On 14/12/16 18:38 -0300, Felipe Magno de Almeida wrote:
Hello Jonathan,
Sorry for the delay, I was in mid-vacation.
Comments are inline.
On Tue, Dec 6, 2016 at 3:45 PM, Jonathan Wakely <jwak...@redhat.com> wrote:
On 16/09/16 02:53 -0300, Felipe Magno de Almeida wrote:
[snip]
I've tried both approaches. Templates were causing problems of not
defined instantations because they were being used as ints too
in other _M_extract functions through a tmp integer. And typedef's
caused the same problem of having to use a tmp value of the right
type but for example _M_extract_wday_or_month could not have the
same type (in AVR they do) and I'd have to use a temporary anyway
then.
This was the least intrusive way.
Did you consider something like this?
I have, but I did not like how it could run code that is not very explicit
in case of exceptions, possibly assigning from a non-initialized variable,
OK, but that assignment from a possibly-uninitialized variable was
also present in your original patch.
causing, theoretically UB. Besides, __mem was already used for
tm_mom, so it seemed best to just mimic for the rest.
Yes, but now we have that overhead of extra reads and writes for every
function, not just that one.
I have another patch attached which fixes the problem you mentioned
and the same problem with the tm_wday in a different switch-case.
I've removed all initializations of __mem, the same way that tm_mon
already did not initialize it, because the value is not actually used at
all in extract_num and variants.
Doesn't this now mean that if an error happens in the _M_extract_xxx
function we set tm.tm_xxx to garbage?
The existing tm.tm_mon case handles that correctly, by checking for an
error first.
I have ran the test and nothing on locale failed, I've ran again with the fixes,
and there are also no errors on locale. Testing with x86_64.
Good, thanks for testing it.
