Re: Jenkins build became unstable: epics-base-3.14-win64 #105

[Andrew Johnson <anj@aps.anl.gov>]

On 02/23/2016 06:31 PM, Andrew Johnson wrote:
> That's interesting, #593 failed but #595 passed; exactly the same
> operation but with swapped inputs. The code is this:
> 
>         case BIT_OR:
>             utop = *ptop--;
>             *ptop = (epicsInt32) ((epicsUInt32) *ptop | utop);
>             break;

Aarrrggggghhh. Stupid F*ing MS compiler/optimizer. This version passes
both tests
>         case BIT_OR:
>             utop = *ptop--;
>             utop |= (epicsUInt32) *ptop;
>             itop = utop;
> printf("result = %d (0x%x)\n", itop, utop);
>             *ptop = itop;
>             break;

but if I remove/comment out the printf() it fails again exactly as
before, i.e. when the first operand (2nd on stack) is 2863311530.1.

The only way I can fix this is to turn off MS global optimizations for
the duration of the calcPerform() function, which does allow the
original code to work. The win32-x86 build doesn't need that though

- Andrew

-- 
There are only two hard problems in distributed systems:
  2. Exactly-once delivery
  1. Guaranteed order of messages
  2. Exactly-once delivery
 -- Mathias Verraes