Basic inheritance question

[Bruno Desthuilliers]

Lie a écrit :
> On Jan 16, 9:23 pm, Bjoern Schliessmann <usenet-
> mail-0306.20.chr0n... at spamgourmet.com> wrote:
>> Lie wrote:
>>> 42.desthuilli... at wtf.websiteburo.oops.com> wrote:
>>>> I used to systematically use it - like I've always systematically
>>>> used 'this' in C++  and Java.
>>> And that is what reduces readability.
>> IMHO not, IOPHO not. This is the nth time (n >> 1) this discussion
>> comes up here. If I have learned one thing from those very lengthy
>> discussions, it's that Python's "self" handling is not going to
>> change.
> 
> And ah... yes, I don't wish it go away either. In VB, excessive Me
> reduces readability, but in Python it doesn't.
> 
>>> A proficient VB/C/Java programmer would frown upon the extra,
>>> unneeded garbage as they thought it was clear already that the
>>> variable refers to a class-level variable.
>> C programmers surely have no opinion concerning C because it has no
>> native classes.
> 
> C-family, ok?

Well... I'm not sure what "C-family" means, but to me it would consist 
in C, C++, Objective C, and D.

> Please stop taking my words to its letters.

So we're supposed to actually guess what you really mean ???

>> Personally, I've seen many C++ programs with complex class designs
>> where it definitely helps to consistently use "this->". I cannot
>> remember all local (and global) variables in bigger methods.
> 
> In that case, you have the _option_ to do it.

Make no sens when maintaining code wrote by someone that didn't use this 
'option'.

(snip)

>>>> it's the first argument of the function - which usually happens to be
>>>> the current instance when the function is used as a method.
>>> And that's the point, self (or anything you name it) is almost always
>>> the current instance
>> # this is a plain function. In this function,
>> # 'obj' can be whatever that happens to have a (numeric)
>> # 'stuff' attribute
>> def func(obj, arg):
>>    return (obj.stuff + arg) / 2.0
>>
>> # this is a class with an instance attribute 'stuff'
>> class Foo(object):
>>     def __init__(self, bar):
>>       self.stuff = bar + 42
>>
>> # this is another (mostly unrelated) class
>> # with a class attribute 'stuff'
>> class Bar(object):
>>    stuff = 42
>>
>> # this is a dummy container class:
>> class Dummy(object): pass
>>
>> # now let's play:
>> import new
>>
>> d = Dummy()
>> d.stuff = 84
>> print func(d, 1)
>>
>> d.baaz = new.instancemethod(func, d, type(d))
>> print d.baaz(2)
>>
>> f = Foo(33)
>> print func(f, 3)
>> Foo.baaz = func
>> f.baaz(4)
>>
>> print func(Bar, 5)
>> Bar.baaz = classmethod(func)
>> Bar.baaz(6)
>>
>>>  and that makes it functionally the same as Me and
>>> this in VB and Java.
>> Depends on the context, cf above !-)
> 
> Please again, stop taking letters to the words, I don't meant them to
> be exactly the same, rather the same would meant that they generally
> can be considered equal,

If you still think that way after having read the above code, then I 
can't help. We obviously don't share the same mental model here.

> exceptions exists of course. And btw, I don't
> understand what you meant by your example, they seemed to be a
> completely OK program for me,

Indeed it's ok (even if totally useless by itself). The point is that 
'self' (or whatever you name it) is just and only the first argument of 
a function, period.

> even though it's a bit confusing to
> follow[2].

Nothing in it should be confusing to anyone having a decent knowledge of 
Python's object model IMHO.

> [2] btw, the reason it's a bit confusing to follow is one of my
> points: It is a Bad Thing(tm) to use the same name for different
> variables

Where do I "use the same name for different variables" here ?

> even in a language like Python that enforce explicit naming
> of classes

Python doesn't "enforce" explicit name of classes - IIRC, there are ways 
to instanciate anonymous class objects. But I definitively don't see how 
this relate to this discussion.

Yes, I know, "please guess what I mean" !-) but sorry, there's no sens 
discussing a technical point without using accurate and appropriate 
technical naming of technical concepts invlved.

>>>>> Most other languages
>>>>> 1) automatically assign the containing class' object
>>>> s/containing class' object/current instance/
>>>>> in a keyword
>>>>> (Java: this, VB: Me) behind the screen,
>>>> That's not very far from what a Python method object does -
>>>> automatically assign the current instance to something. The difference
>>>> is that Python uses functions to implement methods (instead of having
>>>> two distinct contructs), so the only reliable way to "inject" the
>>>> reference to the current instance is to pass it as an argument to the
>>>> function (instead of making it pop from pure air).
>>> It isn't very far, but Python makes it obvious about the assignment
>>> (not behind the screen).
>> Exactly. And given both the simplicity of the solution and what it let
>> you do, that's a *very* GoodThing(tm) IMHO.
> 
> I agree, it's a Good Thing but it doesn't make the point less pointy,
> the difference between Me/this and self is just the explicit
> assignment. Other things that is possible because of the explicit
> assignment is just a "coincidence" of design choice.

Are you sure ? As far as I'm concerned, I think that the design choice 
somehow results from what it makes possible.

(snip pointless discussion wrt/ # of attributes)

>>> And it is always a Bad Thing(tm) to use the same name for two
>>> variable in the class and in function (which is the main and only
>>> source of possible ambiguity) in ANY language, even in Python.
>> Ho, yes.... Like, this would be bad ?
>>
>> class Person(object):
>>    def __init__(self, firstname, lastname, birthdate, gender):
>>      self.firstname = firstname
>>      self.lastname = lastname
>>      self.birthdate = birthdate
>>      self.gender = gender
>>
>> C'mon, be serious. It's often hard enough to come with sensible names,
>> why would one have to find synonyms too ? Try to come with something
>> more readable than the above, and let us know. Seriously, this braindead
>> rule about  "not using the same name for an attribute and a local var"
>> obviously comes from languages where the "this" ref is optional, and
>> FWIW it's obviously the wrong solution to a real problem (the good
>> solution being, of course, to use the fully qualified name for
>> attributes so there's no possible ambiguity).
> 
> The code fragment you've given way above (about the Foo, Bar, bazz,
> and func) also suffers from the bad habits of using the same name for
> different variables.

Where ? And how does this answer the question above ?


> And it's not a "braindead" rule

The way you express it, and as far as i'm concerned, it is, definitively.

> The example you've given IS the most readable form since the function
> is _simple_, consider a function that have complex codes, possibly
> calculations instead of simply assigning initial values I'm sure you'd
> slip up between the self.* variables and the * variables once or
> twice,

*you* would perhaps have this problem. And you would indeed have this 
problem in Java or C++. In Python, this problem just don't exist.

> possibly becoming the source of hard-to-find bugs.

Consistant and intelligible naming is quite another problem. And it's 
too much dependant on the context, language, domain and whatnot for any 
rule like your one above to be universally applyable.


> And in languages that doesn't enforce explicit naming of classes,

I still don't understand how all this relates to the naming of class 
objects ?

Oops, sorry: you meant "in languages that has implicit instance 
reference available in methods" ? Python doesn't have it, so any rule 
deriving from this "feature" is out of scope here.

>  when
> there is the two or more same names, the one with the smallest scope
> is picked, so in _simple_ functions, the trick of using full qualified
> names and overloaded local names is still possible and feasible. In
> complex functions, the trick fails even in Python, because even if
> Python and our full-concentration-brain is aware of the difference
> between self.* and *, our spreaded-concentration-brain that is
> scanning the code for the source of bugs might get stumbled on the
> confusing use of self.* and *.

Here again, *you* may have this problem. I don't, since I always used 
explicit instance reference.

>>>> Anyway, I actually know 3 languages (4 if C# works the same) that has
>>>> this implicit 'this' (or whatever the name) 'feature', and at least 5
>>>> that don't. So I'm not sure that the "most other languages" qualifier
>>>> really applies to point 2 !-)
>>> What's this 5 languages?
>> Smalltalk, Python, PHP, javascript, Ruby. I don't remember how Scheme,
>> CLOS and OCaml handle the case.
> 
> Among all them, only Javascript is considerably mainstream.

Is that a joke ? PHP is probably the most used language for web apps 
(server side of course). And Python is certainly not an obscure unknown 
geek-only language no more. But anyway: you were talking about "most 
other languages" - not "most mainstream languages".

>>> Are they a mainstream, high-level languages
>>> or lesser known, low-level languages? C-family, Java, and Basic are
>>> the Big Three of high-level programming language.
>> None of C, C++, Java nor Basic qualify as "hi-level". C is the lowest
>> possible level above assembly, C++ is often refered to as an "object
>> oriented assembler", Java is way too static, crippled, verbose an
>> unexpressive to qualify as "hi-level" (even if it suffers from some
>> problems usually associated with higher level languages). I won't even
>> comment on basic (is that really a language at all ?).
> 
> Your criteria on being high-level is simply just odd.

My criteria on being hi-level seems quite common: automatic memory 
management, portability, "rich" builtin data types, functions as first 
class citizens, lexical closures, strong introspection features, strong 
metaprogramming support, etc...

> The rest of the
> world recognizes C-family, Java, and Basic as high-level languages.

C was "hi-level" wrt/ assembler, indeed. And Java is "hi-level" wrt/ 
C++. Ho, and _please_ don't get me started on basic !-)

> If I have to say it, Python is actually lower level than Basic.

No comment.

>  While
> Java is just below Python and C and C++ is just below Java. Why do I
> consider Basic the highest-level? Because it is the cleanest to scan
> (no confusing symbols, i.e. no curly braces, no confusing use of
> parens (Python uses (), [], and {}, VB only use ()[3]),

Basic != VB. There are quite a few other basics here.

Now if you choose this criteria, you may want to learn some Lisp dialect.

> 
> In what way C++ resembles an assembler? 

C++ is some OO stuff bolted on a very close-to-the-metal language itself 
designed to write operating systems, drivers and other low-level stuff.

> Have you ever programmed in
> assembly?

I did.

> How hard is it to create a simple program in assembly? And
> how hard is it to create a complex program in C++

Roughly the same, thanks to scoping rules, dependencies hell, lack of 
automatic memory management and overcomplex features.

> (which AFAIK is used
> by hundreds of mega projects including CPython)?

CPython - as the name implies - is written in C.

> And have you ever used Basic at all?

I did. And not only VB.

> Some programmers would instantly frown upon Basic, simply because they
> don't know that Basic is "just another language".

I've used at least four different variants of Basic.

>>>>> In VB, Me is extremely rarely used,
>>>> I used to systematically use it - like I've always systematically used
>>>> 'this' in C++  and Java.
>>> And that is what reduces readability. A proficient VB/C/Java
>>> programmer
>> There are quite a few proficient C/C++/Java programmers here. As far as
>> I'm concerned, I would not pretend being one - I just have a good enough
>> knowledge of C, Java and (alas) VB to be able to get up to speed in a
>> reasonnable time frame.
>>
>> As a side note, the problem just doesn't exists in C, which has
>> absolutely no support for OO.
> 
> When I said C, it might mean C and C-family,

When you say "C", it means "C" to everyone reading you.

> so please stop
> misunderstanding me.

Please learn to express yourself clearly. If you say "X" when you mean 
"Y", *you* are the one responsible for misunderstandings. This is human 
communication 101 skill.

(snip)
> I'm not saying my mindset is better than yours (it have its positives
> and negatives), in fact I apologize for getting you confused.

My "mindset", as you say, is the one you'll find in each and every 
technical communication. You can't discuss technical points without a 
clear, unambiguous naming of technical concepts. And when a community 
exists, it usually has it's own technical idiom, whether highly specific 
(ie : "member variables" in C++, "attributes" in Python), or more 
general (ie: C is not the same thing as C++, whenever language you're 
using).

>>> would frown upon the extra, unneeded garbage as they
>>> thought it was clear already that the variable refers to a class-level
>>> variable.
>> In C++, the canonical way to make this "clear" is to use the m_name
>> convention. There must be some reason C++ programmers feel a need for
>> this "extra, unneeded garbage" ?-)
> 
> In some cases, an extremely complex class that can't be fragmented any
> further, the m_ convention is surely useful, but in most cases you
> could skip them out.

You "could" skip this convention, but it's really considered bad 
practice by quite a lot of C++ programmers.

> And the canonical way to make this "clear" is not
> the m_ convention, it's the name itself. A well-designed class would
> choose names that is recognizable instantly from the name itself, even
> without the pseudo-name appended to it (or prepended).

I await for your exemples.

> btw you must have been memorizing names braindeadly, because the only
> way you could stumble on that is by memorizing names braindeadly.
> Names shouldn't be memorized, it should be inferred and memorized. For
> example, when you met a variable name firstname and lastname inside a
> class called Person, you'd immediately realize that it is Class Level
> variable because you know that the function you're currently working
> on use the name initialfirstname and initiallastname.

Fine, I now have four names to handle, each of them being possibly an 
argument, a local variable, a member variable or a global variable. Great.

Sorry but I won't buy this.

(snip)

>>> As I've pointed out, there is little harm in class-level variable's
>>> implicit reference.
>> Have some working experience on any non-trivial C++ project ?
> 
> No

I would have been surprised if you had answer otherwise.

> (you could say I'm a student so I've never "worked"[1]). But I've
> done some medium-sized projects in other languages.
> 
> [1] If you understand the irony, you'd realized I was deliberately
> misunderstanding you

Not sure.

>>>>> Compare the following codes:
>>>>> VB.NET:
>>>>> Public Class A
>>>>>     Dim var
>>>>>     Public Function aFunction()
>>>>>         return var
>>>> Add three levels of inheritence and a couple globals and you'll find out
>>>> that readability count !-)
>>> It's the mental model that have to be adapted here, if the current
>>> class is inheriting from another class, you've got to think it as
>>> names from parent class as it is a native names, so you don't actually
>>> need to know where the variable comes from
>> In C++ (and VB IIRC), it might as well be a global So sorry but yes, I
>> have to know where it comes from.
> 
> How many times would you use globals, it is a Bad Thing(tm) to use
> globals in the first case.

It is, most of the time, indeed.

The problem is that you rarely start a project from scratch - most of 
the time, you have to work on legacy code. And you really seldom have 
the possibility to do a major rewrite to fix all warts.

> In some exceptional cases globals might be
> unavoidable, but it is trivial to work out that you have to reduce the
> amount of globals to a minimum, in almost any cases to a number you
> can use a hand to count with.

That very nice, from a theoretical POV. That's alas just not how it 
works in real life.

> And applying the hacks mentioned, why
> don't you use the m_ convention for globals, and retains the
> convenience of m_-free variables in your class variable. You use class
> variable much more often than globals, and in most projects class-
> level variable is used just as often as local-variable.

The problem is not what *I* (would) do, but how is the existing code.

>>> since knowing where it
>>> comes from is breaking the encapsulation
>> Nope, it's knowing what you're doing and how the piece of software at
>> hand is working. And FWIW, while data hiding is one possible mean of
>> encapsulation, it's by no way a synonym for encapsulation.
> 
> I agree that knowing an underlying class's implementation is useful
> (in fact, very useful) but what I'm talking is about should-ness, 

> we
> shouldn't _need_ to know the underlying implementation,

How can you hope to extend a class without _any_ knowledge of it's 
implementation ?

(snip)
> 
>>> (which, in Python is very
>>> weakly implemented, which favors flexibility in many cases[1]).
>>> [1] In Python, it is impossible to create a completely private
>>> variable, which is the reason why the mental model of these other
>>> languages doesn't fit Python.
>> Never heard about the infamous '#define private public' hack in C++ ?
>> And don't worry, there are also ways to get at so called 'private' vars
>> in Java.
> 
> No, but it's violating the language's rule.

Nope. It's just making use of some part of the language's rules.

> Python OTOH, provides
> formal ways to got to private vars.

Python doesn't have "private vars" at all.

>>>> In any non-trivial piece of C++ code, and unless the author either used
>>>> the explicit 'this' reference or the 'm_xxx' naming convention, you'll
>>>> have hard time figuring out where a given name comes from when browsing
>>>> a function's code.
>>> If you're used to the implicit naming scheme it's easy to know where a
>>> variable came from, if not the current scope, it's the class' scope
>> You forgot the global scope.
> 
> How many global variables do you have in your projects on average? If
> you always have a pageful list of globals in your projects, then you
> should consider unlearning and relearning the basic programming
> concepts.

I'll come to you when I'll need more training, don't worry.

> It's easy to keep track of globals, as you shouldn't have a
> lot of them even in a huge project.

Can't you understand that starting a new project afresh is *not* the 
common case ?

>>> As a final note:
>>> I don't think implicit class reference is superior to explicit class
>>> reference, neither
>> ...
> 
> I'm sure you don't believe it since I'm talking on implicit's side,
> but that's the fact, I just pointed you out that implicits do have its
> positive side (even if you don't consider them positive in _your_
> book) but that doesn't meant I believe it is better than the other.
> 
> To clear things up:
> As a final note:
> I don't think implicit class reference is superior to explicit class
> reference, but I don't think the vice versa is true either.

Once again : what classes have to do with this ?

Seriously, how can you hope to be taken seriously when you're obviously 
confusing such important concepts as instance and class ?