[Bug libstdc++/89942] New: std::function __is_location_invariant breaks ABI compatibility
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89942
Bug ID: 89942
Summary: std::function __is_location_invariant breaks ABI
compatibility
Product: gcc
Version: 5.4.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: libstdc++
Assignee: unassigned at gcc dot gnu.org
Reporter: kan.liu.229 at gmail dot com
Target Milestone: ---
Created attachment 46079
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=46079=edit
possible fix
in gcc 5.4.0
template
struct __is_location_invariant
: is_trivially_copyable<_Tp>::type
{ };
__is_location_invariant is evaluated to be true for the closure type (lambda
expression)
in gcc 4.9.2
template
struct __is_location_invariant
: integral_constant::value
|| is_member_pointer<_Tp>::value)>
{ };
which is evaluated to be false.
This may break ABI compatibility when pass function object between libs
compiled by 4.9.2 / 5.4.0 gcc.
The attachment is a possible fix.
[Bug libstdc++/61217] New: pop_heap can't guarantee At most 2 * log(last - first) comparisons
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61217 Bug ID: 61217 Summary: pop_heap can't guarantee At most 2 * log(last - first) comparisons Product: gcc Version: 4.9.0 Status: UNCONFIRMED Severity: minor Priority: P3 Component: libstdc++ Assignee: unassigned at gcc dot gnu.org Reporter: kan.liu.229 at gmail dot com pop_heap uses __adjust_heap to percolate down the hole. Inside __adjust_heap, it uses __value stores the value where was in position __result(__last). After percolating down the hole, it uses __push_heap to push back the __value. Process of percolatiing down takes 2 * logN comparasion, and __push_heap takes another 2 * logN, 4 * logN in total, which excceed the requirement 2 * log(last - first) comparisons of standard.
[Bug libstdc++/61166] overflow when parse number in std::duration operator
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61166 --- Comment #12 from Kan Liu kan.liu.229 at gmail dot com --- (In reply to Jonathan Wakely from comment #11) (In reply to Kan Liu from comment #10) _Select_type already does the overflow check, so *template implemented operators* is still redundant I think. You can't use _Select_type on a literal operator that is not a template. Since the standard units (std::hours, milliseconds ...) requires the duration::rep to be *int64_t*, No it doesn't, only chrono::nanoseconds is required to have a 64-bit representation. chrono::hours and chrono::minutes are allowed to have fewer than 32 bits (at least 23 bits and at least 29 bits respectively). there's no need to do too much check. Not all values of uint64_t fit in int64_t Wow... I see
[Bug libstdc++/61166] overflow when parse number in std::duration operator
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61166 --- Comment #10 from Kan Liu kan.liu.229 at gmail dot com --- (In reply to Jonathan Wakely from comment #9) I can see one good reason to implement those operators as templates: it allows us to check if the literal value overflows the duration::rep type, which is required by the standard but impossible to implement without using templates. _Select_type already does the overflow check, so *template implemented operators* is still redundant I think. Since the standard units (std::hours, milliseconds ...) requires the duration::rep to be *int64_t*, there's no need to do too much check.
[Bug libstdc++/61166] overflow when parse number in std::duration operator
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61166
--- Comment #7 from Kan Liu kan.liu.229 at gmail dot com ---
(In reply to emsr from comment #6)
Something like parse_number was in the original doc as an implementation
example. The idea was to select the smallest integral type that could
accommodate the number string with. This is done with _Select_int.
_Select_type picks the most space-efficient duration ratio type using
_Select_int.
Yeah, but it only helps saving space when compiling time for now. For now
implementation, _Select_type takes the default chrono::hour (and other time
units), which uses *int64_t* as its underlying representation. So, it doesn't
save space I think.
Take *ms* as example
Implementation now:
template char... _Digits
constexpr typename
__select_type::_Select_type__select_int::_Select_int_Digits...::value,
chrono::milliseconds::type
operatorms()
{
return __select_type::_Select_type
__select_int::_Select_int_Digits...::value,
chrono::milliseconds::value;
}
space-efficient version (maybe), get the type from _Select_int:
template char... _Digits
constexpr typename
chrono::durationdecltype(__select_int::_Select_int_Digits...::value), milli
operator_ms()
{
typedef typename __select_int::_Select_int_Digits... __selected;
typedef typename chrono::durationdecltype(__selected::value), milli
__duration_type;
return __duration_type {__selected::value};
}
[Bug libstdc++/61166] New: overflow when parse number in std::duration operator
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61166
Bug ID: 61166
Summary: overflow when parse number in std::duration operator
Product: gcc
Version: 4.9.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: libstdc++
Assignee: unassigned at gcc dot gnu.org
Reporter: kan.liu.229 at gmail dot com
overflow may occur when using a number whose size exceeds sizeof(unsigned) in
chrono_literals. Because the parse classes in include/bits/parse_number.h just
use *unsigned* for member type. However, it should accept *unsigned long long*
I think. This issue also exists in trunk I think.
#include chrono
using namespace std;
using namespace std::chrono;
int main() {
// std::numeric_limitsunsigned::max() == 4294967295
cout 429496729510 endl;
cout (429496729510h).count() endl;
return 0;
}
And the result is
429496729510
926648584
[Bug libstdc++/61166] overflow when parse number in std::duration operator
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61166 --- Comment #3 from Kan Liu kan.liu.229 at gmail dot com --- (In reply to emsr from comment #2) Created attachment 32789 [details] Patch to parse_nmber to make it unsigned long long all over. Works on x86_64-linux. Yeah, it works. Thank you!
[Bug libstdc++/61166] overflow when parse number in std::duration operator
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61166 --- Comment #4 from Kan Liu kan.liu.229 at gmail dot com --- btw, is it really necessary to use functionality in parse_number.h to parse manually? What *parse_number* has done is no more than the general *operator(unsigned long long)*, and it just enables *0b* and *0B* prefix for the numbers. I think *operator(unsigned long long)* is quite enough for this.
