Coding: Fleeting Thoughts

[Dason]

Are there any major advantages of 3 over 2 that might persuade somebody to use 3 instead?

[Aaeriele]

Are there any major advantages of 3 over 2 that might persuade somebody to use 3 instead?

The biggest one is probably all string stuff being Unicode by default, which is good.

The "nonlocal" keyword is also occasionally handy.

Extended iterable unpacking (first, *variable, last = foo).

Set literals; dictionary comprehensions.

[RoadieRich]

...dictionary comprehensions.

2.7 has them.

[troyp]

...dictionary comprehensions.

2.7 has them.

Hey, cool! I didn't know that.
2.7 has set literals, as well, btw.

[EvanED]

Some of the stuff I like are the newfound inability to set True and False, and also the fact that things like strings and integers are not orderded between them. (Compare the result of 1 < "hi" in each. In Python 2, this returns true; in Python 3, it throws a TypeError, like a reasonable language.)

[Xeio]

Some of the stuff I like are the newfound inability to set True and False

But then you can't do nonsense like this!

Code: Select all

typeof(string).GetField("Empty").SetValue(null, " ");

[Aaeriele]

...dictionary comprehensions.

2.7 has them.

Hey, cool! I didn't know that.
2.7 has set literals, as well, btw.

Yeah, those are Python 3 features that were backported to the 2.x series (Note that Python 3.0 was actually released 2 years before Python 2.7...).

http://www.python.org/getit/releases/2.7/

Also note that Python 2.7 is the last major release of the 2.x line... so the biggest "feature" that 3.x has is that it's what will be receiving all new versions from here on out.

[MHD]

Has anyone looked at D?
It's like C++, but without all the bad things.

[Jplus]

It's been mentioned several times on these forums.
It does miss some of the bad things of C++, but I think in exchange it's also missing some of the good things. It also doesn't seem to add anything spectacular you couldn't have in C++ some way or another. So until somebody can point out such a thing to me, I'm not enthousiast.

(Off-topic: if that was bad English, anyone please send me a PM...)

[RoadieRich]

Is it just me, or are the streams on sockets in Java backwards?

[Xeio]

Well, I guess you get input from the stream... but yea, I much prefer the Read stream/Write stream idiom.

[phlip]

Is it just me, or are the streams on sockets in Java backwards?

Yeah, the "input stream" is the stream your program gets input from, not the one the socket gets input from.

[Pepve]

Is it backwards? My OS calls them stdin and stdout..

[MHD]

It's been mentioned several times on these forums.
It does miss some of the bad things of C++, but I think in exchange it's also missing some of the good things. It also doesn't seem to add anything spectacular you couldn't have in C++ some way or another. So until somebody can point out such a thing to me, I'm not enthousiast.

(Off-topic: if that was bad English, anyone please send me a PM...)

It has builtin garbage collection and an actual object system and it has compile time programming/type manipulation that puts C++ to shame.
It does still support pointer shenanigans and spectacularly unsafe casts, and even lets you tag code accordingly.
It has a sensible standard library and builtin things that C++ has clunky imports for.
It can link against C code.

It did right from the start (1999) everything the C++11/0x tries to fix now.

One noticable problem is that D 2.0 haven't got a port of Tango yet.
Other things is that it lacks ports for several major C libraries too, such as GMP and many GUI libraries.

[EvanED]

[continuation of this discussion.]

(As a side note, every time I read the manpage for gettimeofday I feel like the next manpage I read is going to just be the lyrics of Never Gonna Give You Up. There are two parameters, but you are required to pass NULL for the second? It returns a value which is guaranteed to be 0? Whose idea of "good API design" is this?)

Sounds like you need a better man page. Mine explains why you should pass NULL (it's obsolete functionality that has never been supported on LInux), and that it returns -1 on failure; looking at the error codes it can return, the implementation on your system may not have any failure modes.

http://pubs.opengroup.org/onlinepubs/00 ... ofday.html

Maybe the Evan was looking at this

That's actually probably exactly what happened. For a few reasons I actually have a tendency to Google for man pages instead of just running 'man blah', and that is apparently the first hit if you do a Google search for gettimeofday().

That being said, I stand by my statement. Why would you put such a poorly-designed function in the standard? I guess I could see including it for backwards compatibility, but it should have been added and immediately deprecated, and an alternative provided which is not stupid. (clock_gettime() was added later, and at least my manpage says "is not widely supported.")

[darkone238]

[continuation of [notaurl=http://forums.xkcd.com/viewtopic.php?p=2880081#p2880081]this discussion[/notaurl].]

(As a side note, every time I read the manpage for gettimeofday I feel like the next manpage I read is going to just be the lyrics of Never Gonna Give You Up. There are two parameters, but you are required to pass NULL for the second? It returns a value which is guaranteed to be 0? Whose idea of "good API design" is this?)

Sounds like you need a better man page. Mine explains why you should pass NULL (it's obsolete functionality that has never been supported on LInux), and that it returns -1 on failure; looking at the error codes it can return, the implementation on your system may not have any failure modes.

http://pubs.opengroup.org/onlinepubs/009604599/functions/gettimeofday.html

Maybe the Evan was looking at this

That's actually probably exactly what happened. For a few reasons I actually have a tendency to Google for man pages instead of just running 'man blah', and that is apparently the first hit if you do [notaurl]http://www.google.com/search?sourceid=chrome&ie=UTF-8&q=man+gettimeofday#pq=man+gettimeofday&hl=en&qe=IGdldHRpbWVvZmRheQ&qesig=YvrQphwe6SbSaLVrhNdgbg&pkc=AFgZ2tn7D4int2HK6y4e-WazF50g7EXbnLstZQLQIZXQFgUuCDJFW73s-2nk1-gUN-es4-ly8lMUoqxmbgVrAkYERrVaXzx4wQ&cp=0&gs_id=5&xhr=t&q=gettimeofday&pf=p&sclient=psy-ab&source=hp&pbx=1&oq=+gettimeofday&aq=0&aqi=g1g-c1g2&aql=&gs_sm=&gs_upl=&bav=on.2,or.r_gc.r_pw.,cf.osb&fp=f2e3bb8212f508f6&biw=1678&bih=942]a Google search[/notaurl] for gettimeofday().

That being said, I stand by my statement. Why would you put such a poorly-designed function in the standard? I guess I could see including it for backwards compatibility, but it should have been added and immediately deprecated, and an alternative provided which is not stupid. (clock_gettime() was added later, and at least my manpage says "is not widely supported.")

http://linux.die.net/man/2/gettimeofday This seems to give a little more context into the function, but I think I agree with you that it should've just been deprecated. Why make decades of code have pointless parameters and depend on poorly defined return values (especially if there's some implementations that actually just return 0 no matter what. Why return anything?)?

[Shivahn]

Is there a newbie thread? Or, like, a newbie question thread? I have a question and I feel sort of weird posting in the fleeting thoughts thread.

Well, not so much weird as I think that I'm likely to have several others, and the "IT DOESN'T WORK" thread isn't appropriate either, as they're often style-related questions since I've learned all I have from the internet rather than formal instruction. If there isn't a thread like that, maybe I'll make one later, unless people think it's a bad idea for whatever reason...

Anyway, I'm on and off working on a simple game engine in C++. There is a source file for chunks that loads them and yada yada. And there's a source file for the creature class and subclasses, which handle all the interactions and such. And some other ones too. They all use (repeatedly) the pixel dimensions of a block, which is the fundamental unit of the world under this particular engine. I'm not sure what the appropriate style and way to handle this is. Do I have each of the header files include a header with constant definitions? Is there some other preferred way of doing this? I'd prefer not to have each header have a definition, since missing one during a change could cause numerous bugs. They'd all be easy to fix, probably, but I'd still rather be safe than sorry...

So, yes. What is the recommendation for something like this?

[EvanED]

Do I have each of the header files include a header with constant definitions? Is there some other preferred way of doing this? I'd prefer not to have each header have a definition, since missing one during a change could cause numerous bugs. They'd all be easy to fix, probably, but I'd still rather be safe than sorry...

You have good sense. You definitely don't want to repeat definitions like that. One header with the constant values that gets included everywhere would be perfectly reasonable if you want to set your dimensions globally like that.

[Shivahn]

Thanks! I'll do that. I'm glad to hear my ideas aren't totally senseless

[PM 2Ring]

Do I have each of the header files include a header with constant definitions? Is there some other preferred way of doing this? I'd prefer not to have each header have a definition, since missing one during a change could cause numerous bugs. They'd all be easy to fix, probably, but I'd still rather be safe than sorry...

You have good sense. You definitely don't want to repeat definitions like that. One header with the constant values that gets included everywhere would be perfectly reasonable if you want to set your dimensions globally like that.

It's a Good Idea to put include guard macros in a header that gets included by other headers, to prevent the header from being included multiple times. FWIW, I tend to put guard macros in all my headers, just to be on the safe side.

[Shivahn]

Do I have each of the header files include a header with constant definitions? Is there some other preferred way of doing this? I'd prefer not to have each header have a definition, since missing one during a change could cause numerous bugs. They'd all be easy to fix, probably, but I'd still rather be safe than sorry...

You have good sense. You definitely don't want to repeat definitions like that. One header with the constant values that gets included everywhere would be perfectly reasonable if you want to set your dimensions globally like that.

It's a Good Idea to put include guard macros in a header that gets included by other headers, to prevent the header from being included multiple times. FWIW, I tend to put guard macros in all my headers, just to be on the safe side.

Code::Blocks does that for me thankfully.

Which is good because otherwise I'd have SDL included like fifteen times. Which would be less than excellent.

[You, sir, name?]

Do I have each of the header files include a header with constant definitions? Is there some other preferred way of doing this? I'd prefer not to have each header have a definition, since missing one during a change could cause numerous bugs. They'd all be easy to fix, probably, but I'd still rather be safe than sorry...

You have good sense. You definitely don't want to repeat definitions like that. One header with the constant values that gets included everywhere would be perfectly reasonable if you want to set your dimensions globally like that.

It's a Good Idea to put include guard macros in a header that gets included by other headers, to prevent the header from being included multiple times. FWIW, I tend to put guard macros in all my headers, just to be on the safe side.

Code::Blocks does that for me thankfully.

Which is good because otherwise I'd have SDL included like fifteen times. Which would be less than excellent.

I'm fairly certain SDL has include guards in it's headers, so that won't be a danger.

[Shivahn]

Good point. Of course they do >.>

I have another question now. In Scheme, you can create function definitions within definitions, as in this sample:

Spoiler:

Code: Select all

(define (quadruple x)
  (define (double x) (* x 2))
  (double
   (double x)))

Is there a way to do this in C++? I wrote some code a while ago for the game engine that, now that I have better theory and understanding of modularity and organization and such, is terrible. It's a public function of the creature class that uses the entity's location, velocity, width, and environmental location and checks for collisions and resolves them. I want to make a helper function that takes those and returns the adjustment that needs to be done to avoid a collision with the environment, and I'd like to make one separately for left, right, up and down. I think doing it that way would be easier than having one big function that has a mess of if trees and keeps track of everything and does it all at once. But it'd be only used for that function: finding and resolving environmental collisions. So if at all possible, I don't want to declare it in the class definition.

Is there a way to do that? And if so, is that just a terrible way to go about things, or all right?

[You, sir, name?]

Short answer: No.

Long answer: Yes, but only in C++11.

[Shivahn]

Ok, I'll just think of some other way to do it for now. It's not a huge amount of code anyway, it just offends my aesthetic sensibilities.

[EvanED]

You can still break it off. If you are implementing your functions inside the class definition, i.e.

Code: Select all

class MyClass {
  public:
    void doSomething() {
        blah();
        blah();
   }
};

then you can at least make the helper functions private.

If you are implementing your functions in a separate .cpp file, then you can make that functions just available in that file, and people who are using your class won't even know they exists. You can do this either by putting those functions inside the unnamed namespace:

Code: Select all

namespace
{
    void helperFunction1()
    {
        ...
    }
}

or by marking them static:

Code: Select all

static void helperFunction1()
{
    ....
}

(This is a different meaning for static from static class members or function variables.) Both of these will make it so that other compilation units can't link against those functions.

[Xanthir]

Fleeting Thoughts seems most appropriate for this.

I just got a new tattoo today - a unix timestamp of my wedding date. Here's the pattern, and the finished result.

[Shivahn]

Thanks, Evan. The static method looks like what I was going for, just in a slightly different place.

That's super awesome Xanthir! Personal in several ways, I like it

[Link]

What I thought would be a 30-minute coding job turned out to be a pretty large can of worms. Well, the naive approach was indeed a 30-minute coding job. Unfortunately, the output is the digital equivalent of the Thousand Young of Shub-Niggurath*. The non-naive approach is of the sort where a Google search on the subject turns up a frightening amount of pages with the text "buy this article".

*Technically speaking, the output is not incorrect as such, but it's one of the ugliest, most inefficient and sanity-shatteringly voluminous ways to represent the data in question.

[Xeio]

Having a private property, and then a private field that stores the value of that property: Good.

Setting the value of the property to the value of the field (which then resets the value of the field to... itself...): Not so good.

I guess at least it isn't broken. Though I'm sort of scared to try and dig deep enough to figure out if there's a reason they're trying to work around the side effects caused by the Set on the property...

Luckily this doesn't seem related to my current bug...

[Jplus]

[not enthusiastic about D]

It has builtin garbage collection

Even if you can switch it off and use unified deterministic resource management instead (and I suspect you can't), I don't consider that an advantage over C++. Although in that case it wouldn't be a drawback either.

and an actual object system

What do you mean by that? What's not real about C++' object system?

and it has compile time programming/type manipulation that puts C++ to shame.

I've heard a similar claim before, but I couldn't find any clear hints at the Digital Mars and D programming language websites on how D would be better at this than C++. Care to give an example?

It does still support pointer shenanigans and spectacularly unsafe casts, and even lets you tag code accordingly.

Does the tagging offer any advantage, apart from documentation (which you could also do with plain comments)? E.g. does it allow the compiler to do anything special?

It has a sensible standard library

Not a difference.

and builtin things that C++ has clunky imports for.

Well that's nice, but I suspect many more libraries are available for C++ overall. It's kind of sweet to provide a big standard library, but it's not of much use if you still have less options overall.

It can link against C code.

Not a difference.

It did right from the start (1999) everything the C++11/0x tries to fix now.

Oh really? Like what? I can find that D has some standard libraries which are now introduced as standard libraries in C++, such as threading and regex, but that again only obscures the fact that more libraries are available for C++ overall. The website doesn't seem to mention any core features that C++ didn't have until now (e.g. lambdas). Even if there are such things they bear little practical relevance because, well, C++ now has them too.
(Oh wait, I can find one thing. C++11 got class member initializers as a new feature, while apparently D already had them.)

One noticable problem is that D 2.0 haven't got a port of Tango yet.
Other things is that it lacks ports for several major C libraries too, such as GMP and many GUI libraries.

Let's just face that less libraries are available for D.

I can find one thing that D has and which C++ lacks completely, and that's the immutable modifier. A rather meager advantage IMO, given how C++ and D compare otherwise. Rust does a better attempt than D to be "better than C++" in that regard. On the other hand it lacks more of the good things as well (such as Turing-complete generics).

[Xeio]

See, now I would think that "bool ValidateX(X)" would return true if X was valid.

Of course, I must be crazy to follow convention like that.

[TheChewanater]

See, now I would think that "bool ValidateX(X)" would return true if X was valid.

Of course, I must be crazy to follow convention like that.

Sometimes in C non-zero means error and zero means success. Assuming this is C++ (or something else with a "bool" primitive), someone may have decided to carry over that convention just for consistency.

Still kind of stupid, though.

[PM 2Ring]

See, now I would think that "bool ValidateX(X)" would return true if X was valid.

Of course, I must be crazy to follow convention like that.

Sometimes in C non-zero means error and zero means success. Assuming this is C++ (or something else with a "bool" primitive), someone may have decided to carry over that convention just for consistency.

Still kind of stupid, though.

Yeah. The integer error code convention is zero for no error, with various non-zero codes for actual errors. If it's a bool then true ought to mean good and false should mean bad.

And then there's the convention where valid data is returned as a non-negative integer, and negative integer(s) signify an error, or no data, eg the EOF code returned by getchar() is (generally) -1.

[Xeio]

Don't worry, neither of those apply in C#, or even in our codebase.

Doesn't matter though, they're passing in "T as IEnumerable<T>" (Hint: T is a non-enumerable-castable class, let alone an enumerable of itself) into the parameter, and the method returns a "valid" if the parameter is null. <_<

(I didn't even notice that casting bit when I wrote my previous post, so, woo...)

[jareds]

I recently noticed that, in Visual Studio 2010, C++ Intellisense (underlining errors, etc. in the editor) is almost surely using the EDG front-end rather than the front-end of the actual MSVC compiler (various cases rejected by EDG and Intellisense but accepted by the MSVC compiler, and vice versa). In fact this can be verified here http://blogs.msdn.com/b/vcblog/archive/2009/05/27/rebuilding-intellisense.aspx.

Nothing against EDG, but I do wonder if they are eventually going to be the only proprietary C++ front-end.

[EvanED]

The EDG folks are crazy good, as are their front ends. (Though looking into it, it seems that at least Comeau is actually lagging well behind both GCC and MSVC on C++11 support.)

But as much as I'd like to see MSVC take up the EDG front end, I sort of doubt they will. That would break a lot of existing code, as the current MS compiler is way off spec in some respects. In particular, it doesn't implement two-phase name lookup for templates; practically speaking this leads to just being way too permissive, but it can lead to silent changes of behavior.

That said, I don't have much experience with using EDG compilers (mostly just an occasional flirt with Intel CC and the Comeau online try-it-out thing), but I actually really like MSVC's error messages. I tend to find them far more readable than GCC's. So hopefully either EDG is up-to-par (which wouldn't surprise me) or they bring it there before taking it up.

[jareds]

But as much as I'd like to see MSVC take up the EDG front end, I sort of doubt they will. That would break a lot of existing code, as the current MS compiler is way off spec in some respects. In particular, it doesn't implement two-phase name lookup for templates; practically speaking this leads to just being way too permissive, but it can lead to silent changes of behavior.

Just to be clear, I meant that Intellisense resembles EDG --microsoft --microsoft_bugs, not default EDG; and presumably default EDG would be named something like --strict under MSVC if they switched front-ends. That said, obviously their compatibility is imperfect or we wouldn't be having this conversation--and it could easily be enough to matter in practice. Overall, I do think the idea of MS replacing their front-end is very speculative, but I was quite surprised to notice that Intellisense was not using the same front-end as the compiler.

That said, I don't have much experience with using EDG compilers (mostly just an occasional flirt with Intel CC and the Comeau online try-it-out thing), but I actually really like MSVC's error messages. I tend to find them far more readable than GCC's. So hopefully either EDG is up-to-par (which wouldn't surprise me) or they bring it there before taking it up.

And I don't have enough experience with MSVC to compare...but I do think EDG's are better than GCC's.

[Steax]

I continuously get this kind of code from my coworkers who try to write PHP:

Code: Select all



function get_user_email($id){
    $q = query("SELECT `email` from `users` where `id`='".mysql_real_escape_string($id)."'");
    if(mysql_num_rows($q) == 0){
        return RESULT_ERROR_NOT_FOUND
    }else{
        return array(mysql_result($q, 0, 'email'), RESULT_OK);
    }
}

function display_user_email($id){
    $email = get_user_email($id);
    if(is_okay($email)) echo "<div>" . $email[0] . "</div>"
}
 

Whoever invented exceptions is now thrashing uncontrollably in his/her grave.

[EvanED]

But as much as I'd like to see MSVC take up the EDG front end, I sort of doubt they will. That would break a lot of existing code, as the current MS compiler is way off spec in some respects. In particular, it doesn't implement two-phase name lookup for templates; practically speaking this leads to just being way too permissive, but it can lead to silent changes of behavior.

Just to be clear, I meant that Intellisense resembles EDG --microsoft --microsoft_bugs, not default EDG; and presumably default EDG would be named something like --strict under MSVC if they switched front-ends.

Oh, EDG already being able to somewhat emulate MSVC didn't occur to me. Do you know how close it gets? E.g. does it at least mostly get their template lookup right? Or is it just some stuff more on the parser side?

Actually what would be really cool is if they implemented both, then had a mode that was like --strict-assist or something. I bet it could even automatically add typename where needed, warn in locations where changing to two-phase lookup changes name bindings, and maybe some other things.

I was quite surprised to notice that Intellisense was not using the same front-end as the compiler.

Same here. I heard about it a while back but don't use VS enough (unfortunately) to really have an opportunity to notice much. I think there have been a few times when it's underlined stuff wrong, but I just chalked that up to not getting preprocessor flags quite right and stuff.