O3 versus O2 weirdness

[John Love-Jensen]

Hi Bartlomiej,

>> Maybe.  Did you have all the warnings turned on?

> by means of -Wall ? yes, they were turned on

Not merely -Wall, which ironically does not enable all the warnings.

But actually enable all the warnings (less whichever ones you've decided
can/should be ignored)?

Note:  GCC does not have -Wall-really-all-I-really-really-mean-it flag.  You
have to enable many warnings by hand.  (Most people probably wouldn't want
ALL those picayune warnings enabled, especially since the standard header
files themselves generate a lot of warnings.)  GCC does not -- and likely
never will -- have a "enable all warnings" flag.

I enable almost all warnings, with the exceptions of:
-Wno-unreachable-code    # templates generate bazillion warnings
-Wno-long-long           # "long long" 64-bit integer language extension
-Wno-four-char-constants # 'ABCD' 32-bit four-char-code language extension
-Wno-aggregate-return    # objects as return types
-Wno-system-headers      # system headers are noisy, otherwise
... parameterized warnings, e.g., -Wlarger-than-<len>
... and all the non-C++ warnings (since I program in C++)

The warnings I enable accounts for about 140 lines of my Makefile.

> it seems that more aggressive optimisation triggers code bugs, maybe it
> could be used for bug finding itself ?

Which is what the warning flags do.  They help detect questionable code.
(Caution: sometime too many spurious warnings.)

But warnings alone cannot prevent bad code.  Bad code can be written in any
language.

Sincerely,
--Eljay