How can I find calls to UIKit instances from a secondary thread? - objective-c

My app is crashing in iOS 5 because I have some code that is calling UIKit instances from a secondary thread. You know you have this problem when you see the following error:
bool _WebTryThreadLock(bool), 0x811bf20: Multiple locks on web thread not allowed! Please file a bug. Crashing now…
So my question is what are some ways that I can find the code that is calling the UIKit instances from a secondary thread?
Here are some things I’ve tried already:
Commented out blocks that could be violating the rule
Added assert([NSThread isMainThread]) in places that might be processing in secondary thread
Added a symbolic breakpoint for _WebTryThreadLock
These things have helped me to find problem areas. However, in my final crash the _WebTryThreadLock breakpoint has no stack trace in any of the other threads. So, how I can find the code that causing the problem without a stack trace?
Thanks for your time!

Your assert() is probably the most valuable tool in this. I've been known to put a similar assertion at the beginning of every method in my Controller classes. If that doesn't find it, I add the assertion to my View classes. If that doesn't find it, I add it to any Model classes that I think are main-thread only.
To #craig's comment, the fact that it claims to be an internal bug might be accurate. But I think you're on the right path to closely examine your own code first.

I adapted the PSPDFUIKitMainThreadGuard.m to allow one to not have to worry about these things. Here: https://gist.github.com/k3zi/98ca835b15077d11dafc :
#import <objc/runtime.h>
#import <objc/message.h>
// Compile-time selector checks.
#define PROPERTY(propName) NSStringFromSelector(#selector(propName))
// A better assert. NSAssert is too runtime dependant, and assert() doesn't log.
// http://www.mikeash.com/pyblog/friday-qa-2013-05-03-proper-use-of-asserts.html
// Accepts both:
// - PSPDFAssert(x > 0);
// - PSPDFAssert(y > 3, #"Bad value for y");
#define PSPDFAssert(expression, ...) \
do { if(!(expression)) { \
NSLog(#"%#", [NSString stringWithFormat: #"Assertion failure: %s in %s on line %s:%d. %#", #expression, __PRETTY_FUNCTION__, __FILE__, __LINE__, [NSString stringWithFormat:#"" __VA_ARGS__]]); \
abort(); }} while(0)
///////////////////////////////////////////////////////////////////////////////////////////
#pragma mark - Helper for Swizzling
BOOL PSPDFReplaceMethodWithBlock(Class c, SEL origSEL, SEL newSEL, id block) {
PSPDFAssert(c && origSEL && newSEL && block);
Method origMethod = class_getInstanceMethod(c, origSEL);
const char *encoding = method_getTypeEncoding(origMethod);
// Add the new method.
IMP impl = imp_implementationWithBlock(block);
if (!class_addMethod(c, newSEL, impl, encoding)) {
NSLog(#"Failed to add method: %# on %#", NSStringFromSelector(newSEL), c);
return NO;
}else {
// Ensure the new selector has the same parameters as the existing selector.
Method newMethod = class_getInstanceMethod(c, newSEL);
PSPDFAssert(strcmp(method_getTypeEncoding(origMethod), method_getTypeEncoding(newMethod)) == 0, #"Encoding must be the same.");
// If original doesn't implement the method we want to swizzle, create it.
if (class_addMethod(c, origSEL, method_getImplementation(newMethod), encoding)) {
class_replaceMethod(c, newSEL, method_getImplementation(origMethod), encoding);
}else {
method_exchangeImplementations(origMethod, newMethod);
}
}
return YES;
}
// This installs a small guard that checks for the most common threading-errors in UIKit.
// This won't really slow down performance but still only is compiled in DEBUG versions of PSPDFKit.
// #note No private API is used here.
__attribute__((constructor)) static void PSPDFUIKitMainThreadGuard(void) {
#autoreleasepool {
for (NSString *selStr in #[PROPERTY(setNeedsLayout), PROPERTY(setNeedsDisplay), PROPERTY(setNeedsDisplayInRect:)]) {
SEL selector = NSSelectorFromString(selStr);
SEL newSelector = NSSelectorFromString([NSString stringWithFormat:#"pspdf_%#", selStr]);
if ([selStr hasSuffix:#":"]) {
PSPDFReplaceMethodWithBlock(UIView.class, selector, newSelector, ^(__unsafe_unretained UIView *_self, CGRect r) {
if(!NSThread.isMainThread){
dispatch_async(dispatch_get_main_queue(), ^{
((void ( *)(id, SEL, CGRect))objc_msgSend)(_self, newSelector, r);
});
}else{
((void ( *)(id, SEL, CGRect))objc_msgSend)(_self, newSelector, r);
}
});
}else {
PSPDFReplaceMethodWithBlock(UIView.class, selector, newSelector, ^(__unsafe_unretained UIView *_self) {
if(!NSThread.isMainThread){
dispatch_async(dispatch_get_main_queue(), ^{
((void ( *)(id, SEL))objc_msgSend)(_self, newSelector);
});
}else
((void ( *)(id, SEL))objc_msgSend)(_self, newSelector);
});
}
}
}
}
It automatically kicks calls into the main thread and thus you wouldn't even have to do anything but plop the code in.

This problem comes because you want to access to UI from secondary thread somehow, it can from webview of whatever else. It is not permitted because UIKit is not thread safe and can be accessed only from MainThread.
The very first thing you can do is to change your thread call to [self performSelectorOnMainThread:#selector(myMethod) withObject:nil waitUntilDone:NO]; (look for documentation).
In case when you have no other choice you can use GCD(Grand Central Dispathc)...

This code (just add to project and compile this file without ARC) causes assertions on UIKit access outside the main thread: https://gist.github.com/steipete/5664345
I've just used it to pickup numerous UIKit/main thread issues in some code I've just picked up.

Related

Casting method to block in Objective C

I'm using a framework that uses a huge amount of block arguments to emit events. Due to licensing I can't release code samples, but I can still get my point across with similar code. Lets say you have the following code:
- (void)begin
{
[MyService doSomething:^{
NSLog(#"Done.");
}];
}
This is fine, but the methods I'm dealing with have 14 block arguments, most of which take several parameters so it makes it very hard to read and difficult to test without creating repetitive boiler-plate code.
Is there a way I can point a block to an instance's method, something like:
- (void)doSomethingDelegate
{
NSLog(#"Done.");
}
- (void)begin
{
[MyService doSomething:CALLBACK(self, #selector(doSomethingDelegate))];
}
Where CALLBACK is a macro or native construct. Is this possible?
No, but you can declare local blocks:
- (void)begin
{
void (^callback)() = ^{ NSLog(#"Done."); };
[MyService doSomething:callback];
}
You could extend that to:
- (void)callbackImpl {
NSLog(#"Done.");
}
- (void)begin {
void (^callback)() = ^() { [self callbackImpl]; };
[MyService doSomething:callback];
}
And you could imagine a macro of the sort:
#define CALLBACK(name) void (^name)() = ^{ [self name##Impl]; };
It gets a little messier when you have arguments to the block, and you need to consider if you want a strong or weak reference to self in those blocks...
Since I also have to worry about arguments that are in a different from this is the best solution I've found so far:
#define wrapper(args, method) ^args { [self method]; }
- (void)onUpdatingDeviceWithMessage:(NSString *)message progress:(float)progress
{
}
- (void)begin
{
[MyService ...
onUpdatingDevice:wrapper((NSString *message, float progress),
onUpdatingDeviceWithMessage:message progress:progress)
}

Is it possible to check if an NSThread is blocked?

I've always been interested in how to write the following code to use it for unit testing:
Is it possible to extend NSThread with a method that would check if a particular thread is blocked?
Right now I'am working with NSCondition: Xcode shows me the chain which is called by -wait to block the thread:
[NSCondition wait]
pthread_cond_wait$UNIX2003
_pthread_cond_wait
__psynch_cvwait
Besides checking the locks done by NSCondition, if it is even possible, I would highly appreciate method working also for any other blocking capabilities (dispatch semaphores, condition locks, sleeping threads and so on, ) - I have no idea about Objective-C internals, if maybe they could be catched by one method or each needs its own.
Here is a simple example of what I would like to achieve. The mysterious method is called isBlocked.
// Some test case
// ...
__block NSThread *thread;
NSCondition *condition = [NSCondition alloc] init];
dispatch_async(someQueue(), ^{
thread = NSThread.currentThread;
[condition lock];
[condition wait];
[condition unlock];
});
while(1) {
NSLog(#"Thread is blocked: %d", thread.isBlocked);
}
Note: I am not good at C and all this low-level POSIX stuff, so, please, be verbose.
Note 2: I am interested in solutions working for dispatch queues as well: if someone can show me how to test the fact that someQueue() is blocked by -[NSCondition wait] (not the fact that it is going to be blocked (fx hacking some code before -[condition wait] is run and the block is set), but the fact that thread/queue is blocked), I will accept this as an answer as much like I would do with working -[NSThread isBlocked] method.
Note 3: Suspecting bad news like "it is not possible", I claim that any ideas about catching the fact that -[condition wait] was run and the thread was set blocked (see Note 2) are appreciated and can be also accepted as an answer!
UPDATE 1 in address to the nice answer by Richard J. Ross III. Unfortunately, his answer does not work in my original example, the version which is closer to my real work (though it does not differ much from the example I've initially provided - sorry that I didn't include it in the first edition of the question):
// Example
// Here I've bootstrapped Richard's isLocking categories for both NSThread and NSCondition
// ...
// somewhere in SenTesting test case...
__block NSThread *thread;
NSCondition *condition = [NSCondition alloc] init];
__block BOOL wePassedBlocking = NO;
dispatch_async(someQueue(), ^{
thread = NSThread.currentThread;
[condition lock];
[condition wait];
[condition unlock];
wePassedBlocking = YES; // (*) This line is occasionally never reached!
});
while(!thread.isWaitingOnCondition); // I want this loop to exit after the condition really locks someQueue() and _thread_ __.
// sleep(1);
[condition lock];
[condition broadcast]; // BUT SOMETIMES this line is called before -[condition wait] is called inside someQueue() so the entire test case becomes blocked!
[condition unlock];
while(!wePassedBlocking); // (*) And so this loop occasionally never ends!
If I uncomment sleep(1) test begins working very stable without any occasional locks!
This leads us to the problem, that Richard's category does set state exactly one line before the actual blocking is done meaning that sometimes test case's main thread catches this new state before we actually have someQueue/thread blocked because Richard's code does not contain any synchronization mechanisms: #synchronized, NSLock or something like that! I hope I am making a clear explanation of this tricky case. For anyone who has doubts about what I've posted here, I would say that I have been also experimenting with multiple queues and even more complex cases, and if needed I'm ready to provide more examples. Richard, thanks again for your effort, let's think more together, if you understand these my points!
UPDATE 2
I see the dead-end paradox: obviously, to really set the state of waitingOnCondition we need to wrap this state's change inside some synchronization closures, but the problem is that the closing one, unlocking the synchronization lock, should be called after -[condition wait], but it can't, because the thread is already blocked. Again, I hope I am describing it pretty clear.
Here you go! It won't detect threads being waited on by anything other than -[NSCondition wait], but it could easily be extended to detect other kinds of waiting.
It's probably not the best implementation out there, but it does in fact work, and will do what you need it to.
#import <objc/runtime.h>
#implementation NSThread(isLocking)
static int waiting_condition_key;
-(BOOL) isWaitingOnCondition {
// here, we sleep for a microsecond (1 millionth of a second) so that the
// other thread can catch up, and actually call 'wait'. This time
// interval is so small that you will never notice it in an actual
// application, it's just here because of how multithreaded
// applications work.
usleep(1);
BOOL val = [objc_getAssociatedObject(self, &waiting_condition_key) boolValue];
// sleep before and after so it works on both edges
usleep(1);
return val;
}
-(void) setIsWaitingOnCondition:(BOOL) value {
objc_setAssociatedObject(self, &waiting_condition_key, #(value), OBJC_ASSOCIATION_RETAIN);
}
#end
#implementation NSCondition(isLocking)
+(void) load {
Method old = class_getInstanceMethod(self, #selector(wait));
Method new = class_getInstanceMethod(self, #selector(_wait));
method_exchangeImplementations(old, new);
}
-(void) _wait {
// this is the replacement for the original wait method
[[NSThread currentThread] setIsWaitingOnCondition:YES];
// call the original implementation, which now resides in the same name as this method
[self _wait];
[[NSThread currentThread] setIsWaitingOnCondition:NO];
}
#end
int main()
{
__block NSCondition *condition = [NSCondition new];
NSThread *otherThread = [[NSThread alloc] initWithTarget:^{
NSLog(#"Thread started");
[condition lock];
[condition wait];
[condition unlock];
NSLog(#"Thread ended");
} selector:#selector(invoke) object:nil];
[otherThread start];
while (![otherThread isWaitingOnCondition]);
[condition lock];
[condition signal];
[condition unlock];
NSLog(#"%i", [otherThread isWaitingOnCondition]);
}
Output:
2013-03-20 10:43:01.422 TestProj[11354:1803] Thread started
2013-03-20 10:43:01.424 TestProj[11354:1803] Thread ended
2013-03-20 10:43:01.425 TestProj[11354:303] 0
Here is a solution using dispatch_semaphore_t
PGFoo.h
#import <Foundation/Foundation.h>
#interface PGFoo : NSObject
- (void)longRunningAsynchronousMethod:(void (^)(NSInteger result))completion;
#end
PGFoo.m
#import "PGFoo.h"
#implementation PGFoo
- (void)longRunningAsynchronousMethod:(void (^)(NSInteger))completion {
dispatch_async(dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_DEFAULT, 0), ^{
sleep(5);
completion(1);
});
}
#end
Test Methods
- (void)testThatFailsBecauseItIsImpatient {
PGFoo *foo = [[PGFoo alloc] init];
__block NSInteger theResult = 0;
[foo longRunningAsynchronousMethod:^(NSInteger result) {
theResult = result;
}];
STAssertEquals(theResult, 1, nil);
}
- (void)testThatPassesBecauseItIsPatient {
PGFoo *foo = [[PGFoo alloc] init];
__block NSInteger theResult = 0;
dispatch_semaphore_t semaphore = dispatch_semaphore_create(0);
[foo longRunningAsynchronousMethod:^(NSInteger result) {
theResult = result;
dispatch_semaphore_signal(semaphore);
}];
dispatch_semaphore_wait(semaphore, DISPATCH_TIME_FOREVER);
STAssertEquals(theResult, 1, nil);
}
By using a dispatch_semaphore_t you can "track" whether a thread that is waiting on that semaphore is blocked. For every call of dispatch_semaphore_wait the semaphore's count is decremented and the thread waits until a call of dispatch_semaphore_signal is made, when dispatch_semaphore_signal is called the semaphore's count is incremented, if the count is incremented to a value greater than -1 the thread continues.
This solution fails to answer your question about checking whether an NSThread is "blocked" but I think it provides what you are reaching for, assuming you're not reaching to check on NSThread instances that are maintained within an existing framework.

Calling Obj-C Code from JavaScript via Console: Arguments get dropped?

Having a heck of a time with this one.
I've got a super-simple Cocoa app containing one WebView, a WebScripting API defined in the page, and a single NSObject defined on that API. When I turn on the debugger tools (in the embedded WebView), I can see the API on the JavaScript window object, and I can see my "api" property defined on that -- but when I call the API's "get" method, the arguments aren't being serialized -- when the Obj-C method gets called, the arguments are missing. See below, which hopefully illustrates:
I've combed through the docs, I've (apparently) set the appropriate methods to expose everything that needs to be exposed, and I can see the method being called. There has to be something stupid I'm missing, but as a relative newbie to this environment, I'm not seeing it.
Thanks in advance for your help!
Have you set WebKitDeveloperExtras to YES in your default user defaults when you send -[NSUserDefaults registerDefaults:]?
Depending on what version of Xcode you're using you could be getting a known error. If you're using LLDB on anything but the most recent version, it might not be giving you the right variables in the debugger. The solution has been to use GDB instead of LLDB until Apple fixes the problem. But I think they fixed the problem in the latest version. I'd change the debugger to use GDB and see if you're getting the right variables in Xcode. (Product-> Edit Scheme...-> Run -> Debugger). I came across this problem in iOS, though, so I don't know its applicability to OSX. Worth a try anyway.
I originally came across the problem here: https://stackoverflow.com/a/9485349/1147934
I process javascript in the main thread of my app from a local file stored in the apps directory. I check for beginning and ending tokens for the js functions I am executing and whether the function contains a variable.
Hopefully this can give you some good ideas for your issue. You could also do alerts in the js to see if the values post correctly as you run the app (I am sure you thought of that already, but it's worth mentioning.) Happy coding! I hope this helps!
in the .h file define:
NSMutableString *processedCommand;
NSArray *commandArguments;
In the .m file:
// tokens
#define kOpenToken #"<%%"
#define kCloseToken #"%%>"
// this will throw
-(void)executeJScriptCommand:(NSString *)aCommand {
[self performSelectorOnMainThread:#selector(executeThisCommand:) withObject:aCommand waitUntilDone:YES];
}
// this will throw
-(NSString *)executeCommand:(NSString *)command {
NSString *aCommand = [[[command stringByReplacingOccurrencesOfString:kOpenToken withString:#""]
stringByReplacingOccurrencesOfString:kCloseToken withString:#""]
stringByTrimmingLeadingAndTrailingWhitespaces];
if ([aCommand hasPrefix:#"="])
{
// variable. get value
[self getVariableFromCommand:aCommand];
}
else {
[self executeThisCommand:aCommand];
}
NSString *returnValue = [NSString stringWithString:processedCommand];
self.processedCommand = nil;
self.commandArguments = nil;
return returnValue;
}
-(void)executeThisCommand:(NSString *)aCommand {
BOOL hasError = NO;
// clear result
self.processedCommand = nil;
self.commandArguments = nil;
BOOL isFromJS = NO;
NSString *function = nil;
NSMutableArray *commandParts = nil;
#try {
// first, break the command into its parts and extract the function that needs to be called, and the (optional) arguments
commandParts = [[NSMutableArray alloc] initWithArray:[aCommand componentsSeparatedByString:#":"]];
if ([[[commandParts objectAtIndex:0] lowercaseString] isEqualToString:#"js-call"]) {
isFromJS = YES;
[commandParts removeObjectAtIndex:0];
}
// get our function, arguments
function = [[commandParts objectAtIndex:0] retain];
[commandParts removeObjectAtIndex:0];
if ([commandParts count] > 0){
if (isFromJS == YES) {
NSString *arguments = [[commandParts objectAtIndex:0] stringByReplacingPercentEscapesUsingEncoding:NSUTF8StringEncoding];
if ([arguments length] > 0) {
self.commandArguments = [arguments JSONValue];
}
}
else {
self.commandArguments = [NSArray arrayWithArray:commandParts];
}
}
// build invoke
SEL sel = NSSelectorFromString(function);
if ([self respondsToSelector:sel]) {
[self performSelectorOnMainThread:sel withObject:nil waitUntilDone:YES];
// using invocation causes a SIGABORT because the try/catch block was not catching the exception.
// using perform selector fixed the problem (i.e., the try/catch block now correctly catches the exception, as expected)
}
else {
[appDelegate buildNewExceptionWithName:#"" andMessage:[NSString stringWithFormat:#"Object does not respond to selector %#", function]];
}
}
#catch (NSException * e) {
hasError = YES;
[self updateErrorMessage:[NSString stringWithFormat:#"Error processing command %#: %#", aCommand, [e reason]]];
}
#finally {
[function release];
[commandParts release];
}
if (hasError == YES) {
[appDelegate buildNewExceptionWithName:#"executeThisCommand" andMessage:self.errorMessage];
}
}
// this can return nil
-(NSString *)getQueryStringValue:(NSString *)name {
NSString *returnValue = nil;
if (queryString != nil) {
returnValue = [queryString objectForKey:[name lowercaseString]];
}
return returnValue;
}

asiHttpRequest under iOS5

I am succesfuly using AsiHttpRequest library to make url connections in my apps. However, I upgrade to iOS5 and Reachability.m file is reporting some errors (4) on following functions:
static void ReachabilityCallback(SCNetworkReachabilityRef target, SCNetworkReachabilityFlags flags, void* info) {
#pragma unused (target, flags)
NSCAssert(info, #"info was NULL in ReachabilityCallback");
NSCAssert([(NSObject*) info isKindOfClass: [Reachability class]], #"info was the wrong class in ReachabilityCallback");
// Post a notification to notify the client that the network reachability changed.
[[NSNotificationCenter defaultCenter] postNotificationName: kReachabilityChangedNotification object: (Reachability *) info];
} // ReachabilityCallback()
- (BOOL) startNotifier {
SCNetworkReachabilityContext context = {0, self, NULL, NULL, NULL};
if(SCNetworkReachabilitySetCallback(reachabilityRef, ReachabilityCallback, &context)) {
if(SCNetworkReachabilityScheduleWithRunLoop(reachabilityRef, CFRunLoopGetCurrent(), kCFRunLoopDefaultMode)) {
return YES;
}
}
return NO;
} // startNotifier
1sr error: in NSCAssert line, "Cast of C pointer type 'void' to Objective-C pointer type NSObject requires a bridged cast". Why happens and how to solve it?
ANSWER: you can disable ARC for each file. Go to build settings of your project and set the -fno-objc-arc flag on all the ASIHTTPRequest files (double click to edit text). Then you must remove ASIAuthenticationDialog and any references to it that are still generating errors. It works for me.
EDIT: I remember now and yes, the problem is ARC. But you can exclude files from being complied under ARC by setting the following compiler flag in Build Phases >> Compile Sources: -fno-objc-arc. If you select all the ASIHTTPRequest files and double-click, you can set the flag for all of them in one fell swoop.
ORIGINAL POST:
I've been using ASIHTTPRequest for a couple of weeks now and I remember reading a post somewhere about problems with Reachability, I just can't remember what it was exactly.
Anyway, this is what those lines in my Reachability.m look like:
//Start listening for reachability notifications on the current run loop
static void ReachabilityCallback(SCNetworkReachabilityRef target, SCNetworkReachabilityFlags flags, void* info) {
#pragma unused (target, flags)
NSCAssert(info, #"info was NULL in ReachabilityCallback");
NSCAssert([(NSObject*) info isKindOfClass: [Reachability class]], #"info was the wrong class in ReachabilityCallback");
//We're on the main RunLoop, so an NSAutoreleasePool is not necessary, but is added defensively
// in case someone uses the Reachablity object in a different thread.
NSAutoreleasePool* pool = [NSAutoreleasePool new];
// Post a notification to notify the client that the network reachability changed.
[[NSNotificationCenter defaultCenter] postNotificationName: kReachabilityChangedNotification
object: (Reachability *) info];
[pool release];
} // ReachabilityCallback()
- (BOOL) startNotifier {
SCNetworkReachabilityContext context = {0, self, NULL, NULL, NULL};
if(SCNetworkReachabilitySetCallback(reachabilityRef, ReachabilityCallback, &context)) {
if(SCNetworkReachabilityScheduleWithRunLoop(reachabilityRef, CFRunLoopGetCurrent(), kCFRunLoopDefaultMode)) {
return YES;
}
}
return NO;
} // startNotifier

NSInvocation Leaks

I am trying to setup an NSInovcation system to launch selectors into background threads using performSelectorInBackground: - So far everything is successful when running the system on instance methods (-), but I also want to support class methods (+). I have adjusted my code to provide an invokeInBackgroundThread for both types of class and everything worked except for one problem. When the class methods are invoked I get my console flooded with "autoreleased with no pool in place" messages. No idea what is causing it. The code which is based off the DDFoundation open source project is shown below.
#implementation NSObject (DDExtensions)
...
+ (id)invokeInBackgroundThread
{
DDInvocationGrabber *grabber = [DDInvocationGrabber invocationGrabber];
[grabber setInvocationThreadType:INVOCATION_BACKGROUND_THREAD];
return [grabber prepareWithInvocationTarget:self];
}
- (id)invokeInBackgroundThread
{
DDInvocationGrabber *grabber = [DDInvocationGrabber invocationGrabber];
[grabber setInvocationThreadType:INVOCATION_BACKGROUND_THREAD];
return [grabber prepareWithInvocationTarget:self];
}
...
...
- (void)forwardInvocation:(NSInvocation *)ioInvocation
{
[ioInvocation setTarget:[self target]];
[self setInvocation:ioInvocation];
if (_waitUntilDone == NO) {
[_invocation retainArguments];
}
if (_threadType == INVOCATION_MAIN_THREAD)
{
[_invocation performSelectorOnMainThread:#selector(invoke)
withObject:nil
waitUntilDone:_waitUntilDone];
} else {
[_invocation performSelectorInBackground:#selector(invoke)
withObject:nil];
}
}
...
+(void)doSomething;
[[className invokeOnBackgroundThread] doSomething];
Main thread has autorelease pool by default, if you start extra thread - it's your job to create the pool. Actually, nothing complicated here, just
NSAutoreleasePool* pool = [[NSAutoreleasePool alloc] init];
// Work...
[pool release];
Also, if you have a lot of threads, I'd suggest you to take a look at NSOperation instead of running threads with [performSelectorInBackground]. NSOperation (with wrapping queue) is more flexible solution for such tasks.