Re: Simple Out-of-Box Demo: Private Framework Crashes in 10.3.9

FROM : Jerry Krinock
DATE : Wed Mar 26 01:59:49 2008

An off-list respondent suggested that possibly Panther is not 
correctly initializing the NSString category in my framework into the 
NSString class.  I added a +initialize method to my category, and got 
a strange result which might give someone a clue...

@implementation NSString (Crash)

+ (void)initialize {
    NSLog(@"Class: %@ initializing.  self: %p", 
NSStringFromClass(self), self);
}

... // My other method which crashes when it sends an internal 
message, as in
... // http://www.cocoabuilder.com/archive/message/cocoa/2008/3/21/201996

@end

Now, please understand that I'm not sure whether or not it's kosher to 
add +initialize in a category on a Cocoa class, but the interesting 
thing is that when I run this in Leopard, seven classes (it's a 
"cluster", I suppose) are initialized without complaint...

    TestApp[88066:10b] Class: NSCFString initializing.  self: 0xa00a14a0
    TestApp[88066:10b] Class: NSMutableString initializing.  self: 
0xa03f7480
    TestApp[88066:10b] Class: NSPlaceholderString initializing.  self: 
0xa03f8f40
    TestApp[88066:10b] Class: NSString initializing.  self: 0xa03f8f00
    TestApp[88066:10b] Class: NSCheapMutableString initializing. 
self: 0xa03f74c0
    TestApp[88066:10b] Class: NSPathStore2 initializing.  self: 
0xa03f7ac0
    TestApp[88066:10b] Class: NSPlaceholderMutableString 
initializing.  self: 0xa03f7500

When I run this in Panther, ten classes are initialized (looks like 
Apple has trimmed some of the fat since then), but in the first four 
of them, +initialize seems to run before there is an autorelease pool...

    TestApp[4105] *** _NSAutoreleaseNoPool(): Object 0x301270 of class 
NSCFString autoreleased with no pool in place - just leaking
    TestApp[4105] Class: NSCFString initializing.  self: 0xa01c05f4

    TestApp[4105] *** _NSAutoreleaseNoPool(): Object 0x309600 of class 
NSCFString autoreleased with no pool in place - just leaking
    TestApp[4105] Class: %NSCFString initializing.  self: 0xa0a35c9c

    TestApp[4105] *** _NSAutoreleaseNoPool(): Object 0x301100 of class 
NSCFString autoreleased with no pool in place - just leaking
    TestApp[4105] Class: NSMutableString initializing.  self: 0xa0a35fcc

    TestApp[4105] *** _NSAutoreleaseNoPool(): Object 0x307010 of class 
NSCFString autoreleased with no pool in place - just leaking
    TestApp[4105] Class: NSString initializing.  self: 0xa0a360ec

    TestApp[4105] Class: NSCheapMutableString initializing.  self: 
0xa0a35f9c

    TestApp[4105] Class: NSPathStore2 initializing.  self: 0xa0a356cc

    TestApp[4105] Class: NSPlaceholderString initializing.  self: 
0xa0a360bc

    TestApp[4105] Class: NSPlaceholderMutableString initializing. 
self: 0xa0a35f6c

    TestApp[4105] Class: NSBigMutableString initializing.  self: 
0xa0a35c6c

    TestApp[4105] Class: NSBulletString initializing.  self: 0xa72188d0

Now, it sort of makes sense.  The first four classes to be 
+initialized are NSCFString, %NSCFString, NSMutableString and 
NSString.  If indeed it is not kosher to add a +initialize method in a 
category, then I suppose it is possible that the autorelease pool 
might not be in place for an NSString before NSString is initialized. 
(I realize the previous sentence is 50% nonsense, but I'm groping in 
the dark!)  But why the difference in Panther?  Maybe it's the same 
thing causing my other methods to crash.

Anyhow, if this gives anyone a clue as to why a simple method in a 
category on NSString in a private framework would crash the first time 
it sends an internal message, please let us know.  In case you missed 
this thread last week, want to see the details and/or download build 
transcripts and/or a demo project, here's a handy link...

http://www.cocoabuilder.com/archive/message/cocoa/2008/3/21/201996

Jerry Krinock