Re: NSTask stealth bug in readDataOfLength!! :(

FROM : Chris Suter
DATE : Mon Oct 30 02:01:30 2006

> My guess is that what's probably happening is that your program is 
> getting a signal (probably SIGCHILD or SIGPIPE or something), which 
> is causing the read() call inside -readDataOfLength: to return 
> prematurely with EINTR, hence the exception you're seeing.  It's 
> quite annoying that the -readData methods don't handle this 
> better.  I think in at least one place in our code we intentionally 
> avoid using NSFileHandle's -read methods for this very reason.
>
> I'm sure we must have filed a bug report about this, but I can't 
> find anything.  Maybe my colleague Chris did, I'm not sure.


I did file a bug.

I also changed all our code to use:

@implementation NSFileHandle (CSFileHandleExtensions)

- (NSData *)availableDataOrError:(NSException **)returnError;
{
  for (;;) {
    @try {
      return [self availableData];
    } @catch (NSException *e) {
      if ([[e name] isEqualToString:NSFileHandleOperationException]) {
   if ([[e reason] isEqualToString:
     @"*** -[NSConcreteFileHandle availableData]: Interrupted system 
call"]) {
     continue;
   }
   if (returnError)
     *returnError = e;
   return nil;
      }

      @throw;
    }
  }
}

@end

Not particularly pleasant, I'll admit.

- Chris