Re: Fix operator precedence in _ShiftMin1.

[Paolo Carlini <paolo dot carlini at oracle dot com>]

Ed Smith-Rowland wrote:
> This doesn't really have much to do with 39629.  Sorry for the noise.
Ok, I'll try to fix 39629 asap. In any case, since the tests apparently
have some issues on 32-bit machines, please make sure the carefully
regression test any further patch also in multilibed 32-bit mode (I
suppose you are on x86_64), something like:

  make -k check-target-libstdc++-v3
RUNTESTFLAGS="--target_board='unix/\{-m32,-m64\}'"

Then, If you can confirm that no further fails are introduced (by
enabling that by-hand disabled VERIFY), then the patch is ok of course.

Finally, about the stdint.h issue which I mentioned in the audit trail
of 39629, yet another  unrelated issue in fact, I see that bits/random.h
uses many stdint.h types, like uint_fast64_t, and that implies that 
(until Joseph will be able to close PR 448), the *entire* facility
should be disabled when we cannot assume that stdint.h is available,
similarly to e.g., <thread>, all the tests guarded with
dg-require-cstdint "", etc. Please let me know if you want to deal with
this last issue, otherwise I'll only minimally fix 39629, for the moment.

Paolo.