Recurring C++ and Qt anti-patterns

kshegunov

@JonB said in Recurring C++ and Qt anti-patterns:

Depressingly, for those of us forced to use Python (with Qt) we are encouraged to use exceptions for flow handling etc. Under the mantra that Python exceptions are different from/more lightweight than C++ exceptions....

They're not, obviously. However in all fairness exceptions have their place, just not in the way they're abused. I use them for quick escape through the stack in numerical code, where it'd be borderline stupid to sprinkle std::optional with if-ery everywhere to handle an error deep in the code. As everything, though, one should apply the very old and tested "common sense" ...

artwaw

@Kent-Dorfman I still have C64 (my first computer bought in 1988) and I still use Basic for the fun of it :)

JonB

@kshegunov said in Recurring C++ and Qt anti-patterns:

They're not, obviously.

I take it that is against my "Python exceptions are different from/more lightweight than C++ exceptions"(?) Well, apparently, this is true. I could (if required) give several references. I guess we cannot assume that a Python exception maps to a C++/OS/native exception....

kshegunov

@JonB said in Recurring C++ and Qt anti-patterns:

@kshegunov said in Recurring C++ and Qt anti-patterns:

They're not, obviously.

I take it that is against my "Python exceptions are different from/more lightweight than C++ exceptions"(?)

Yes, that's correct. More so I was referring to the "lightweight"-ness, not so much about them being different. I could agree if you can show a VM's exception to be lighter than a hardware's stack unwinding, but that's rather dubious. Basically that'd be like saying that Python's exceptions are faster than a C longjmp, arguable at best.

Well, apparently, this is true. I could (if required) give several references.

Eh, fine, amaze me.

I guess we cannot assume that a Python exception maps to a C++/OS/native exception....

Nor had I done that. The underlying tech is different to jump to such a conclusion.

JonB

@kshegunov said in Recurring C++ and Qt anti-patterns:

Eh, fine, amaze me.

Well, of course now that you want them I can't find as many clear statements as I have come across in the past :) But we could start with this accepted answer on SO:

In the Python world, using exceptions for flow control is common and normal.

The Python cultural norm is somewhat different. In many cases, you must use exceptions for control-flow. Also, the use of exceptions in Python does not slow the surrounding code and calling code as it does in some compiled languages

In other words, your understanding that "exceptions are for the exceptional" is a rule that makes sense in some other languages, but not for Python.

Or https://stackoverflow.com/a/3743528/489865

If you are using the exception as part of the standard control flow - which is the Pythonic "ask forgiveness, not permission" way

For the "efficiency" question, although this may not be the post I had in mind I find in https://softwareengineering.stackexchange.com/a/351121

The general consensus “don't use exceptions!” mostly comes from other languages and even there is sometimes outdated.
In C++, throwing an exception is very costly due to “stack unwinding”.
So in those languages exceptions are “too expensive” to be used as control flow. In Python this is less of an issue and exceptions are a lot cheaper.

In Python, if I want to know if a key is in a dictionary the "Pythonic" way is

try:
    abc = dict["key"]
except:
    ...

utilizing dict exception throwing on non-existent key rather than testing for existence....

Want to test for a divisor being 0? Don't test it, divide and catch the exception.

Am I good Python citizen? Am I, heck no! If nothing else, it gets in the way of having the choice to break on any exception in a debugger.

J.Hilk

@JonB said in Recurring C++ and Qt anti-patterns:

Want to test for a divisor being 0? Don't test it, divide and catch the exception.

oh my goodness, that's like exploding the first atomic bomb without doing the math to check if it would ignite the atmosphere!
(yes, those calculations where done, no matter how silly the idea seems now🙈)

kshegunov

@JonB said in Recurring C++ and Qt anti-patterns:

In C++, throwing an exception is very costly due to “stack unwinding”.
So in those languages exceptions are “too expensive” to be used as control flow. In Python this is less of an issue and exceptions are a lot cheaper.

Returning from a function causes stack unwinding; exiting a block causes stack unwinding. So while I see why it may be costly to throw (for example a heavy destructor running), it's hardly costlier than to just return (the little coming from the compiler generating the appropriate exception handling code). So while the quoted argument may seem reasonable to you, even creating an object in Python is already costly, probably more so than the whole throw machinery of C++, so take "expensive" and "cheap" as real relative terms, just like in real life.

JonB

@kshegunov
I like this: your claim/view is that Python is so slow in the first place that it can't get much worse with exceptions :)

kshegunov

@JonB said in Recurring C++ and Qt anti-patterns:

I like this: your claim/view is that Python is so slow in the first place that it can't get much worse with exceptions :)

No, not exactly. I claim that you should compare things that are comparable to begin with. Python runs in a VM, while C++ runs on metal. It is context that defines the term, and "costly" in C++ terms isn't applicable, or justifiable in python. Think of it like this, fopen calls into the kernel, do you consider this costly? It can be, if you do it all the time, like opening a file reading a couple of bytes and closing it. But the point is, is this costly? "Well, it depends" is the correct answer. Or, as I mentioned already, is longjmp costly?

What I'm trying to convey is that in python you don't even consider this stuff, because the almighty VM shields you from it, and you can say your exceptions are cheap, which I don't know, they might very well be. But you still pay for the VM, exceptions or no exceptions. On the other hand I can choose to use exceptions where they make sense and pay the handling code price, or I may choose not to in some places and declare things with noexcept, or I can do what Qt does and disable them altogether.

But even when you're talking only C++ it's not absolute. As I stated, and I do stand by it, throwing isn't that much different than unwinding stack frame by stack frame, until you meet the handler. And in C/C++ the stack is king, it's god and spirit and the holy mother, and all that; and it's very fast, and it is hardware supported for a reason. I would hardly believe anyone claiming push and pop are coincidences, a technical curiosity if you will. So yeah, unwinding the stack "is as inevitable as the demise of capitalism", but it's done all the time for all reasons, some of which were mentioned, and it is by far a bad reasoning, rotten logic if I may, to say it makes throwing "costly".

PS.
That's one of my better missives, if I may be so bold to say so myself. ;P

jsulm

@JonB said in Recurring C++ and Qt anti-patterns:

try:
abc = dict["key"]
except:

Isn't pythonic way

if "key" in dict:
   ...

?!

JonB

@jsulm
No, the whole point is that the "Pythonic" way is precisely to go try ... except instead of checking via in, that's my point! Same with division by 0, and other cases. This is Python's "Ask for forgiveness, not for permission" philosophy. Note that personally I wrote

Am I good Python citizen? Am I, heck no! If nothing else, it gets in the way of having the choice to break on any exception in a debugger.

:)

kshegunov

@JonB said in Recurring C++ and Qt anti-patterns:

Same with division by 0, and other cases.

Out of curiosity: How does python handle that, because as far as I know (I don't really know that much, but bear with me for a second) there's no (strong) typing and while dividing by int(0) is invalid, division by double(0) is valid?

JonB

@kshegunov
I am just reporting that the "Pythonic" way to do division, where the divisor might be zero, is to do the divide unconditionally and catch the exception. As soon as I see "Pythonic" I tend to ignore it, and do what I'd do in C++ instead, but that's just me :)

What I can show you is the following output:

>>> z = 1 / 0
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ZeroDivisionError: division by zero
>>> z = 1.5 / 0.0
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ZeroDivisionError: float division by zero
>>>

So you get a ZeroDivisionError either way (which you could catch in a try ... except ZeroDivisionError), though by the look of it the error message distinguishes between plain division by zero versus float division by zero :)

kshegunov

@JonB said in Recurring C++ and Qt anti-patterns:

As soon as I see "Pythonic" I tend to ignore it, and do what I'd do in C++ instead, but that's just me

You're a wise man ... ;)

What I can show you is the following output

Thanks, curiosity satisfied. So python just raises an exception even if dividing by a double(0) is a valid operation. Fair enough.

JonB

@kshegunov
I don't want to get into a debate (I know what you're like :) ), and I do know about floating point numbers being approximate representations (though zero/0.0 does have an exact representation), but (IMHO!) it is only in your physics/quantum mechanics area that "dividing by a double(0) is a valid operation" (the area where you can magic-away infinities and so on!). In a program it is not. (What have I let myself in for...!)

kshegunov

@JonB said in Recurring C++ and Qt anti-patterns:

I know what you're like

Hey! Words can hurt, you know! ;)

though zero/0.0 does have an exact representation

Yes, actually two representations, as with the actual zero. You have +0.0 and -0.0.

it is only in your physics/quantum mechanics area that "dividing by a double(0) is a valid operation"

Eh, I didn't write the IEEE standard. Take your beef with prof. Kahan.

In a program it is not

Actually if you look through the math.h implementations you're going to see a lot of handling for such cases. For example the people who wrote them had the decency to actually handle these special cases like log(0) returning -inf. While I agree it's not often useful to divide by zero it sometimes can be, so that's the reason to handle it like that, I assume.

JonB

@kshegunov said in Recurring C++ and Qt anti-patterns:

You have +0.0 and -0.0

The next time someone asks how much money I have in my pocket I will remember to give this answer.

I have $123.45 to give away. I want to hand each person $0.00. How many people do I need to meet to get rid of all my cash? :)

kshegunov

Here: https://en.wikipedia.org/wiki/Riemann_series_theorem
Knock yourself out ... ;P

fcarney

Be careful with not doing things the pythonic way in python. A lot of the time doing it the pythonic way leverages the internals of the language. In other words it pushes the execution from the interpreter to the built in methods that are written in C. So it can have an effect on performance. I don't think the exception example does this though. There may be other reasons I am not aware of.

Kent-Dorfman

wait a cotton pickin minute! there is no explicit cast to double in python so the x/double(0) argument is invalid on that basis alone...and x/float(0) behaves as expected.