Re: Bug or feature?

["Randy Brukardt" <randy@rrsoftware.com>]

"Georg Bauhaus" <rm-host.bauhaus@maps.futureapps.de> wrote in message 
news:5374819a$0$6696$9b4e6d93@newsspool2.arcor-online.net...
> On 14/05/14 23:48, Randy Brukardt wrote:
>> To compile ACATS tests in GNAT, I have to use a small boatload of 
>> options:
>>
>> gnatmake
>> C457003.adb -eS -gnat12 -O0 -gnatE -gnato -gnatv -gnatws -gnatd7 -bargs - 
>> T0
>
> Would it be meaningful when testing any compiler, to include
> the optimizers typically used when translating production code?
> -O2 seems commonly used with GNAT.

I picked these options originally because they were the ones used during the 
latest formal conformity assessment of GNAT. (I've since modified them a bit 
on the advice of the AdaCore ACATS test person.) That's the only set of 
options that anyone ever guaranteed actually met the Standard.

One would expect that internally, AdaCore tests other sets of options as 
well. Optimization is sometimes a problem, as really powerful optimizers can 
sometimes eliminate or invalidate ACATS tests. ACATS tests have been 
repaired to avoid optimization effects, but its a never-ending game of 
whack-a-mole. (As optimizers get better, new problems emerge, which require 
still more test repairs, etc.) In addition, some optimization modes probably 
aren't standards-conformant. (For instance, Janus/Ada has a mode where all 
objects are assumed to be in range. This matches our Ada 83 compiler, but 
it's not correct for Ada 95 and later.)

For Janus/Ada, I run 3 different sets of optimization options, combined with 
3 different language settings. But that's for in-house use only; a formal 
conformity assessment would be done with the optimization off. The in-house 
goal is to minimize failures with the optimizer on but there are a few 
failures that are effectively unfixable, so I doubt it would ever be 
perfect.

                                 Randy.