Optimizer bug or programming error? - c++

First of all: I know that most optimization bugs are due to programming errors or relying on facts which may change depending on optimization settings (floating point values, multithreading issues, ...).
However I experienced a very hard to find bug and am somewhat unsure if there is any way to prevent these kind of errors from happening without turning the optimization off. Am I missing something? Could this really be an optimizer bug? Here's a simplified example:
struct Data {
int a;
int b;
double c;
};
struct Test {
void optimizeMe();
Data m_data;
};
void Test::optimizeMe() {
Data * pData; // Note that this pointer is not initialized!
bool first = true;
for (int i = 0; i < 3; ++i) {
if (first) {
first = false;
pData = &m_data;
pData->a = i * 10;
pData->b = i * pData->a;
pData->c = pData->b / 2;
} else {
pData->a = ++i;
} // end if
} // end for
};
int main(int argc, char *argv[]) {
Test test;
test.optimizeMe();
return 0;
}
The real program of course has a lot more to do than this. But it all boils down to the fact that instead of accessing m_data directly, a (previously unitialized) pointer is being used. As soon as I add enough statements to the if (first)-part, the optimizer seems to change the code to something along these lines:
if (first) {
first = false;
// pData-assignment has been removed!
m_data.a = i * 10;
m_data.b = i * m_data.a;
m_data.c = m_data.b / m_data.a;
} else {
pData->a = ++i; // This will crash - pData is not set yet.
} // end if
As you can see, it replaces the unnecessary pointer dereference with a direct write to the member struct. However it does not do this in the else-branch. It also removes the pData-assignment. Since the pointer is now still unitialized, the program will crash in the else-branch.
Of course there are various things which could be improved here, so you might blame it on the programmer:
Forget about the pointer and do what the optimizer does - use m_data directly.
Initialize pData to nullptr - that way the optimizer knows that the else-branch will fail if the pointer is never assigned. At least it seems to solve the problem in my test-environment.
Move the pointer assignment in front of the loop (effectively initializing pData with &m_data, which then could also be a reference instead of a pointer (for good measure). This makes sense because pData is needed in all cases so there is no reason to do this inside the loop.
The code is obviously smelly, to say the least, and I'm not trying to "blame" the optimizer for doing this. But I'm asking: What am I doing wrong? The program might be ugly, but it's valid code...
I should add that I'm using VS2012 with C++/CLI and v110_xp-Toolset. Optimization is set to /O2. Please also note that if you really want to reproduce the problem (that's not really the point of this question though) you need to play around with the complexity of the program. This is a very simplified example and the optimizer sometimes doesn't remove the pointer assignment. Hiding &m_data behind a function seems to "help".
EDIT:
Q: How do I know that the compiler is optimizing it to something like the example provided?
A: I'm not very good at reading assembler, I have looked at it however and have made 3 observations which make me believe that it's behaving this way:
As soon as optimization kicks in (adding more assignments usually does the trick) the pointer assignment has no associated assembler statement. It also hasn't been moved up to the declaration, so it's really left uninitialized it seems (at least to me).
In cases where the program crashes, the debugger skips the assignment statement. In cases where the program runs without problems, the debugger stops there.
If I watch the content of pData and the content of m_data while debugging, it clearly shows that all assignments in the if-branch have an effect on m_data and m_data receives the correct values. The pointer itself it still pointing to the same uninitialized value it had from the beginning. Therefore I have to assume that it is in fact not using the pointer to make the assignments at all.
Q: Does it have to do anything with i (Loop unrolling)?
A: No, the actual program actually uses do { ... } while() to loop over a SQL SELECT-resultset so the iteration count is completely runtime-specific and cannot be predetermined by the compiler.

It sure looks like an bug to me. It's fine for the optimizer to eliminate the unnecessary redirection, but it should not eliminate the assignment to pData.
Of course, you can work around the problem by assigning to pData before the loop (at least in this simple example). I gather that the problem in your actual code isn't as easily resolved.

I also vote for an optimizer bug if it is really reproducible in this example. To overrule the optimizer you could try to declare pData as volatile.

Related

How can I pass a C++ array of structs to a CUDA device?

I've spent 2 days trying to figure this out and getting nowhere. Say I had a struct that looks like this:
struct Thing {
bool is_solid;
double matrix[9];
}
I want to create an array of that struct called things and then process that array on the GPU. Something like:
Thing *things;
int num_of_things = 100;
cudaMallocManaged((void **)&things, num_of_things * sizeof(Thing));
// Something missing here? Malloc individual structs? Everything I try doesn't work.
things[10].is_solid = true; // Segfaults
Is it even best practice to do it this way rather than pass a single struct with arrays that are num_of_things large? It seem to me that can get pretty nasty especially when you have arrays already (like matrix, which would need to be 9 * num_of_things.
Any info would be much appreciated!
After some dialog in the comments, it seems that OP's posted code has no issues. I was able to successfully compile and run this test case built around that code, and so was OP:
$ cat t1005.cu
#include <iostream>
struct Thing {
bool is_solid;
double matrix[9];
};
int main(){
Thing *things;
int num_of_things = 100;
cudaError_t ret = cudaMallocManaged((void **)&things, num_of_things * sizeof(Thing));
if (ret != cudaSuccess) {
std::cout << cudaGetErrorString(ret) << std::endl;
return 1;}
else {
things[10].is_solid = true;
std::cout << "Success!" << std::endl;
return 0;}
}
$ nvcc -arch=sm_30 -o t1005 t1005.cu
$ ./t1005
Success!
$
Regarding this question:
Is it even best practice to do it this way rather than pass a single struct with arrays that are num_of_things large?
Yes, this is a sensible practice and is usable whether managed memory is being used or not. An array of more or less any structure that does not contain embedded pointers to dynamically allocated data elsewhere can be transferred to the GPU in a simple fashion using a single cudaMemcpy call (for example, if managed memory were not being used.)
To address the question about the 3rd (flags) parameter to cudaMallocManaged:
If it is specified, it is not correct to pass zero (although OP's posted code gives no evidence of that.) You should use one of the documented choices.
If it is not specified, this is still valid, and a default argument of cudaMemAttachGlobal is provided. This can be confirmed by reviewing the cuda_runtime.h file or else simply compiling/running the test code above. This particular point appears to be an oversight in the documentation, and I've filed an internal issue at NVIDIA to take a look at that. So it's possible the documentation may change in the future with respect to this.
Finally, proper cuda error checking is always in order any time you are having trouble with a CUDA code, and the use of such may shed some light on any errors that are made. The seg fault that the OP reported in code comments was almost certainly due to the cudaMallocManaged call failing (perhaps because a zero parameter was supplied incorrectly) and as a result the pointer in question (things) had no actual allocation. Subsequent usage of that pointer would lead to a seg fault. My test code demonstrates how to avoid that seg fault, even if the cudaMallocManaged call fails for some reason, and the key is proper error checking.

Why is foo->bar->foobar considered bad style? And how to avoid without adding code?

Our C++ professor mentioned that using the result of operator-> as input into another operator-> is considered bad style.
So instead of writing:
return edge->terminal->outgoing_edges[0];
He would prefer:
Node* terminal = edge->terminal;
return terminal->outgoing_edges[0];
Why is this considered bad style?
How could I restructure my program to avoid the 'bad style' but also avoid the extra line of code that is created as per the above suggestion?
There's a number of reasons.
The Law of Demeter gives a structural reason (note that your C++ professors code still violates this though!). In your example, edge has to know about terminal and outgoing_edges. That makes it tightly coupled.
As an alternative
class Edge {
private:
Node* terminal;
public:
Edges* outgoing_edges() {
return terminal->outgoing_edges;
}
}
Now you can change the implementation of outgoing_edges in one place without changing everywhere. To be honest, I don't really buy this in the case of a data structure like a graph (it is tightly coupled, edges and nodes can't escape each other). This'd be over-abstraction in my book.
There's the null dereference problem too, in the expression a->b->c, what if b is null?
You should ask your professor as to why he considers it bad style. I don't. I would however consider his omission of the const in the declaration of terminal to be bad style.
For a single snippet like that, it's probably not bad style. However consider this:
void somefunc(Edge *edge)
{
if (edge->terminal->outgoing_edges.size() > 5)
{
edge->terminal->outgoing_edges.rezize(10);
edge->terminal->counter = 5;
}
++edge->terminal->another_value;
}
This is starting to get unwieldy - it is difficult to read, it is difficult to write (I made approximately 10 typos when typing that). And it requires a lot of evaluation of the operator-> on those 2 classes. OK if the operator is trivial, but if the operator does anything exciting, it's going to end up doing a lot of work.
So there's 2 possible answers:
Maintanability
Efficiency
And a single snippet like that, you can't avoid the extra line. In something like the above, it'd have resulted in less typing.
Well, I can't be sure about what your professor did mean, bu I have some thoughts.
First of all, code like this can be "not-so-obvious":
someObjectPointer->foo()->bar()->foobar()
You can't realy say what is the result of foo(), and thus you can't really say on what is the bar() being called. Is that the same object? Or maybe its a temp object that is being created? Or maybe it's something else?
Another thing is if you have to repeat your code. Consider an example like this:
complexObject.somePart.someSubObject.sections[i].doSomething();
complexObject.somePart.someSubObject.sections[i].doSomethingElse();
complexObject.somePart.someSubObject.sections[i].doSomethingAgain();
complexObject.somePart.someSubObject.sections[i].andAgain();
Compare it with such:
section * sectionPtr = complexObject.somePart.someSubObject.sections[i];
sectionPtr->doSomething();
sectionPtr->doSomethingElse();
sectionPtr->doSomethingAgain();
sectionPtr->andAgain();
You can see how the second example is not only shorter, but easier to read.
Sometimes the chain of functions is easier, because you don't need to bother with what they return:
resultClass res = foo()->bar()->foobar()
vs
classA * a = foo();
classB * b = a->bar();
classC * c = b->foobar();
Another thing is debugging. It is very hard to debug long chains of function calls, because you simply can't understand which of them causes the error. In this case, breaking the chain helps a lot, not to mention you can do some extra checks too, like
classA * a = foo();
classB * b;
classC * c;
if(a)
b = a->bar();
if(b)
c = b->foobar();
So, to summarize, you can't say that it is a "bad" or "good" style in general. You have to consider your circumstances first.
The two code snippets you have shown is (of course) semantically identical. When it comes to matters of style, suggest you simply follow what your professor wants.
Slightly better than your second code snippet would be:
Node* terminal = (edge ? edge->terminal : NULL);
return (terminal ? terminal->outgoing_edges[0] : NULL);
I am assuming that outgoing_edges is an array; if not, you have to check for that being NULL or empty as well.
Neither are necessarily better in their current form. However, if you were to add a check for null in your second example, then this would make sense.
if (edge != NULL)
{
Node* terminal = edge->terminal;
if (terminal != NULL) return terminal->outgoing_edges[0];
}
Though even that is still bad because who is to say that outgoing_edges is properly populated or allocated. A better bet would be to call a function called outgoingEdges() that does all that nice error checking for you and never causes you undefined behavior.

C/C++: goto into the for loop

I have a bit unusual situation - I want to use goto statement to jump into the loop, not to jump out from it.
There are strong reasons to do so - this code must be part of some function which makes some calculations after the first call, returns with request for new data and needs one more call to continue. Function pointers (obvious solution) can't be used because we need interoperability with code which does not support function pointers.
I want to know whether code below is safe, i.e. it will be correctly compiled by all standard-compliant C/C++ compilers (we need both C and C++).
function foo(int not_a_first_call, int *data_to_request, ...other parameters... )
{
if( not_a_first_call )
goto request_handler;
for(i=0; i<n; i++)
{
*data_to_request = i;
return;
request_handler:
...process data...
}
}
I've studied standards, but there isn't much information about such use case. I also wonder whether replacing for by equivalent while will be beneficial from the portability point of view.
Thanks in advance.
UPD: Thanks to all who've commented!
to all commenters :) yes, I understand that I can't jump over initializers of local variables and that I have to save/restore i on each call.
about strong reasons :) This code must implement reverse communication interface. Reverse communication is a coding pattern which tries to avoid using function pointers. Sometimes it have to be used because of legacy code which expects that you will use it.
Unfortunately, r-comm-interface can't be implemented in a nice way. You can't use function pointers and you can't easily split work into several functions.
Seems perfectly legal.
From a draft of the C99 standard http://std.dkuug.dk/JTC1/SC22/WG14/www/docs/n843.htm in the section on the goto statement:
[#3] EXAMPLE 1 It is sometimes convenient to jump into the
middle of a complicated set of statements. The following
outline presents one possible approach to a problem based on
these three assumptions:
1. The general initialization code accesses objects only
visible to the current function.
2. The general initialization code is too large to
warrant duplication.
3. The code to determine the next operation is at the
head of the loop. (To allow it to be reached by
continue statements, for example.)
/* ... */
goto first_time;
for (;;) {
// determine next operation
/* ... */
if (need to reinitialize) {
// reinitialize-only code
/* ... */
first_time:
// general initialization code
/* ... */
continue;
}
// handle other operations
/* ... */
}
Next, we look at the for loop statement:
[#1] Except for the behavior of a continue statement in the |
loop body, the statement
for ( clause-1 ; expr-2 ; expr-3 ) statement
and the sequence of statements
{
clause-1 ;
while ( expr-2 ) {
statement
expr-3 ;
}
}
Putting the two together with your problem tells you that you are jumping past
i=0;
into the middle of a while loop. You will execute
...process data...
and then
i++;
before flow of control jumps to the test in the while/for loop
i<n;
Yes, that's legal.
What you're doing is nowhere near as ugly as e.g. Duff's Device, which also is standard-compliant.
As #Alexandre says, don't use goto to skip over variable declarations with non-trivial constructors.
I'm sure you're not expecting local variables to be preserved across calls, since automatic variable lifetime is so fundamental. If you need some state to be preserved, functors (function objects) would be a good choice (in C++). C++0x lambda syntax makes them even easier to build. In C you'll have no choice but to store state into some state block passed in by pointer by the caller.
First, I need to say that you must reconsider doing this some other way. I've rarely seen someone using goto this days if not for error management.
But if you really want to stick with it, there are a few things you'll need to keep in mind:
Jumping from outside the loop to the middle won't make your code loop. (check the comments below for more info)
Be careful and don't use variables that are set before the label, for instance, referring to *data_to_request. This includes iwhich is set on the for statement and is not initialized when you jump to the label.
Personally, I think in this case I would rather duplicate the code for ...process data... then use goto. And if you pay close attention, you'll notice the return statement inside your for loop, meaning that the code of the label will never get executed unless there's a goto in the code to jump to it.
function foo(int not_a_first_call, int *data_to_request, ...other parameters... )
{
int i = 0;
if( not_a_first_call )
{
...process data...
*data_to_request = i;
return;
}
for (i=0; i<n; i++)
{
*data_to_request = i;
return;
}
}
No, you can't do this. I don't know what this will do exactly, but I do know that as soon as you return, your call stack is unwound and the variable i doesn't exist anymore.
I suggest refactoring. It looks like you're pretty much trying to build an iterator function similar to yield return in C#. Perhaps you could actually write a C++ iterator to do this?
It seems to me that you didn't declare i. From the point of declaration completely depends whether or not this is legal what you are doing, but see below for the initialization
In C you may declare it before the loop or as loop variable. But if it is declared as loop variable its value will not be initialized when you use it, so this is undefined behavior. And if you declare it before the for the assignment of 0 to it will not be performed.
In C++ you can't jump across the constructor of the variable, so you must declare it before the goto.
In both languages you have a more important problem, this is if the value of i is well defined, and if it is initialized if that value makes sense.
Really if there is any way to avoid this, don't do it. Or if this is really, really, performance critical check the assembler if it really does what you want.
If I understand correctly, you're trying to do something on the order of:
The first time foo is called, it needs to request some data from somewhere else, so it sets up that request and immediately returns;
On each subsequent call to foo, it processes the data from the previous request and sets up a new request;
This continues until foo has processed all the data.
I don't understand why you need the for loop at all in this case; you're only iterating through the loop once per call (if I understand the use case here). Unless i has been declared static, you lose its value each time through.
Why not define a type to maintain all the state (such as the current value of i) between function calls, and then define an interface around it to set/query whatever parameters you need:
typedef ... FooState;
void foo(FooState *state, ...)
{
if (FirstCall(state))
{
SetRequest(state, 1);
}
else if (!Done(state))
{
// process data;
SetRequest(state, GetRequest(state) + 1);
}
}
The initialisation part of the for loop will not occur, which makes it somewhat redundant. You need to initialise i before the goto.
int i = 0 ;
if( not_a_first_call )
goto request_handler;
for( ; i<n; i++)
{
*data_to_request = i;
return;
request_handler:
...process data...
}
However, this is really not a good idea!
The code is flawed in any case, the return statment circumvents the loop. As it stands it is equivalent to:
int i = 0 ;
if( not_a_first_call )
\\...process_data...
i++ ;
if( i < n )
{
*data_to_request = i;
}
In the end, if you think you need to do this then your design is flawed, and from the fragment posted your logic also.

Using "assert" with pointers in C++

When do we need to use "assert" for pointers in C++, and when they are used, how are they most commonly implemented?
Generally you would use an assert to check a condition that, if false, would indicate a bug in your application. So if a NULL pointer shouldn't ever be encountered at some point in the application, unless there's a bug, then assert it. If it might be encountered due to some invalid input then you need to do proper error handling.
You don't need to use assert on pointers at all. The idea is to ensure you don't crash when dereferencing your pointers when they're null.
You can do this with assert but it's not a very professional way to handle errors like this since it invariably terminates the program - not a good idea if the user hasn't, for example, saved their last three hours worth of data entry.
What you should do with pointers is to check them for null-ness and fail gracefully. In other words, have your function return an error of some sort or do nothing (not everyone will agree with this approach but it's perfectly acceptable if it's documented).
The assert stuff is meant, in my opinion, for catching problems during development which is why you'll find assert does nothing in release builds under some compilers. It is not a substitute for defensive programming.
As to how to do it:
#include <assert.h>
void doSomethingWithPointer (int *p) {
assert (p != 0);
cout << *p << endl;
}
but this would be better done as:
void doSomethingWithPointer (int *p) {
if (p != 0)
cout << *p << endl;
}
In other words, even if your "contract" (API) states that you're not allowed to receive null pointers, you should still handle them gracefully. An old quote: be conservative in what you give, liberal in what you accept (paraphrased).
ASSERT statements are great as "enforced documentation" - that is, they tell the reader something about the code ("This should never happen") and then enforces it by letting you know if they don't hold true.
If it's something that could happen (invalid input, memory not able to be allocated), that's not a time to use ASSERT. Asserts are only for things that can not possibly happen if everyone is obeying pre-conditions and such.
You can do it thusly:
ASSERT(pMyPointer);
From experience if you assert on null conditions that should never happen under normal conditions you program is in a really bad state. Recovering from such null condition will more likely than not mask the original problem.
Unless you code with exception guarantee in mind (linky) I say let it crash, then you know you have a problem.
I would use an ASSERT where a null pointer wouldn't immediately cause a crash but might lead to somethign wrong later that's hard to spot.
eg:
ASSERT(p);
strcpy(p, "hello");
Is a little unnecessary, it simply replaces a fatal exception with a fatal assert!
But in more complex code, particulalrly things like smart pointers, it might be useful to know check if the pointer is what you thing it is.
Remember ASSERTs only run in debug builds, they dissapear in the release.
In C, there also assert function..
in debug mode, if assert(x), x condition is false, there will pop up an alert...
But remember it works only in debug mode...
in release mode, all assert functions are all skipped
Assertions are used to to define how the program should function. That being said, the most common use of Assert()s when dealing with pointers is going to either be that they are valid (non-NULL and point towards valid memory) or that their internal state is valid if they point to an object/class instance, for example.
Assertions are not for replacing or acting as error condition code, but instead to enforce rules that you are placing on the functioning of your code, such as what conditions should be at given points in time.
For example,
function int f(int x, int * pY)
{
// These are entrance conditions expected in the function. It would be
// a BUG if this happened at all.
Assert(x >= 0);
Assert(pY != nullptr);
Assert(*pY >= 0, "*pY should never be less than zero");
// ...Do a bunch of computations with x and pY and return the result as z...
int z = x * 2 / (x + 1) + pow(*pY, x); // Maybe z should be always positive
// after these calculations:
Assert(x >= 0, "X should always be positive after calculations);
// Maybe *pY should always be non-zero after calculations
Assert(*pY != 0, "y should never be zero after computation");
Assert(z > 0):
return z;
}
Many users of Asserts choose to apply Assertions to internal state validation once they become familiar with them. We call these Invariants() which are methods on a class that assert many things about the internals of the object that should always hold true.
For example:
class A
{
public:
A(wchar_t * wszName)
{
_cch = 0;
_wszName = wszName;
}
// Invariant method to be called at times to verify that the
// internal state is consistent. This means here that the
// internal variable tracking the length of the string is
// matching the actual length of the string.
void Invariant()
{
Assert(pwszName != nullptr);
Assert(_cch == wcslen(pwszName));
}
void ProcessABunchOfThings()
{
...
}
protected:
int _cch;
wchar_t * pwszName;
}
// Call to validate internal state of object is consistent/ok
A a(L"Test Object");
a.Invariant();
a.ProcessABunchOfThings();
a.Invariant();
The important thing to remember is that this is to make sure that when bugs do happen that mean the program is not working as you would expect, then the effect of the bug happens as close to where it happened in the code as possible in order to make debugging easier. I have used Asserts extensively in my own code and while at Microsoft and I swear by them since they have saved me so much time in debugging and even knowing the defect is there.

Inadvertent use of = instead of ==

This question's answers are a community effort. Edit existing answers to improve this post. It is not currently accepting new answers or interactions.
It seems that
if (x=y) { .... }
instead of
if (x==y) { ... }
is a root of many evils.
Why don't all compilers mark it as error instead of a configurable warning?
I'm interested in finding out cases where the construct if (x=y) is useful.
One useful construct is for example:
char *pBuffer;
if (pBuffer = malloc(100))
{
// Continue to work here
}
As mentioned before, and downvoted several times now, I might add this is not specially good style, but I have seen it often enough to say it's useful. I've also seen this with new, but it makes more pain in my chest.
Another example, and less controversial, might be:
while (pointer = getNextElement(context))
{
// Go for it. Use the pointer to the new segment of data.
}
which implies that the function getNextElement() returns NULL when there is no next element so that the loop is exited.
Most of the time, compilers try very hard to remain backward compatible.
Changing their behavior in this matter to throw errors will break existing legitimate code, and even starting to throw warnings about it will cause problems with automatic systems that keep track of code by automatically compiling it and checking for errors and warnings.
This is an evil we're pretty much stuck with at the moment, but there are ways to circumvent and reduce the dangers of it.
Example:
void *ptr = calloc(1, sizeof(array));
if (NULL = ptr) {
// Some error
}
This causes a compilation error.
Simple answer: An assignment operation, such as x=y, has a value, which is the same as the newly assigned value in x. You can use this directly in a comparison, so instead of
x = y; if (x) ...
you can write
if (x = y) ...
It is less code to write (and read), which is sometimes a good thing, but nowadays most people agree that it should be written in some other way to increase readability. For example, like this:
if ((x = y) != 0) ...
Here is a realistic example. Assume you want to allocate some memory with malloc, and see if it worked. It can be written step by step like this:
p = malloc(4711); if (p != NULL) printf("Ok!");
The comparison to NULL is redundant, so you can rewrite it like this:
p = malloc(4711); if (p) printf("Ok!");
But since the assignment operation has a value, which can be used, you could put the entire assignment in the if condition:
if (p = malloc(4711)) printf("Ok!");
This does the same thing, but it is more concise.
Because it's not illegal (in C or C++ anyway) and sometimes useful...
if ( (x = read(blah)) > 0)
{
// now you know how many bits/bytes/whatever were read
// and can use that info. Esp. if you know, say 30 bytes
// are coming but only got 10
}
Most compilers kick up a real stink if you don't put parenthesis around the assignment anyway, which I like.
About the valid uses of if(i = 0)
The problem is that you're taking the problem upside down. The "if" notation is not about comparing two values like in some other languages.
The C/C++ "if" instruction waits for any expression that will evaluate to either a boolean, or a null/non-null value. This expression can include two values comparison, and/or can be much more complex.
For example, you can have:
if(i >> 3)
{
std::cout << "i is less than 8" << std::endl
}
Which proves that, in C/C++, the if expression is not limited to == and =. Anything will do, as long as it can be evaluated as true or false (C++), or zero non-zero (C/C++).
Another C++ valid use:
if(MyObject * pObject = dynamic_cast<MyInterface *>(pInterface))
{
pObject->doSomething();
}
And these are simple uses of the if expression (note that this can be used, too, in the for loop declaration line). More complex uses do exist.
About advanced uses of if(i = 0) in C++ (Quoted from myself)
After discovering a duplicate of this question at In which case is if(a=b) a good idea?, I decided to complete this answer with an additional bonus, that is, variable injection into a scope, which is possible in C++, because if will evaluate its expression, including a variable declaration, instead of limiting itself to compare two operands like it is done in other languages:
So, quoting from myself:
Another use would be to use what is called C++ variable injection. In Java, there is this cool keyword:
synchronized(p)
{
// Now, the Java code is synchronized using p as a mutex
}
In C++, you can do it, too. I don't have the exact code in mind (nor the exact Dr. Dobb's Journal's article where I discovered it), but this simple define should be enough for demonstration purposes:
#define synchronized(lock) \
if (auto_lock lock_##__LINE__(lock))
synchronized(p)
{
// Now, the C++ code is synchronized using p as a mutex
}
This is the same way, mixing injection with an if and for declaration. You can declare a primitive foreach macro (if you want an industrial-strength foreach, use Boost's).
See the following articles for a less naive, more complete and more robust implementation:
FOR_EACH and LOCK
Exception Safety Analysis
Concurrent Access Control & C++
How many errors of this kind really happens?
Rarely. In fact, I have yet to remember one, and I have been a professional for the past 8 years.
I guess it happened, but then, in 8 years, I did produce a sizeable quantity of bugs. It's just that this kind of bugs did not happen enough to have me remember them in frustration.
In C, you'll have more bugs because of buffer overruns, like:
void doSomething(char * p)
{
strcpy(p, "Hello, World! How are you \?\n");
}
void doSomethingElse()
{
char buffer[16];
doSomething(buffer);
}
In fact, Microsoft was burned so hard because of that they added a warning in Visual C++ 2008 deprecating strcpy!
How can you avoid most errors?
The very first "protection" against this error is to "turn around" the expression: As you can't assign a value to a constant, this:
if(0 = p) // ERROR: It should have been if(0 == p). IT WON'T COMPILE!
It won't compile.
But I find this quite a poor solution, because it tries to hide behind a style what should be a general programming practice, that is: Any variable that is not supposed to change should be constant.
For example, instead of:
void doSomething(char * p)
{
if(p == NULL) // POSSIBLE TYPO ERROR
return;
size_t length = strlen(p);
if(length == 0) // POSSIBLE TYPO ERROR
printf("\"%s\" length is %i\n", p, length);
else
printf("the string is empty\n");
}
Trying to "const" as many variables as possible will make you avoid most typo errors, including those not inside "if" expressions:
void doSomething(const char * const p) // CONST ADDED HERE
{
if(p == NULL) // NO TYPO POSSIBLE
return;
const size_t length = strlen(p); // CONST ADDED HERE
if(length == 0) // NO TYPO POSSIBLE
printf("\"%s\" length is %i\n", p, length);
else
printf("the string is empty\n");
}
Of course, it is not always possible (as some variables do need to change), but I found than most of the variables I use are constants (I keep initializing them once, and then, only reading them).
Conclusion
Usually, I see code using the if(0 == p) notation, but without the const-notation.
To me, it's like having a trash can for recyclables, and another for non-recyclable, and then in the end, throw them together in the same container.
So, do not parrot an easy style habit hoping it will make your code a lot better. It won't. Use the language constructs as much as possible, which means, in this case, using both the if(0 == p) notation when available, and using of the const keyword as much as possible.
The 'if(0 = x)' idiom is next to useless because it doesn't help when both sides are variables ('if(x = y)') and most (all?) of the time you should be using constant variables rather than magic numbers.
Two other reasons I never use this idiom, IMHO it makes code less readable and to be honest I find the single '='to be the root of very little evil. If you test your code thouroughly (which we all do, obviously) this sort of syntax error turns up very quickly.
Standard C idiom for iterating:
list_elem* curr;
while ( (curr = next_item(list)) != null ) {
/* ... */
}
Many compilers will detect this and warn you, but only if you set the warning level high enough.
For example:
~> gcc -c -Wall foo.c
foo.c: In function ‘foo’:
foo.c:5: warning: suggest parentheses around assignment used as truth value
Is this really such a common error? I learned about it when I learned C myself, and as a teacher I have occasionally warned my students and told them that it is a common error, but I have rarely seen it in real code, even from beginners. Certainly not more often than other operator mistakes, such as for example writing "&&" instead of "||".
So the reason that compilers don't mark it as an error (except for it being perfectly valid code) is perhaps that it isn't the root of very many evils.
The assignment as conditional is legal C and C++, and any compiler that doesn't permit it isn't a real C or C++ compiler. I would hope that any modern language not designed to be explicitly compatible with C (as C++ was) would consider it an error.
There are cases where this allows concise expressions, such as the idiomatic while (*dest++ = *src++); to copy a string in C, but overall it's not very useful, and I consider it a mistake in language design. It is, in my experience, easy to make this mistake, and hard to spot when the compiler doesn't issue a warning.
I think the C and C++ language designers noticed there is no real use in forbidding it because
Compilers can warn about it if they want anyway
Disallowing it would add special cases to the language, and would remove a possible feature.
There isn't complexity involved in allowing it. C++ just says that an expression implicitly convertible to bool is required. In C, there are useful cases detailed by other answers. In C++, they go one step further and allowed this one in addition:
if(type * t = get_pointer()) {
// ....
}
Which actually limits the scope of t to only the if and its bodies.
It depends on the language. Java flags it as an error as only Boolean expressions can be used inside the if parenthesis (and unless the two variables are Boolean, in which case the assignment is also a Boolean).
In C, it is a quite common idiom for testing pointers returned by malloc or if after a fork we are in the parent or child process:
if ( x = (X*) malloc( sizeof(X) ) {
// 'malloc' worked, pointer != 0
if ( pid = fork() ) {
// Parent process as pid != 0
C/C++ compilers will warn with a high enough warning level if you ask for it, but it cannot be considered an error as the language allows it. Unless, then again, you ask the compiler to treat warnings as errors.
Whenever comparing with constants, some authors suggest using the test constant == variable so that the compiler will detect if the user forgets the second equality sign.
if ( 0 == variable ) {
// The compiler will complaint if you mistakenly
// write =, as you cannot assign to a constant
Anyway, you should try to compile with the highest possible warning settings.
Try viewing
if( life_is_good() )
enjoy_yourself();
as
if( tmp = life_is_good() )
enjoy_yourself();
Part of it has to do with personal style and habits. I am agnostic to reading either if (kConst == x) or if (x == kConst). I don't use the constant on the left because historically I don't make that error and I write code as I would say it or would like to read it. I see this as a personal decision as part of a personal responsibility to being a self-aware, improving engineer. For example, I started analyzing the types of bugs that I was creating and started to re-engineer my habits so as not to make them - similar to constant on the left, just with other things.
That said, compiler warnings, historically, are pretty crappy and even though this problem has been well known for years, I didn't see it in a production compiler until the late 80's. I also found that working on projects that were portable helped clean up my C a great deal, as different compilers and different tastes (ie, warnings) and different subtle semantic differences.
I, personally, consider this the most useful example.
Say that you have a function read() that returns the number of bytes read, and you need to use this in a loop. It's a lot simpler to use
while((count = read(foo)) > 0) {
//Do stuff
}
than to try and get the assignment out of the loop head, which would result in things like
while(1) {
count = read(foo);
if(!(count > 0))
break;
//...
}
or
count = read(foo);
while(count > 0) {
//...
count = read(foo);
}
The first construct feels awkward, and the second repeats code in an unpleasant way.
Unless, of course, I've missed some brilliant idiom for this...
There are a lot of great uses of the assignment operator in a conditional statement, and it'd be a royal pain in the ass to see warnings about each one all the time. What would be nice would be a function in your IDE that let you highlight all the places where assignment has been used instead of an equality check - or - after you write something like this:
if (x = y) {
then that line blinks a couple of times. Enough to let you know that you've done something not exactly standard, but not so much that it's annoying.
if ((k==1) || (k==2)) is a conditional
if ((k=1) || (k=2) ) is BOTH a conditional AND an assignment statement
Here's the explanation
Like most languages, C works inner-most to outermost in order by operator precedence.
First, it tries to set k to 1, and succeeds.
Result: k = 1 and Boolean = 'true'
Next: it sets k to 2, and succeeds.
Result: k = 2 and Boolean = 'true'
Next: it evaluates (true || true)
Result: k still = 2, and Boolean = true
Finally, it then resolves the conditional: If (true)
Result: k = 2 and the program takes the first branch.
In nearly 30 years of programming I have not seen a valid reason for using this construct, though if one exists it probably has to do with a need to deliberately obfuscate your code.
When one of our new people has a problem, this is one of the things I look for, right along with not sticking a terminator on a string, and copying a debug statement from one place to another and not changing the '%i to '%s' to match the new field they are dumping.
This is fairly common in our shop because we constantly switch between C and Oracle PL/SQL; if( k = 1) is the correct syntax in PL/SQL.
It is very common with "low level" loop constructs in C/C++, such as with copies:
void my_strcpy(char *dst, const char *src)
{
while((*dst++ = *src++) != '\0') { // Note the use of extra parentheses, and the explicit compare.
/* DO NOTHING */
}
}
Of course, assignments are very common with for loops:
int i;
for(i = 0; i < 42; ++i) {
printf("%d\n", i);
}
I do believe it is easier to read assignments when they are outside of if statements:
char *newstring = malloc(strlen(src) * sizeof(char));
if(newstring == NULL) {
fprintf(stderr, "Out of memory, d00d! Bailing!\n");
exit(2);
}
// Versus:
if((newstring = malloc(strlen(src) * sizeof(char))) == NULL) // ew...
Make sure the assignment is obvious, thuogh (as with the first two examples). Don't hide it.
As for accidental uses ... that doesn't happen to me much. A common safeguard is to put your variable (lvalues) on the right hand side of the comparison, but that doesn't work well with things like:
if(*src == *dst)
because both oprands to == are lvalues!
As for compilers ... who can blame 'em? Writing compilers is difficult, and you should be writing perfect programs for the compiler anyway (remember GIGO?). Some compilers (the most well-known for sure) provide built-in lint-style checking, but that certainly isn't required. Some browsers don't validate every byte of HTML and Javascript it's thrown, so why would compilers?
There are several tactics to help spot this .. one is ugly, the other is typically a macro. It really depends on how you read your spoken language (left to right, right to left).
For instance:
if ((fp = fopen("foo.txt", "r") == NULL))
Vs:
if (NULL == (fp = fopen(...)))
Sometimes it can be easier to read/write (first) what your testing for, which makes it easier to spot an assignment vs a test. Then bring in most comp.lang.c folks that hate this style with a passion.
So, we bring in assert():
#include <assert.h>
...
fp = fopen("foo.txt", "r");
assert(fp != NULL);
...
when your at the midst, or end of a convoluted set of conditionals, assert() is your friend. In this case, if FP == NULL, an abort() is raised and the line/file of the offending code is conveyed.
So if you oops:
if (i = foo)
insted of
if (i == foo)
followed by
assert (i > foo + 1)
... you'll quickly spot such mistakes.
Hope this helps :)
In short, reversing arguments sometimes helps when debugging .. assert() is your long life friend and can be turned off in compiler flags in production releases.
As pointed out in other answers, there are cases where using assignment within a condition offers a brief-but-readable piece of code that does what you want. Also, a lot of up-to-date compilers will warn you if they see an assignment where they expect a condition. (If you're a fan of the zero-warnings approach to development, you'll have seen these.)
One habit I've developed that keeps me from getting bitten by this (at least in C-ish languages) is that if one of the two values I'm comparing is a constant (or otherwise not a legal lvalue), I put it on the left-hand side of the comparator: if (5 == x) { whatever(); } Then, if I should accidentally type if (5 = x), the code won't compile.
You asked why it was useful, but keep questioning examples people are providing. It's useful because it's concise.
Yes, all the examples which use it can be rewritten - as longer pieces of code.
I have only had this typo once in my 15 years of development. I would not say it is on the top of my list of things to look out for. I also avoid that construct anyway.
Note also that some compilers (the one I use) issue a warning on that code. Warnings can be treated as errors for any compiler worth its salt. They can also be ignored.
Placing the constant on the left side of a comparison is defensive programming. Sure you would never make the silly mistake of forgetting that extra '=', but who knows about the other guy.
The D programming language does flag this as an error. To avoid the problem with wanting to use the value later, it allows declarations sort of like C++ allows with for loops.
if(int i = some_fn())
{
another_fn(i);
}
The compiler won't flag it as an error because it is valid C/C++. But what you can do (at least with Visual C++) is turn up the warning level so that it flags it as a warning and then tell the compiler to treat warnings as errors. This is a good practice anyway so that developers don't ignore warnings.
If you had actually meant = instead of == then you need to be more explicit about it. E.g.,
if ((x = y) != 0)
Theoretically, you're supposed to be able to do this:
if ((x = y))
to override the warning, but that doesn't seem to always work.
In practice I don't do it, but a good tip is to do:
if ( true == $x )
In the case that you leave out an equals, assigning $x to true will obviously return an error.
RegEx sample
RegEx r;
if(((r = new RegEx("\w*)).IsMatch()) {
// ... do something here
}
else if((r = new RegEx("\d*")).IsMatch()) {
// ... do something here
}
Assign a value test
int i = 0;
if((i = 1) == 1) {
// 1 is equal to i that was assigned to a int value 1
}
else {
// ?
}
That's why it's better to write:
0 == CurrentItem
Instead of:
CurrentItem == 0
so that the compiler warns you if you type = instead of ==.