Performance difference between branch prediction and branch target prediction? - c++

I'm writing some audio code where basically everything is a tiny loop. Branch prediction failures as I understand them is a big enough performance issue that I struggle to keep the code branch free. But there is only so far that can take me, which got me wondering about the different kinds of branching.
In c++, the conditional branch to fixed target:
int cond_fixed(bool p) {
if (p) return 10;
return 20;
}
And (if I understand this question correctly), the unconditional branch to variable target:
struct base {
virtual int foo() = 0;
};
struct a : public base {
int foo() { return 10; }
};
struct b : public base {
int foo() { return 20; }
};
int uncond_var(base* p) {
return p->foo();
}
Are there performance differences? It seems to me that if one of the two methods were obviously faster than the other, the compiler would simply transform the code to match.
For those cases where branch prediction is of very high importance, what details regarding performance are useful to know?
EDIT: The actual operation of x : 10 ? 20 is merely a place holder. The actual operation following the branch is at least complex enough that doing both is inefficient. Additionally, if I had enough information to sensibly use __builtin_expect, branch prediction would be a non-issue in this case.

Side note: if you have a code like
if (p) a = 20; else a = 10;
then there isn't any branch. The compiler is using a conditional move (see: Why is a conditional move not vulnerable for Branch Prediction Failure?)

You didn't mention your compiler. I once used GCC for a performance critical application (a contest at my university actually) and I remember that GCC has the __builtin_expect macro. I went through all the conditions in my code and ended up with 5-10% speedup, which I found to be amazing, given the fact that I payed attention to pretty much everything I knew (memory-layout etc.) and that I didn't change anything regarding the algorithm itself.
The algorithm was a pretty basic depth-search by the way. And I ran it on a Core 2 Duo, not sure which ones though.

Related

Is it a good practice to use naked brackets inside functions?

Technically, in C++ we have the possibility to use curly-braces to declare a new scope. For example in this function, which swaps two numbers
void swap_int(int& first, int& second)
{
int temp = first;
first = second;
second = temp;
}
we could also declare temp inside its own block:
void swap_int(int& first, int& second)
{
// Do stuf...
{
int temp = first;
first = second;
second = temp;
}
// Do other stuff...
}
This obviously has the advantage that temp is deleted directly when it is not needed anymore.
However, in the code I write I never use it. Also, in code from 3rd-party libraries I almost never see it at all.
Why is it not used publically? Does it bring any performance increase at all, or does it just mean additional typing-work?
I don't see anything wrong per-se with naked brackets. They're a part of the language, and they're well defined. Historically, one place I have found them useful is when working with code that uses status codes instead of exceptions, while keeping const goodness:
const StatusCode statusCode = DoThing();
if (statusCode == STATUS_SUCCESS)
Foo();
else
Bar();
const StatusCode statusCode2 = DoAnotherThing(); // Eww variable name.
...
The alternative would be:
{
const StatusCode statusCode = DoThing();
if (statusCode == STATUS_SUCCESS)
Foo();
else
Bar();
}
{
// Same variable name, used for same purpose, easy to
// find/replace, and has const guarantees. Great success.
const StatusCode statusCode = DoAnotherThing();
...
}
The same applies to objects such as thread lockers that use RAII as well (Mutex objects, semaphores, etc), or generally any kind of resource that you may want to have an extremely short lifetime (file handles, for example).
Personally, I think the reason it's rare is that it can be indicative of a code smell (although not always). Where there are naked brackets, there may be an opportunity to factor out a function.
To take your example, if there is more than one job for swap_int, then the function is doing more than one thing. By extracting the actual swap code into another function, you can encourage reuse! For example:
template <typename T>
void swap_anything(T &first, T& second)
{
T temp = first;
first = second;
second = temp;
}
// -------------------------------------------
void swap_int(int& first, int& second)
{
// Do stuff...
swap_anything(first, second);
// Do other stuff...
}
Sometimes it's good practice (although I dislike that term, as it's subjective and context-specific), but like many things in life, taking it to the extreme (on either end of the spectrum) is a bad idea.
You'll see new scopes introduced sometimes in C++ where RAII is important, like when dealing with thread locks. Sometimes the precise moment of when an object is created and destroyed is very important and needs to be controlled. In those situations, introducing a new scope is a very useful and often-used way of accomplishing this.
But that's not frequently the case. Most of the objects we (the broad programming community) use don't have strict lifetimes that need to be so carefully managed. As such, it's not worth arbitrarily introducing new scopes to manage lifetimes of objects whose lifetimes aren't worth micromanaging.
If you do, you'll decrease the signal-to-noise ratio, and people will have a hard time telling which scopes are introduced to carefully control important resources, and which scopes are not. This can make it easy to introduce bugs when refactoring or developing code across teams. At the very least, you'll make programming a whole lot more tedious, which sucks, and you should generally avoid that if you can or else a violent psychopath may take it out on you.

How to organize time invariant checking with D contracts?

For example, I have to assure that a certain function for a certain real-time system works for 20 ms or less. I can simply measure time at the beginning of a function and at the end of it, then assert the difference to be satisfactory. And I do this in C++.
But this look pretty much like contract, except time checking is a post-condition, and time measurement at the beginning is not a condition at all. It would be nice to put it into contract not only for the notation of it, but for building reasons as well.
So I wonder, can I use contract capabilities to check the time of function working?
Sort of, but not really well. The reason is variables declared in the in{} block are not visible in the out{} block. (There's been some discussing about changing this, so it can check pre vs post state by making a copy in the in block, but nothing has been implemented.)
So, this will not work:
void foo()
in { auto before = Clock.currTime(); }
out { assert(Clock.currTime - before < dur!"msecs"(20)); }
body { ... }
The variable from in won't carry over to out, giving you an undefined identifier error. But, I say "sort of" though because there is a potential workaround:
import std.datetime;
struct Foo {
SysTime test_before;
void test()
in {
test_before = Clock.currTime();
}
out {
assert(Clock.currTime - test_before < dur!"msecs"(20));
}
body {
}
}
Declaring the variable as a regular member of the struct. But this would mean a lot of otherwise useless variables for each function, wouldn't work with recursion, and just pollutes the member namespace.
Part of me is thinking you could do your own stack off to the side and have in{} push the time, then out{} pops it and checks.... but a quick test shows that it is liable to break once inheritance gets involved. If you repeat the in{} block each time, it might work. But this strikes me as awfully brittle. The rule with contract inheritance is ALL of the out{} blocks of the inheritance tree need to pass, but only any ONE of the in{} blocks needs to pass. So if you had a different in{} down the chain, it might forget to push the time, and then when out tries to pop it, your stack would underflow.
// just for experimenting.....
SysTime[] timeStack; // WARNING: use a real stack here in production, a plain array will waste a *lot* of time reallocating as you push and pop on to it
class Foo {
void test()
in {
timeStack ~= Clock.currTime();
}
out {
auto start = timeStack[$-1];
timeStack = timeStack[0 .. $-1];
assert(Clock.currTime - start < dur!"msecs"(20));
import std.stdio;
// making sure the stack length is still sane
writeln("stack length ", timeStack.length);
}
body { }
}
class Bar : Foo {
override void test()
in {
// had to repeat the in block on the child class for this to work at all
timeStack ~= Clock.currTime();
}
body {
import core.thread;
Thread.sleep(10.msecs); // bump that up to force a failure, ensuring the test is actually run
}
}
That seems to work, but I think it is more trouble than it's worth. I expect it would break somehow as the program got bigger, and if your test breaks your program, that kinda defeats the purpose.
I'd probably do it as a unittest{}, if only checking with explicit tests fulfills you requirements (however, note that contracts, like most asserts in D, are removed if you compile with the -release switch, so they won't actually be checked in release versions either. If you need it to reliably fail, throw an exception rather than assert, since that will always work, in debug and release modes.).
Or you could do it with an assert in the function or a helper struct or whatever, similar to C++. I'd use a scope guard:
void test() {
auto before = Clock.currTime();
scope(exit) assert(Clock.currTime - before < dur!"msecs"(20)); // or import std.exception; and use enforce instead of assert if you want it in release builds too
/* write the rest of your function */
}
Of course, here you'll have to copy it in the subclasses too, but it seems like you'd have to do that with the in{} blocks anyway, so meh, and at least the before variable is local.
Bottom line, I'd say you're probably best off doing it more or less the same way you have been in C++.

C/C++ (Other Languages Too?) Conditional Early Return Good Code Practice

Recently, I was reviewing some code I maintain and I noticed a practice different than what I am used to. As a result, I'm wondering which method to use when performing an early return in a function.
Here's some example:
Version 1:
int MyFunction(int* ptr)
{
if(!ptr) { // oh no, NULL pointer!
return -1; // what was the caller doing? :(
}
// other code goes here to do work on the pointer
// ...
return 0; // we did it!
}
Version 2:
int MyFunction(int* ptr)
{
if(!ptr) { // oh no, NULL pointer!
return -1; // what was the caller doing? :(
} else { // explicitly show that this only gets call when if statement fails
// other code goes here to do work on the pointer
// ...
return 0; // hooray!
}
}
As a result, I'm wondering which is considered the "best practice" for those of you who have endured (and survived) many code reviews. I know each effectively does the same thing, but does the "else" add much in terms of readability and clarity? Thanks for the help.
The else would only add clarity if the else clause is short, a few lines of code at best. And if you have several initial conditions you want to check, the source gets cluttered very quickly.
The only time I would use an else if it is a small function with a small else, meaning less than about 10 source lines, and there are no other initial checks to make.
In some cases I have used a single loop so that a series of initial checks can use a break to leave.
do {
...
} while (0);
I am loathe to use a goto which is practically guaranteed to get at least one true believer of goto less programming up in arms.
So much would depend on any code standards of your organization. I tend to like minimalism so I use the first version you provide without the else.
I might also do something like the following in a smaller function say less than 20 or 30 lines:
int MyFunction(int* ptr)
{
int iRetStatus = -1; // we have an error condition
if (ptr) { // good pointer
// stuff to do in this function
iRetStatus = 0;
}
return iRetStatus; // we did it!
}
The only problem with returns in the body of the function is that sometimes people scanning the function do not realize that there is a return. In small functions where everything can be pretty much seen on a single screen, the chance of missing a return is pretty small. However for large functions, returns in the middle can be missed especially large complex functions that have gone through several maintenance cycles and had a lot of cruft and work arounds put into them.

goto Optimization Refactor

I have a "MyFunction" I keep obsessing over if I should or shouldn't use goto on it and in similar (hopefully rare) circumstances. So I'm trying to establish a hard-and-fast habit for this situation. To-do or not-to-do.
int MyFunction()
{ if (likely_condition)
{
condition_met:
// ...
return result;
}
else /*unlikely failure*/
{ // meet condition
goto condition_met;
}
}
I was intending to net the benefits of the failed conditional jump instruction for the likely case. However I don't see how the compiler could know which to streamline for case probability without something like this.
it works right?
are the benefits worth the confusion?
are there better (less verbose, more structured, more expressive) ways to enable this optimization?
It appears to me that the optimization you're trying to do is mostly obsolete. Most modern processors have branch prediction built in, so (assuming it's used enough to notice) they track how often a branch is taken or not and predict whether the branch is likely to be taken or not based on its past pattern of being taken or not. In this case, speed depends primarily on how accurate that prediction is, not whether the prediction is for taken vs. not taken.
As such, you're probably best off with rather simpler code:
int MyFunction() {
if (!likely_condition) {
meet_condition();
}
// ...
return result;
}
A modern CPU will take that branch either way with equal performance if it makes the correct branch prediction. So if that is in an inner loop, the performance of if (unlikely) { meet condition } common code; will match what you have written.
Also, if you spell out the common code in both branches the compiler will generate code that is identical to what you have written: The common case will be emitted for the if clause and the else clause will jmp to the common code. You see this all the time with simpler terminal cases like *out = whatever; return result;. When debugging it can be hard to tell which return you're looking at because they've all been merged.
It looks like the code should work as you expect as long as condition_met: doesn't skip variable initializations.
No, and you don't even know that the obfuscated version compiles into more optimal code. Compiler optimizations (and processor branch prediction) are getting very smart in recent times.
3.
int MyFunction()
{
if (!likely_condition)
{
// meet condition
}
condition_met:
// ...
return result;
}
or, if it helps your compiler (check the assembly)
int MyFunction()
{
if (likely_condition); else
{
// meet condition
}
condition_met:
// ...
return result;
}
I would highly recommend using the __builtin_expect() macro (GCC) or alike for your particular C++ compiler (see Portable branch prediction hints) instead of using goto:
int MyFunction()
{ if (__builtin_expect(likely_condition))
{
// ...
return result;
}
else /*unlikely failure*/
{ // meet condition
}
}
As others also mentioned goto is error prone and evil from the bones.

To GOTO or not to GOTO? [closed]

Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 4 years ago.
Improve this question
Currently I am working on a project where goto statements are heavely used. The main purpose of goto statements is to have one cleanup section in a routine rather than multiple return statements.
Like below:
BOOL foo()
{
BOOL bRetVal = FALSE;
int *p = NULL;
p = new int;
if (p == NULL)
{
cout<<" OOM \n";
goto Exit;
}
// Lot of code...
Exit:
if(p)
{
delete p;
p = NULL;
}
return bRetVal;
}
This makes it much easier as we can track our clean up code at one section in code, that is, after the Exit label.
However, I have read many places it's bad practice to have goto statements.
Currently I am reading the Code Complete book, and it says that we need to use variables close to their declarations. If we use goto then we need to declare/initialize all variables before first use of goto otherwise the compiler will give errors that initialization of xx variable is skipped by the goto statement.
Which way is right?
From Scott's comment:
It looks like using goto to jump from one section to another is bad as it makes the code hard to read and understand.
But if we use goto just to go forward and to one label then it should be fine(?).
I am not sure what do you mean by clean up code but in C++ there is a concept called "resource acquisition is initialization" and it should be the responsibility of your destructors to clean up stuff.
(Note that in C# and Java, this is usually solved by try/finally)
For more info check out this page:
http://www.research.att.com/~bs/bs_faq2.html#finally
EDIT: Let me clear this up a little bit.
Consider the following code:
void MyMethod()
{
MyClass *myInstance = new MyClass("myParameter");
/* Your code here */
delete myInstance;
}
The problem: What happens if you have multiple exits from the function? You have to keep track of each exit and delete your objects at all possible exits! Otherwise, you will have memory leaks and zombie resources, right?
The solution: Use object references instead, as they get cleaned up automatically when the control leaves the scope.
void MyMethod()
{
MyClass myInstance("myParameter");
/* Your code here */
/* You don't need delete - myInstance will be destructed and deleted
* automatically on function exit */
}
Oh yes, and use std::unique_ptr or something similar because the example above as it is is obviously imperfect.
I've never had to use a goto in C++. Ever. EVER. If there is a situation it should be used, it's incredibly rare. If you are actually considering making goto a standard part of your logic, something has flown off the tracks.
There are basically two points people are making in regards to gotos and your code:
Goto is bad. It's very rare to encounter a place where you need gotos, but I wouldn't suggest striking it completely. Though C++ has smart enough control flow to make goto rarely appropriate.
Your mechanism for cleanup is wrong: This point is far more important. In C, using memory management on your own is not only OK, but often the best way to do things. In C++, your goal should be to avoid memory management as much as possible. You should avoid memory management as much as possible. Let the compiler do it for you. Rather than using new, just declare variables. The only time you'll really need memory management is when you don't know the size of your data in advance. Even then, you should try to just use some of the STL collections instead.
In the event that you legitimately need memory management (you have not really provided any evidence of this), then you should encapsulate your memory management within a class via constructors to allocate memory and deconstructors to deallocate memory.
Your response that your way of doing things is much easier is not really true in the long run. Firstly, once you get a strong feel for C++ making such constructors will be 2nd nature. Personally, I find using constructors easier than using cleanup code, since I have no need to pay careful attention to make sure I am deallocating properly. Instead, I can just let the object leave scope and the language handles it for me. Also, maintaining them is MUCH easier than maintaining a cleanup section and much less prone to problems.
In short, goto may be a good choice in some situations but not in this one. Here it's just short term laziness.
Your code is extremely non-idiomatic and you should never write it. You're basically emulating C in C++ there. But others have remarked on that, and pointed to RAII as the alternative.
However, your code won't work as you expect, because this:
p = new int;
if(p==NULL) { … }
won't ever evaluate to true (except if you've overloaded operator new in a weird way). If operator new is unable to allocate enough memory, it throws an exception, it never, ever returns 0, at least not with this set of parameters; there's a special placement-new overload that takes an instance of type std::nothrow and that indeed returns 0 instead of throwing an exception. But this version is rarely used in normal code. Some low-level codes or embedded device applications could benefit from it in contexts where dealing with exceptions is too expensive.
Something similar is true for your delete block, as Harald as said: if (p) is unnecessary in front of delete p.
Additionally, I'm not sure if your example was chose intentionally because this code can be rewritten as follows:
bool foo() // prefer native types to BOOL, if possible
{
bool ret = false;
int i;
// Lots of code.
return ret;
}
Probably not a good idea.
In general, and on the surface, there isn't any thing wrong with your approach, provided that you only have one label, and that the gotos always go forward. For example, this code:
int foo()
{
int *pWhatEver = ...;
if (something(pWhatEver))
{
delete pWhatEver;
return 1;
}
else
{
delete pWhatEver;
return 5;
}
}
And this code:
int foo()
{
int ret;
int *pWhatEver = ...;
if (something(pWhatEver))
{
ret = 1;
goto exit;
}
else
{
ret = 5;
goto exit;
}
exit:
delete pWhatEver;
return ret;
}
really aren't all that different from each other. If you can accept one, you should be able to accept the other.
However, in many cases the RAII (resource acquisition is initialization) pattern can make the code much cleaner and more maintainable. For example, this code:
int foo()
{
Auto<int> pWhatEver = ...;
if (something(pWhatEver))
{
return 1;
}
else
{
return 5;
}
}
is shorter, easier to read, and easier to maintain than both of the previous examples.
So, I would recommend using the RAII approach if you can.
Your example is not exception safe.
If you are using goto to clean up the code then, if an exception happens before the cleanup code, it is completely missed. If you claim that you do not use exceptions then you are mistaken because the new will throw bad_alloc when it does not have enough memory.
Also at this point (when bad_alloc is thrown), your stack will be unwound, missing all the cleanup code in every function on the way up the call stack thus not cleaning up your code.
You need to look to do some research into smart pointers. In the situation above you could just use a std::auto_ptr<>.
Also note in C++ code there is no need to check if a pointer is NULL (usually because you never have RAW pointers), but because new will not return NULL (it throws).
Also in C++ unlike (C) it is common to see early returns in the code. This is because RAII will do the cleanup automatically, while in C code you need to make sure that you add special cleanup code at the end of the function (a bit like your code).
I think other answers (and their comments) have covered all the important points, but here's one thing that hasn't been done properly yet:
What your code should look like instead:
bool foo() //lowercase bool is a built-in C++ type. Use it if you're writing C++.
{
try {
std::unique_ptr<int> p(new int);
// lots of code, and just return true or false directly when you're done
}
catch (std::bad_alloc){ // new throws an exception on OOM, it doesn't return NULL
cout<<" OOM \n";
return false;
}
}
Well, it's shorter, and as far as I can see, more correct (handles the OOM case properly), and most importantly, I didn't need to write any cleanup code or do anything special to "make sure my return value is initialized".
One problem with your code I only really noticed when I wrote this, is "what the hell is bRetVal's value at this point?". I don't know because, it was declared waaaaay above, and it was last assigned to when? At some point above this. I have to read through the entire function to make sure I understand what's going to be returned.
And how do I convince myself that the memory gets freed?
How do I know that we never forget to jump to the cleanup label? I have to work backwards from the cleanup label, finding every goto that points to it, and more importantly, find the ones that aren't there. I need to trace through all paths of the function just to be sure that the function gets cleaned up properly. That reads like spaghetti code to me.
Very fragile code, because every time a resource has to be cleaned up you have to remember to duplicate your cleanup code. Why not write it once, in the type that needs to be cleaned up? And then rely on it being executed automatically, every time we need it?
In the eight years I've been programming I've used goto a lot, most of that was in the first year when I was using a version of GW-BASIC and a book from 1980 that didn't make it clear goto should only be used in certain cases. The only time I've used goto in C++ is when I had code like the following, and I'm not sure if there was a better way.
for (int i=0; i<10; i++) {
for (int j=0; j<10; j++)
{
if (somecondition==true)
{
goto finish;
}
//Some code
}
//Some code
}
finish:
The only situation I know of where goto is still used heavily is mainframe assembly language, and the programmers I know make sure to document where code is jumping and why.
As used in the Linux kernel, goto's used for cleanup work well when a single function must perform 2 or more steps that may need to be undone. Steps need not be memory allocation. It might be a configuration change to a piece of code or in a register of an I/O chipset. Goto's should only be needed in a small number of cases, but often when used correctly, they may be the best solution. They are not evil. They are a tool.
Instead of...
do_step1;
if (failed)
{
undo_step1;
return failure;
}
do_step2;
if (failed)
{
undo_step2;
undo_step1;
return failure;
}
do_step3;
if (failed)
{
undo_step3;
undo_step2;
undo_step1;
return failure;
}
return success;
you can do the same with goto statements like this:
do_step1;
if (failed) goto unwind_step1;
do_step2;
if (failed) goto unwind_step2;
do_step3;
if (failed) goto unwind_step3;
return success;
unwind_step3:
undo_step3;
unwind_step2:
undo_step2;
unwind_step1:
undo_step1;
return failure;
It should be clear that given these two examples, one is preferable to the other. As to the RAII crowd... There is nothing wrong with that approach as long as they can guarantee that the unwinding will always occur in exactly reverse order: 3, 2, 1. And lastly, some people do not use exceptions in their code and instruct the compilers to disable them. Thus not all code must be exception safe.
You should read this thread summary from the Linux kernel mailing lists (paying special attention to the responses from Linus Torvalds) before you form a policy for goto:
http://kerneltrap.org/node/553/2131
In general, you should design your programs to limit the need for gotos. Use OO techniques for "cleanup" of your return values. There are ways to do this that don't require the use of gotos or complicating the code. There are cases where gotos are very useful (for example, deeply nested scopes), but if possible should be avoided.
The downside of GOTO is pretty well discussed. I would just add that 1) sometimes you have to use them and should know how to minimize the problems, and 2) some accepted programming techniques are GOTO-in-disguise, so be careful.
1) When you have to use GOTO, such as in ASM or in .bat files, think like a compiler. If you want to code
if (some_test){
... the body ...
}
do what a compiler does. Generate a label whose purpose is to skip over the body, not to do whatever follows. i.e.
if (not some_test) GOTO label_at_end_of_body
... the body ...
label_at_end_of_body:
Not
if (not some_test) GOTO the_label_named_for_whatever_gets_done_next
... the body ...
the_label_named_for_whatever_gets_done_next:
In otherwords, the purpose of the label is not to do something, but to skip over something.
2) What I call GOTO-in-disguise is anything that could be turned into GOTO+LABELS code by just defining a couple macros. An example is the technique of implementing finite-state-automata by having a state variable, and a while-switch statement.
while (not_done){
switch(state){
case S1:
... do stuff 1 ...
state = S2;
break;
case S2:
... do stuff 2 ...
state = S1;
break;
.........
}
}
can turn into:
while (not_done){
switch(state){
LABEL(S1):
... do stuff 1 ...
GOTO(S2);
LABEL(S2):
... do stuff 2 ...
GOTO(S1);
.........
}
}
just by defining a couple macros. Just about any FSA can be turned into structured goto-less code. I prefer to stay away from GOTO-in-disguise code because it can get into the same spaghetti-code issues as undisguised gotos.
Added: Just to reassure: I think one mark of a good programmer is recognizing when the common rules don't apply.
Using goto to go to a cleanup section is going to cause a lot of problems.
First, cleanup sections are prone to problems. They have low cohesion (no real role that can be described in terms of what the program is trying to do ), high coupling (correctness depends very heavily on other sections of code), and are not at all exception-safe. See if you can use destructors for cleanup. For example, if int *p is changed to auto_ptr<int> p, what p points to will be automatically released.
Second, as you point out, it's going to force you to declare variables long before use, which will make it harder to understand the code.
Third, while you're proposing a fairly disciplined use of goto, there's going to be the temptation to use them in a looser manner, and then the code will become difficult to understand.
There are very few situations where a goto is appropriate. Most of the time, when you are tempted to use them, it's a signal that you're doing things wrong.
The entire purpose of the every-function-has-a-single-exit-point idiom in C was to put all the cleanup stuff in a single place. If you use C++ destructors to handle cleanup, that's no longer necessary -- cleanup will be done regardless of how many exit points a function has. So in properly-designed C++ code, there's no longer any need for this kind of thing.
Since this is a classic topic, I will reply with Dijkstra's Go-to statement considered harmful (originally published in ACM).
Goto provides better don't repeat yourself (DRY) when "tail-end-logic" is common to some-but-not-all-cases. Especially within a "switch" statement I often use goto's when some of the switch-branches have tail-end-commonality.
switch(){
case a: ... goto L_abTail;
case b: ... goto L_abTail;
L_abTail: <commmon stuff>
break://end of case b
case c:
.....
}//switch
You have probably noticed than introducing additional curly-braces is enough to satisfy the compiler when you need such tail-end-merging in-the-middle of a routine. In other words, you don't need to declare everything way up at the top; that's inferior readability indeed.
...
goto L_skipMiddle;
{
int declInMiddleVar = 0;
....
}
L_skipMiddle: ;
With the later versions of Visual Studio detecting the use of uninitialized variables, I find myself always initializing most variables even though I think they may be assigned in all branches - it's easy to code a "tracing" statement which refs a variable that was never assigned because your mind doesn't think of the tracing statement as "real code", but of course Visual Studio will still detect an error.
Besides don't repeat yourself, assigning label-names to such tail-end-logic even seems to help my mind keep things straight by choosing nice label names. Without a meaningful label your comments might end up saying the same thing.
Of course, if you are actually allocating resources then if auto-ptr doesn't fit, you really must use a try-catch, but tail-end-merge-don't-repeat-yourself happens quite often when exception-safety is not an issue.
In summary, while goto can be used to code spaghetti-like structures, in the case of a tail-end-sequence which is common to some-but-not-all-cases then the goto IMPROVES the readability of the code and even maintainability if you would otherwise be copy/pasting stuff so that much later on someone might update one-and-not-the-other. So it's another case where being fanatic about a dogma can be counterproductive.
The only two reasons I use goto in my C++ code are:
Breaking a level 2+ nested loops
Complicated flows like this one (a comment in my program):
/* Analysis algorithm:
1. if classData [exporter] [classDef with name 'className'] exists, return it,
else
2. if project/target_codename/temp/classmeta/className.xml exist, parse it and go back to 1 as it will succeed.
3. if that file don't exists, generate it via haxe -xml, and go back to 1 as it will succeed.
*/
For code readability here, after this comment, I defined the step1 label and used it in step 2 and 3. Actually, in 60+ source files, only this situation and one 4-levels nested for are the places I used goto. Only two places.
A lot of people freak out with gotos are evil; they are not. That said, you will never need one; there is just about always a better way.
When I find myself "needing" a goto to do this type of thing, I almost always find that my code is too complex and can be easily broken up into a few method calls that are easier to read and deal with. Your calling code can do something like:
// Setup
if(
methodA() &&
methodB() &&
methodC()
)
// Cleanup
Not that this is perfect, but it's much easier to follow since all your methods will be named to clearly indicate what the problem might be.
Reading through the comments, however, should indicate that your team has more pressing issues than goto handling.
The code you're giving us is (almost) C code written inside a C++ file.
The kind of memory cleaning you're using would be OK in a C program not using C++ code/libraries.
In C++, your code is simply unsafe and unreliable. In C++ the kind of management you're asking for is done differently. Use constructors/destructors. Use smart pointers. Use the stack. In a word, use RAII.
Your code could (i.e., in C++, SHOULD) be written as:
BOOL foo()
{
BOOL bRetVal = FALSE;
std::auto_ptr<int> p = new int;
// Lot of code...
return bRetVal ;
}
(Note that new-ing an int is somewhat silly in real code, but you can replace int by any kind of object, and then, it makes more sense). Let's imagine we have an object of type T (T could be an int, some C++ class, etc.). Then the code becomes:
BOOL foo()
{
BOOL bRetVal = FALSE;
std::auto_ptr<T> p = new T;
// Lot of code...
return bRetVal ;
}
Or even better, using the stack:
BOOL foo()
{
BOOL bRetVal = FALSE;
T p ;
// Lot of code...
return bRetVal;
}
Anyway, any of the above examples are magnitudes more easy to read and secure than your example.
RAII has many facets (i.e. using smart pointers, the stack, using vectors instead of variable length arrays, etc.), but all in all is about writing as little code as possible, letting the compiler clean up the stuff at the right moment.
All of the above is valid, you might also want to look at whether you might be able to reduce the complexity of your code and alleviate the need for goto's by reducing the amout of code that is in the section marked as "lot of code" in your example. Additionaly delete 0 is a valid C++ statement
Using GOTO labels in C++ is a bad way to program, you can reduce the need by doing OO programming (deconstructors!) and trying to keep procedures as small as possible.
Your example looks a bit weird, there is no need to delete a NULL pointer. And nowadays an exception is thrown when a pointer can't get allocated.
Your procedure could just be wrote like:
bool foo()
{
bool bRetVal = false;
int p = 0;
// Calls to various methods that do algorithms on the p integer
// and give a return value back to this procedure.
return bRetVal;
}
You should place a try catch block in the main program handling out of memory problems that informs the user about the lack of memory, which is very rare... (Doesn't the OS itself inform about this too?)
Also note that there is not always the need to use a pointer, they are only useful for dynamic things. (Creating one thing inside a method not depending on input from anywhere isn't really dynamic)
I am not going to say that goto is always bad, but your use of it most certainly is. That kind of "cleanup sections" was pretty common in early 1990's, but using it for new code is pure evil.
The easiest way to avoid what you are doing here is to put all of this cleanup into some kind of simple structure and create an instance of it. For example instead of:
void MyClass::myFunction()
{
A* a = new A;
B* b = new B;
C* c = new C;
StartSomeBackgroundTask();
MaybeBeginAnUndoBlockToo();
if ( ... )
{
goto Exit;
}
if ( ... ) { .. }
else
{
... // what happens if this throws an exception??? too bad...
goto Exit;
}
Exit:
delete a;
delete b;
delete c;
StopMyBackgroundTask();
EndMyUndoBlock();
}
you should rather do this cleanup in some way like:
struct MyFunctionResourceGuard
{
MyFunctionResourceGuard( MyClass& owner )
: m_owner( owner )
, _a( new A )
, _b( new B )
, _c( new C )
{
m_owner.StartSomeBackgroundTask();
m_owner.MaybeBeginAnUndoBlockToo();
}
~MyFunctionResourceGuard()
{
m_owner.StopMyBackgroundTask();
m_owner.EndMyUndoBlock();
}
std::auto_ptr<A> _a;
std::auto_ptr<B> _b;
std::auto_ptr<C> _c;
};
void MyClass::myFunction()
{
MyFunctionResourceGuard guard( *this );
if ( ... )
{
return;
}
if ( ... ) { .. }
else
{
...
}
}
A few years ago I came up with a pseudo-idiom that avoids goto, and is vaguely similar to doing exception handling in C. It has been probably already invented by someone else so I guess I "discovered it independently" :)
BOOL foo()
{
BOOL bRetVal = FALSE;
int *p=NULL;
do
{
p = new int;
if(p==NULL)
{
cout<<" OOM \n";
break;
}
// Lot of code...
bRetVal = TRUE;
} while (false);
if(p)
{
delete p;
p= NULL;
}
return bRetVal;
}
I think using the goto for exit code is bad since there's a lot of other solutions with low overhead such as having an exit function and returning the exit function value when needed. Typically in member functions though, this shouldn't be needed, otherwise this could be indication that there's a bit too much code bloat happening.
Typically, the only exception I make of the "no goto" rule when programming is when breaking out of nested loops to a specific level, which I've only ran into the need to do when working on mathematical programming.
For example:
for(int i_index = start_index; i_index >= 0; --i_index)
{
for(int j_index = start_index; j_index >=0; --j_index)
for(int k_index = start_index; k_index >= 0; --k_index)
if(my_condition)
goto BREAK_NESTED_LOOP_j_index;
BREAK_NESTED_LOOP_j_index:;
}
That code has a bunch of problems, most of which were pointed out already, for example:
The function is too long; refactoring out some code into separate functions might help.
Using pointers when normal instances will probably work just fine.
Not taking advantage of STL types such as auto_ptr
Incorrectly checking for errors, and not catching exceptions. (I would argue that checking for OOM is pointless on the vast majority of platforms, since if you run out of memory you have bigger problems than your software can fix, unless you are writing the OS itself)
I have never needed a goto, and I've always found that using goto is a symptom of a bigger set of problems. Your case appears to be no exception.
Using "GOTO" will change the "logics" of a program and how you enterpret or how you would imagine it would work.
Avoiding GOTO-commands have always worked for me so guess when you think you might need it, all you maybe need is a re-design.
However, if we look at this on an Assmebly-level, jusing "jump" is like using GOTO and that's used all the time, BUT, in Assembly you can clear out, what you know you have on the stack and other registers before you pass on.
So, when using GOTO, i'd make sure the software would "appear" as the co-coders would enterpret, GOTO will have an "bad" effect on your software imho.
So this is more an explenation to why not to use GOTO and not a solution for a replacement, because that is VERY much up to how everything else is built.
I may have missed something: you jump to the label Exit if P is null, then test to see if it's not null (which it's not) to see if you need to delete it (which isn't necessary because it was never allocated in the first place).
The if/goto won't, and doesn't need to delete p. Replacing the goto with a return false would have the same effect (and then you could remove the Exit label).
The only places I know where goto's are useful are buried deep in nasty parsers (or lexical analyzers), and in faking out state machines (buried in a mass of CPP macros). In those two cases they've been used to make very twisted logic simpler, but that is very rare.
Functions (A calls A'), Try/Catches and setjmp/longjmps are all nicer ways of avoiding a difficult syntax problem.
Paul.
Ignoring the fact that new will never return NULL, take your code:
BOOL foo()
{
BOOL bRetVal = FALSE;
int *p=NULL;
p = new int;
if(p==NULL)
{
cout<<" OOM \n";
goto Exit;
}
// Lot of code...
Exit:
if(p)
{
delete p;
p= NULL;
}
return bRetVal;
}
and write it like this:
BOOL foo()
{
BOOL bRetVal = FALSE;
int *p = new int;
if (p!=NULL)
{
// Lot of code...
delete p;
}
else
{
cout<<" OOM \n";
}
return bRetVal;
}