pegasus/doc/Attic/changelog.txt - diff

Return to changelog.txt CVS log

Up to [Pegasus] / pegasus / doc / Attic

version 1.130, 2002/05/11 01:35:42

version 1.132, 2002/05/14 03:27:23

Line 213

is changed to (__int64), the tv_usec variable of timeval struct has

the correct value.

38. (Roger Kumpf - HP) 13 May 2002 - To eliminate the possibility of

collisions (and subsequent corruption) when different threads

concurrently update the ArrayRep reference count, I added a Mutex

to the ArrayRep structure. This Mutex gets locked in the inc()

and dec() methods that manage the reference count.

After adding the Mutex, the "mysterious" memory problems seemed to clear

up. This seems like a very good thing. However, note that the addition

of the Mutex adversely affects performance (nearly 2:1). I tried using

an AtomicInt for the reference count, but that seemed to yield even worse

performance on my system.

Anyone should feel free to optimize the performance of the locking in

the ArrayRep structure. Since I believe this locking is required for

proper operation, I've added the Mutex for all platforms. Note that I

had to comment out the assert statements in internal_dq.h to avoid a

circular header file dependency (Array.h->IPC.h->internal_dq.h->

Exception.h->String.h->Array.h).

39. (Roger Kumpf - HP) 13 May 2002 - Removed operator int() methods from

all CIM* classes. These methods had been causing a number of difficult

to find defects, including the use of the wrong versions of the methods

in Memory.h by the Array template (leading to major memory leaks).

In general, it is dangerous to define a casting operator method that

loses so much information, because the compiler will apply the cast

at will in an effort to match types. I've replaced all instances of

the int cast operator with isNull() methods.

========================================================================

Version 1.08 working towards 1.1 - Started 27 March 2002

Legend: