I declared NSBlockOperation with an object declared inside that operation. My app constantly is crashing because of memory issue. Appreciate any hint with a great explanation on this spent several hours still no success.
runtime: Memory Issues - (5 leaked types): 1 instance of NSExactBlockVariable leaked
- (EMUserInfoOperation*)loadingLocalModelOperationWithColor:(EMOutfitColor)outfitColor gender:(EMGender)gender {
__block EMUserInfoOperation* operation = [EMUserInfoOperation blockOperationWithBlock:^{
NSURL* remoteURL = [NSURL URLWithString:self.settings[kEMRemoteUrlKey]];
EMOutfitModel* model = nil;
if (remoteURL == nil) {
model = [[EMDomainDataLoader sharedLoader] loadEmbededOutfitNamed:self.name gender:gender];
} else {
model = [[EMDomainDataLoader sharedLoader] loadCachedOutfitNamed:self.name withVersion:self.version gender:gender];
}
[model syncApplyTextureFromPath:[self texturePathForColor:outfitColor] textureSampler:EMTextureSamplerColor];
NSString *alphaPath = [self texturePathForAlpha];
if(alphaPath.length > 0) {
[model syncApplyTextureFromPath:alphaPath textureSampler:EMTextureSamplerAlpha];
}
operation.userInfo = model;
}];
return operation;
}
I am guessing that your EMUserInfoOperation object has a strong reference to the block that the operation is created with. And this block also has a strong reference to the EMUserInfoOperation object because it captures the operation variable. So you have a retain cycle.
You can have the block only weakly reference the EMUserInfoOperation object by doing the following:
EMUserInfoOperation* operation;
__block __weak typeof(operation) weakOperation;
weakOperation = operation = [EMUserInfoOperation blockOperationWithBlock:^{
typeof(operation) strongOperation = weakOperation;
if (strongOperation) {
// ...
strongOperation.userInfo = model;
}
}];
return operation;
Related
In my ARC iOS app I am running a for loop that ends up with a large memory allocation overhead. I want to somehow end my for loop with minimal/no extra memory allocated. In this instance I am using the SSKeychain library which lets me fetch things from a keychain. I usually just use autorelease pools and get my memory removed properly but here I don't know what is wrong because I end up with 70 mb + of memory allocated at the end of the loop. I have been told that I should start/end a run loop to properly deal with this. Thoughts?
for (int i = 0; i < 10000; ++i) {
#autoreleasepool {
NSError * error2 = nil;
SSKeychainQuery* query2 = [[SSKeychainQuery alloc] init];
query2.service = #"Eko";
query2.account = #"loginPINForAccountID-2";
query2.password = nil;
[query2 fetch:&error2];
}
}
What are you using to measure memory usage?
Results of a very simple test...
Running in the simulator, measure only resident memory before and after.
Without autoreleasepool...
Started with 27254784, ended with 30212096, used 2957312
With autoreleasepool...
Started with 27316224, ended with 27443200, used 126976
Obviously, the autoreleasepool is preventing memory from growing too bad, and I don't see anything close to 70MB being used under any circumstance.
You should run instruments and get some good readings on the behavior.
Here is the code I hacked and ran...
The memchecker
static NSUInteger available_memory(void) {
NSUInteger result = 0;
struct task_basic_info info;
mach_msg_type_number_t size = sizeof(info);
if (task_info(mach_task_self(), TASK_BASIC_INFO, (task_info_t)&info, &size) == KERN_SUCCESS) {
result = info.resident_size;
}
return result;
}
And the code...
#define USE_AUTORELEASE_POOL 1
- (BOOL)application:(UIApplication *)application didFinishLaunchingWithOptions:(NSDictionary *)launchOptions {
dispatch_async(dispatch_get_main_queue(), ^{
NSUInteger beginMemory = available_memory();
for (int i = 0; i < 10000; ++i) {
#ifdef USE_AUTORELEASE_POOL
#autoreleasepool
#endif
{
NSError * error2 = nil;
SSKeychainQuery* query2 = [[SSKeychainQuery alloc] init];
query2.service = #"Eko";
query2.account = #"loginPINForAccountID-2";
query2.password = nil;
[query2 fetch:&error2];
}
}
NSUInteger endMemory = available_memory();
NSLog(#"Started with %u, ended with %u, used %u", beginMemory, endMemory, endMemory-beginMemory);
});
return YES;
}
I started writing a simple JSON RPC TCP library in Objective C.
I have a method that invokes a RPC Method:
- (void)invokeMethod:(NSString *)method
withParameters:(id)parameters
requestId:(id)requestId
success:(void (^)(id responseObject))success
failure:(void (^)(NSError *error))failure
{
NSAssert(NSClassFromString(#"NSJSONSerialization"), #"NSJSONSerialization not found!");
NSDictionary *requestObject = #{#"jsonrpc": #"2.0",
#"method": method,
#"params": parameters,
#"id": requestId};
NSError *error = nil;
NSData *jsondData = [NSJSONSerialization dataWithJSONObject:requestObject options:0 error:&error];
if (error){
return failure(error);
}
[self->callbacks setObject:#{#"success": success ? [success copy] : [NSNull null],
#"failure": failure ? [failure copy] : [NSNull null]}
forKey:requestId];
NSString *str = [[NSString alloc] initWithData:jsondData encoding:NSUTF8StringEncoding];
NSLog(#"Sending: %#", str);
[self.socket writeData:jsondData withTimeout:-1 tag:1];
}
The class basically represents a TCP connection, when calling the above method, the JSON data is sent with an id over TCP to the server which either returns a success or a failure:
- (void) socket:(GCDAsyncSocket *)sender didReadData:(NSData *)data withTag:(long)tag
{
NSError *error = nil;
[self.socket readDataWithTimeout:-1 tag:2];
// … rpc response parsing code here, removed for simplicity …
// detect if error or success
NSDictionary *cbs = [self->callbacks objectForKey:JSONRPCObjectId];
void(^success)(id resultObject) = [cbs objectForKey:#"success"];
success ? success(JSONRPCObjectResult) : nil;
return;
}
Now, I am unsure how to keep track of the success and failure blocks, currently I am storing them in an NSMutableDict, using the requestId as key. Is it fine to do this or is there a better approach that I should use?
Blocks in objective-c are objects and you can treat the same way as other object, so storing them in NSDictionarys, NSArrays etc is perfectly fine. The only catch is that blocks when initially created exist in the same memory scope as local variable do and so they are no longer valid when the method that the block is defined in returns, just like all other local variables so you have to copy them first, just copy them and put the copy in the collection. There is a block copy function but you can just send them a copy message [myBlock copy];
Quick answer, seeing as you don't have anything workable yet...
This is more than you asked for; so, you'll probably have to pair it down to meet your specific need. Basically, it stores as many blocks as you specify at contiguous memory addresses. Paste this into a header file or somewhere global to the method from which you will call these:
typedef const typeof(id(^)(void)) retained_object;
static id (^retainable_object)(id(^)(void)) = ^ id (id(^object)(void)) {
return ^{
return object();
};
};
typeof (retained_object) *(^(^retain_object)(id (^__strong)(void)))(void) = ^ (id(^retainable_object)(void)) {
typeof(retained_object) * object_address;
object_address = &retainable_object;
typeof(retained_object) * persistent_object = (typeof(retained_object) *)CFBridgingRetain(retainable_object);
return ^ typeof(retained_object) * {
return persistent_object;
};
};
static void (^(^iterator)(const unsigned long))(id(^)(void)) = ^ (const unsigned long object_count) {
id const * retained_objects_ref[object_count];
return ^ (id const * retained_objects_t[]) {
return ^ (id(^object)(void)) {
object();
int index = 0UL;
int * index_t = &index;
for (; (*index_t) < object_count; ((*index_t) = (*index_t) + 1UL)) printf("retained_object: %p\n", (*((id * const)retained_objects_t + (object_count - index)) = retain_object(retainable_object(object()))));
};
}(retained_objects_ref);
};
From some method, add:
iterator(1000)(^ id { return (^{ printf("stored block\n"); }); });
This should store 1,000 blocks at as many unique memory addresses.
My map object has a set of coordinates. It doesn't always have the same number of coordinates.
In java I'd just declare the object as Double[] xpoints and would set it's size when instantiating a map like this: xpoints = new double[npoints];
How can I do this with objective-c?
I tried doing this: #property(nonatomic) double * xpoints; but somehow all of it's values turn to 0 when I print it with NSLog.
Map's init:
-(id)initWithXpoints:(double[]) xpointss Ypoints:(double[]) ypointss Npoints:(int)npointss
{
self = [super init];
if (self)
{
self.xpoints = xpointss;
self.ypoints = ypointss;
self.npoints = npointss;
}
return self;
}
Something weird happens though. The values are changed to zero when I print xpoints[0] from the object that created the map. The first time I print it it works. The second time it just prints zero.
I think it happens because xpointss sent to init is removed from the memory. How can I "instantiate" the xpoints property if it's a pointer?
Is there a better way to do this?
added: I tried creating a temporary xpoints like this:
double tempxpoints[npointss];
double tempypoints[npointss];
for (int i = 0; i < npointss; i++)
{
tempxpoints[i] = xpointss[i];
tempypoints[i] = ypointss[i];
}
self.xpoints = tempxpoints;
self.ypoints = tempypoints;
But it still didn't work.
Edit: Thanks for all the answers. This ended up being my final Init code:
-(id)initWithXpoints:(double[]) xpointss Ypoints:(double[]) ypointss Npoints:(int)npointss
{
self = [super init];
if (self)
{
_xpoints = [[NSMutableArray alloc] init];
_ypoints = [[NSMutableArray alloc] init];
for (int i = 0; i < npointss; i++)
{
NSNumber *tempx = [NSNumber numberWithDouble:xpointss[i]];
NSNumber *tempy = [NSNumber numberWithDouble:ypointss[i]];
[_xpoints addObject:tempx];
[_ypoints addObject:tempy];
}
_npoints = npointss;
}
return self;
}
If you allocate the arrays as local variables, then they will be allocated on the stack. When execution leaves the function, those memory areas are freed up. You must use malloc() to allocate arrays that you can pass around and use free() to free them up.
// to allocate
double[] tempxpoints = (double[])malloc(sizeof(double) * npointss);
// to free when not used any more
free(tempxpoints);
But actually NSArray has been designed to handle these cases. And with ARC you don't even have to care about freeing the memory.
NSMutableArray *tempxpoints = [[NSMutableArray alloc] init];
[tempxpoints addObject:#2]; // wrap the double in an NSNumber object
If you were being fully Objective-C about it, you'd use an NSArray, fill it with NSNumbers and never specify a length. You can usually give them hints about how much space is likely to be required but Objective-C's collections all always size dynamically.
As of recent versions of the compiler, you can use array[x] notation on NSArray and write direct NSNumber constants as e.g. #4.5f if that sweetens the deal at all.
If you literally want C-style arrays then you'll need to descend to the C level of thought. So, something like:
#property(nonatomic, readonly) double * xpoints;
And:
-(id)initWithXpoints:(double[]) xpointss Ypoints:(double[]) ypointss Npoints:(int)npointss
{
self = [super init];
if (self){
size_t sizeOfArraysInBytes = sizeof(double)*npointss;
_xpoints = (double *)malloc(sizeOfArraysInBytes);
memcpy(_xpoints, xpointss, sizeOfArraysInBytes);
/* ... etc ... */
/* you never use self. notation in an init because it's a method call,
and method calls on objects that are not yet fully instantiated aren't
safe. Sample cause of failure: a subclass overrides the setter */
}
return self;
}
- (void)dealloc
{
free(_xpoints);
/* ... etc ... */
}
The array itself will be read/write elsewhere (it's the pointer that's read-only, not the things it points to) as class.xpoints[0], etc.
So I'm using recursive blocks. I understand that for a block to be recursive it needs to be preceded by the __block keyword, and it must be copied so it can be put on the heap. However, when I do this, it is showing up as a leak in Instruments. Does anybody know why or how I can get around it?
Please note in the code below I've got references to a lot of other blocks, but none of them are recursive.
__block NSDecimalNumber *(^ProcessElementStack)(LinkedList *, NSString *) = [^NSDecimalNumber *(LinkedList *cformula, NSString *function){
LinkedList *list = [[LinkedList alloc] init];
NSDictionary *dict;
FormulaType type;
while (cformula.count > 0) {
dict = cformula.pop;
type = [[dict objectForKey:#"type"] intValue];
if (type == formulaOperandOpenParen || type == formulaListOperand || type == formulaOpenParen) [list add:ProcessElementStack(cformula, [dict objectForKey:#"name"])];
else if (type == formulaField || type == formulaConstant) [list add:NumberForDict(dict)];
else if (type == formulaOperand) [list add:[dict objectForKey:#"name"]];
else if (type == formulaCloseParen) {
if (function){
if ([function isEqualToString:#"AVG("]) return Average(list);
if ([function isEqualToString:#"MIN("]) return Minimum(list);
if ([function isEqualToString:#"MAX("]) return Maximum(list);
if ([function isEqualToString:#"SQRT("]) return SquareRoot(list);
if ([function isEqualToString:#"ABS("]) return EvaluateStack(list).absoluteValue;
return EvaluateStack(list);
} else break;
}
}
return EvaluateStack(list);
} copy];
NSDecimalNumber *number = ProcessElementStack([formula copy], nil);
UPDATE
So in my own research I've discovered that the problem apparently does have to do with the references to the other blocks this block uses. If I do something simple like this, it doesn't leak:
__block void (^LeakingBlock)(int) = [^(int i){
i++;
if (i < 100) LeakingBlock(i);
} copy];
LeakingBlock(1);
However, if I add a another block in this, it does leak:
void (^Log)(int) = ^(int i){
NSLog(#"log sub %i", i);
};
__block void (^LeakingBlock)(int) = [^(int i){
Log(i);
i++;
if (i < 100) LeakingBlock(i);
} copy];
LeakingBlock(1);
I've tried using the __block keyword for Log() and also tried copying it, but it still leaks. Any ideas?
UPDATE 2
I found a way to prevent the leak, but it's a bit onerous. If I convert the passed in block to a weak id, and then cast the weak id back into a the block type, I can prevent the leak.
void (^Log)(int) = ^(int i){
NSLog(#"log sub %i", i);
};
__weak id WeakLogID = Log;
__block void (^LeakingBlock)(int) = [^(int i){
void (^WeakLog)(int) = WeakLogID;
WeakLog(i);
if (i < 100) LeakingBlock(++i);
} copy];
LeakingBlock(1);
Surely there's a better way?
Ok, I found the answer on my own...but thanks to those who tried to help.
If you're referencing/using other blocks in a recursive block, you must pass them in as weak variables. Of course, __weak only applies to block pointer types, so you must typedef them first. Here's the final solution:
typedef void (^IntBlock)(int);
IntBlock __weak Log = ^(int i){
NSLog(#"log sub %i", i);
};
__block void (^LeakingBlock)(int) = ^(int i){
Log(i);
if (i < 100) LeakingBlock(++i);
};
LeakingBlock(1);
The above code doesn't leak.
Aaron,
As your code appears to be single threaded, why are you copying the block? If you don't copy the block, you don't have a leak.
Andrew
Without further context information, I can say this:
You are leaking that block because you are copying it and not releasing it elsewhere. You need to copy it to move it to the heap, that's ok. But the way you've chosen is not entirely ok.
A correct way to do it is to store it as some object instance variable, copy it, and then release it inside dealloc. At least, that's a way to do it without leaking.
I have a simple loop with an int counter that gets incremented inside a while loop when a special case exists. My question is simply - how should I manage memory inside this function with regards to the int specifically? I've been using NSNumber almost exclusively and what little time I've spent with int seems to make me think I'm not doing releasing it correctly.
Any other improvements are also welcome but I'm very interested in the int question
- (NSArray *)parseJson:(NSArray *) items
{
NSMutableArray* hats = [[NSMutableArray alloc] init];
NSEnumerator *enumerator = [items objectEnumerator];
NSDictionary* item;
int counterz = 0;
while (item = (NSDictionary*)[enumerator nextObject]) {
Hat* hat = [[Hat alloc] init];
hat.addr = [item objectForKey:#"Address"];
BOOL* hasHat = [item objectForKey:#"HasHat"];
if ([hasHat boolValue]) {
if (counterz < 10) {
[hats addObject:hat];
counterz++;
}
}
}
return hats;
}
Thank you in advance!
You don't need to release a "normal" (i.e.: non-object based) int - it'll happily life out its (brief, tragic) life on the stack until it falls out of scope.
You've got a couple unnecessary things and some memory leaks...
- (NSArray *)parseJson:(NSArray *) items {
NSMutableArray *hats = [NSMutableArray array];
int counter = 0;
for (NSDictionary *item in items) {
Hat *hat = [[Hat alloc] init];
[hat setAddr:[item objectForKey:#"Address"]];
BOOL hasHat = [[item objectForKey:#"HasHat"] boolValue];
if (hasHat && counter < 10) {
[hats addObject:hat];
counter++;
}
[hat release];
}
return hats;
}
And heck, once you reach a counter of 10, you could break out of the loop, because you're never going to do anything useful once 10 is reached.
Some other comments:
The name of the method is wrong. Nothing about this method has to do with parsing JSON. At best you're interpreting an array of dictionaries that happened to originate from a JSON string, but there's nothing about the nature of this code that says "this is parsing JSON".
-[NSDictionary objectForKey:] returns an object. A BOOL is not an object, it's a primitive (like an int or char). Appending * to the type does not make it an object either. :)
Since the method name does not begin with new or alloc and does not contain the word copy, you're supposed to return an autoreleased object from it. The method in the question was returning an owned object (+1 retain count), since you invoked alloc, but never autorelease. Using the convenience constructor +array fixes this.
In your loop, you allocated a Hat object, but never released it. This is a classic memory leak.