object. For example:
-- C:
int a = 0;
-- Lua:
print(ent.a)
-- C:
int a()
{
return 0;
}
-- Lua:
print(ent:a())
same function on a static member would be:
print(entlib.a())
Basically, each non-static object that has a function that can be called will be converted to : for better use.
In a MATLAB class I am writing, if the constructor is given 0 arguments, the user is asked to provide a file using uigetfile. If the user cancels the prompt, uigetfile returns 0. In that case it makes no sense to make the object. Is there some way of cancelling the object construction without throwing an exception? If I do an early return I get a malformed object that cannot be used. Here is what the code looks like:
classdef MyClass
methods
function self = MyClass(filename)
if nargin == 0
filename = uigetfile;
if filename == 0
%cancel construction here
return; %I still get a MyClass object with self.filename == []
end
end
self.filename = filename;
end
end
properties
filename;
end
end
However I am unsure if using uigetfile in the constructor is the right thing to do. Maybe it should be the resposibility of another part of my code.
In modern Matlab objects, I don't think it's possible to get out of the constructor without either returning a constructed object or throwing an error. (In the old style classes, the constructor was actually allowed to return whatever it wanted, including objects or primitives of other types, and oh man could that turn in to a mess.) When a constructor is called, the output argument is already intialized with an object with the default property values, so when you call return there, it just skips the rest of initialization and returns the object. If you try to replace out with something besides a MyClass object, that's an error.
Just reorganize the control flow to pull the GUI code out of the constructor, like you're speculating on at the end. Mixing it in to the constructor, especially conditionally, can cause problems. In particular, Matlab expects the zero-arg constructor to always return a scalar object with some sort of default values, because the zero-arg gets called implicitly when filling in elements during array expansion and so on. It's basically used as a prototype.