Re: Signed overflow Was: Re: GNU poke 2.0.92 with ubsan

[Jose E. Marchesi]

> Jose E. Marchesi wrote:
>> So I tried with ubsan with your patch, plus these changes:
>
> Looks good.
>
>> However, the changes in pkl-fold.c trigger a compile-time warning
>> (-Woverflow) in the gnulib macros INT_*_OVERFLOW.
>> 
>> This is one of these warnings:
>> 
>> In file included from ../../gl-libpoke/xalloc.h:27:0,
>>                  from ../../libpoke/pkl-fold.c:28:
>> ../../libpoke/pkl-fold.c: In function 'pkl_fold_or':
>> ../../gl-libpoke/intprops.h:310:65: warning: integer overflow in expression 
>> [-Woverflow]
>>         : (a) % (_GL_INT_CONVERT (a, _GL_SIGNED_INT_MAXIMUM (b)) + 1))   \
>>                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~
>> ../../gl-libpoke/intprops.h:300:19: note: in expansion of macro 
>> '_GL_UNSIGNED_NEG_MULTIPLE'
>>     : (b) < 0 && ! _GL_UNSIGNED_NEG_MULTIPLE (a, b, max))
>>                    ^~~~~~~~~~~~~~~~~~~~~~~~~
>> ../../gl-libpoke/intprops.h:388:3: note: in expansion of macro 
>> '_GL_REMAINDER_OVERFLOW'
>>    op_result_overflow (a, b,                                     \
>>    ^~~~~~~~~~~~~~~~~~
>> ../../gl-libpoke/intprops.h:378:3: note: in expansion of macro 
>> '_GL_BINARY_OP_OVERFLOW'
>>    _GL_BINARY_OP_OVERFLOW (a, b, _GL_REMAINDER_OVERFLOW)
>>    ^~~~~~~~~~~~~~~~~~~~~~
>> ../../libpoke/pkl-fold.c:613:23: note: in expansion of macro 
>> 'INT_REMAINDER_OVERFLOW'
>>                    if (INT_REMAINDER_OVERFLOW (op1_val,                  \
>
> Hmm. I recall that some warnings may occur in dead code in intprops.h,
> depending on GCC versions. This doesn't look serious to me. But will follow 
> up.

Note that the warning does not appear if -fsanitize=undefined is not
used.