PHP: why GOTO instead of recursion?

https://github.com/igorw/retry/issues/3

by napolux • 12 years ago

84 74 12 years ago

github.com

Guvante• 12 years ago

So the short version is "PHP doesn't perform optimizations so I had to manually include them".

canadaj• 12 years ago

Sure looks like it to me. Compiler can't figure out how to replace 'true' with '\true'? Better do it myself.

Implicated• 12 years ago

I stand corrected. Compiler.

Someone12349• 12 years ago

In the context there was no compiler, there was an interpreter. PHP compilers (which do exist) might remove this type of redundancy, the PHP interpreter likely won't for reasons stated elsewhere in the comments.

kstrauser4• 12 years ago

That whole thread made me cringe. Yes, it's a clever microoptimization, but he never mentions whether a profiler showed it to actually be beneficial. That might save 5 cycles out of a billion, for instance, at the cost of making the code "weird".

Don't try to out-clever the compiler. You might be able to do it, but there's no guarantee that tomorrow's compiler update won't be smarter than your today's workaround, leaving you with code that is both "weird" and slow.

Crito• 12 years ago

"there's no guarantee that tomorrow's compiler update won't be smarter than your today's workaround, leaving you with code that is both "weird" and slow."

Yeah, there is no guarantee, but let's be realistic here. That bug for PHP's incorrect parsing of octal integer literals has been unfixed for what, a decade now?

ajanuary3• 12 years ago

Don't try to out-clever the compiler. You might be able to do it, but there's no guarantee that tomorrow's compiler update won't be smarter than your today's workaround, leaving you with code that is both "weird" and slow.

I think that's definitely true in the general case.

However, for a 19 line library whose focus is "fast retrying" where the goto construct is "goto beginning of function", I don't think it's really hugely "weird".

In this case it's also extremely unlikely to get slower.

martin_• 12 years ago

To add context, igorw is a contributor on php-internals has committed multiple patches/improvements[1] to php-src as recent as PHP 5.6

Great explanation!

https://github.com/php/php-src/commits?author=igorw

realusername• 12 years ago

I knew that the php interpreter was not very clever but the most surprising part is this one :

There is actually a difference between "while (true)" and "while (\true)", and it's not even capable to optimize this as a standard JMP. I can't believe that really simple things like this are still not optimized in the interpreter.

psuter• 12 years ago

From the reply: "This requires doing a namespace lookup against igorw\true".

It has been a long time since I last wrote PHP code, but is it possible to dynamically redefine the value of `true`? That would (partially) explain why you can't statically evaluate it to `\true`.

jbrooksuk• 12 years ago

I've never tried, but this (http://us2.php.net/manual/en/function.runkit-constant-redefi...) seems relevant.

masklinn10• 12 years ago

That would (partially) explain why you can't statically evaluate it to `\true`.

Yes, but that only requires a guard clause. That's what CPython 2 does in the same situation:

    >>> import dis
    >>> def foo():
    ...     while True:
    ...             print "ok"
    ... 
    >>> dis.dis(foo)
      2           0 SETUP_LOOP              15 (to 18)
            >>    3 LOAD_GLOBAL              0 (True)
                  6 POP_JUMP_IF_FALSE       17

      3           9 LOAD_CONST               1 ('ok')
                 12 PRINT_ITEM          
                 13 PRINT_NEWLINE       
                 14 JUMP_ABSOLUTE            3
            >>   17 POP_BLOCK           
            >>   18 LOAD_CONST               0 (None)
                 21 RETURN_VALUE

(in Python 3, True and False are keywords so the loop prelude can be elided entirely)

Someone12349• 12 years ago

The key word there is interpreter. When you compile you have near infinite time to try and optimise and remove redundancy. When you interpret (i.e. real-time compilation) you have to decide if the redundancy check is more expensive than removing the redundancy.

Take this example, it is one extra OP Code instruction, which might generate a handful of ASM that actually gets run. How many ASM instructions would it take to detect this (and make sure there are no externalities) and remove it? I'd imagine considerably more.

And worse still you're also performing all of these redundancy checks even when there is nothing to remove. So now you've increased the cost to ALL loops, not just static ones with no purpose.

TL;DR: It is more expensive to detect and remove than to ignore.

masklinn10• 12 years ago

Take this example, it is one extra OP Code instruction, which might generate a handful of ASM that actually gets run. How many ASM instructions would it take to detect this (and make sure there are no externalities) and remove it? I'd imagine considerably more.

The detection is only done one while generating the bytecode, but the extraneous bytecode instructions have to be executed every time the loop… loops.

Not only that but it's a very simple optimisation to apply (even CPython is able to discover it).

excitom• 12 years ago

It seems to me that if you're this worried about performance you should be looking at something like HipHop.

peterwwillis• 12 years ago

Longest explanation of "function calls [and loops] are expensive" ever. Also the best example of why over-optimization is a completely valid thing to do if you're bored and have a debugger handy.

masklinn10• 12 years ago

The explanation is mostly "Zend's bytecode compiler is retarded and can't do trivial optimisations so you have to handroll them".

Piskvorrr• 12 years ago

Very...clever[1].

http://www.catb.org/jargon/html/K/kluge.html

wyager2• 12 years ago

If you need to use GOTOs in your code for speed and you're not coding in C, you should probably find a faster platform.

quotemstr• 12 years ago

Not all useful control flow graphs can be described with while and if. It's perfectly fine to use goto. Look at some of the contortions people need to go through when avoiding it. Avoiding goto often makes programs harder to follow.

Why is it inexperienced programmers go through a phase where they think goto is evil incarnate? I wish Dijkstra had never written that damn paper.

wyager2• 12 years ago

Not all useful control flow graphs can be described with while and if.

They can be described with if, while, for, case, pattern matching, recursion, etc. There is almost always a better alternative.

Why is it inexperienced programmers go through a phase where they think goto is evil incarnate?

It seems to me that most people who support general usage of GOTO are inexperienced and don't necessarily realize it. They lack knowledge of better control structures. See the blub paradox.

Besides, the justification here is performance, not readability. I'm saying that performance in particular is a bad justification for using GOTO.

jtc3312• 12 years ago

Agreed that sometimes it's right for control-flow purposes. Hence why I pointed out in the OP thread that I've used it before. But that's not the justification being used by the code author; his justification is purely performance.

> "Why is it inexperienced programmers go through a phase where they think goto is evil incarnate? I wish Dijkstra had never written that damn paper."

You seem to be referencing my comments. I'm not inexperienced, and I don't think goto is evil incarnate. I do believe that optimizing for individual CPU cycles in PHP is generally pointless.

Hawkee2• 12 years ago

Great read, but I think it's a few years too late. When you compare how PHP is trending against languages like Python it's following a similar path as perl, https://www.google.com/trends/explore#q=php%2C%20javascript%... - I'm surprised JavaScript isn't having much of an upturn lately.

goshx• 12 years ago

On a side note, Monty Python (or the snake) are not programming languages. Check the "Related searches" section at the bottom and click "Python".

Hawkee2• 12 years ago

Silly me. I forget that not everybody lives in the world of development. Considering this it seems every language is experiencing a downturn. To me this looks more like a fault in Google's trend calculation. Maybe it doesn't consider search inflation from year to year.

sebs__• 12 years ago

Using the same kind of metric, certain male bodyparts seem to be trending as programming lang. as well.

https://www.google.com/trends/explore#q=php%2C%20javascript%...

justincormack2• 12 years ago

Languages without tail recursion elimination are very annoying. Looking at you JavaScript.

phkahler• 12 years ago

> Languages without tail recursion elimination are very annoying. Looking at you JavaScript.

Programmers who use tail recursion instead of a loop are very annoying. Looking at you justincormack ;-)

Seriously, when I first saw recursion explained using the fibonacci sequence I was really scratching my head. I wondered why anyone would do that. Now that I'm older and wiser, I can see that it's natural to a mathematician. Why anyone who claims to be a programmer or compsci person would do it is beyond me. I suppose it's a simple example, but I'm a fan of introducing concepts with examples that really want the technique in question as a solution.

emodendroket2• 12 years ago

How about a depth-first search, then?

toast0• 12 years ago

Some languages don't have loops, so you have to use tail recursion. Sometimes, tail recursion is more clear or concise than a loop:

Erlang:

  fib(1) -> 1;
  fib(2) -> 1;

  fib(N) when N > 2 -> fib(fib(1), fib(2), N - 2).
  fib(_A, B, 0) -> B;
  fib(A, B, N) when N > 0 -> fib(B, A + B, N - 1).

  int fib(int count) {
     int a = 1, b =1;
     while (--count >= 2) {
        int next = a + b;
        a = b;
        b = next;
     }
     return b;
  }

Notes: Erlang version will crash if given value is less than 1; C version will just return 1 for anything outside of accepted range. Erlang integers become large integers automatically, C integers will overflow. If you prefer fib(0) == 0, you can adjust it, with more work for C, because I cheated the initial conditions/bounds check.

(I shrunk the C code a bit more than originally written, and it's now pretty similar in length... I guess it's about the same... which is the point :)

justincormack2• 12 years ago

There are excellent uses. State machines as tail recursive functions are great, but in JavaScript you need a trampoline handler, which is much messier.

powera• 12 years ago

The real question is: if you care that much about performance, why would you use the un-optimized PHP runtime?

thathonkey• 12 years ago

Sometimes people optimize for fun, out of sheer curiosity, or some combination of things like that. Not everything we do as programmers has to be for some solely practical purpose. Having some fun with opcode optimization is certainly permissible and if you program a lot in PHP, learning how things work under the hood will make you a better PHP programmer in general.

krazydad• 12 years ago

I like including a few GOTOs just to piss off the Dijkstra acolytes, one of whom is quite vocal in the thread.

jtc3312• 12 years ago

Since I see you're referencing my posts in that thread, you'll notice that I also point out that I understand that GOTO is not always bad, and, in fact, have used it before in production code.

seccess• 12 years ago

In my experience, PHP can have really unexpected runtime performance. This site has some interesting benchmarks highlighting this: http://www.phpbench.com/

I'd like to hear of any other PHP performance gotchas people know of. One of my favorites: passing function arguments by value is faster than passing by reference.

PHP: why GOTO instead of recursion?

Related Stories