[ace-bugs] ACE_Time_Value: set(double) hangs when setting large values
Markus Hossner
markus.hossner at innovative-navigation.de
Mon Feb 22 10:49:01 CST 2016
ACE VERSION: 6.3.3
HOST MACHINE and OPERATING SYSTEM:
Red Hat Enterprise Linux Server release 6.5 (Santiago) 64 Bit
TARGET MACHINE and OPERATING SYSTEM, if different from HOST:
Red Hat Enterprise Linux Server release 6.5 (Santiago) 64 Bit
gcc version 4.4.7 20120313 (Red Hat 4.4.7-4) (GCC)
THE $ACE_ROOT/ace/config.h FILE
#include "ace/config-linux.h"
THE $ACE_ROOT/include/makeinclude/platform_macros.GNU FILE
include $(ACE_ROOT)/include/makeinclude/platform_linux.GNU
INSTALL_PREFIX=/usr/local/ACE+TAO-6.3.1
LD_RPATH_FLAGS += -Wl,-rpath,/usr/local/in/lib
CONTENTS OF $ACE_ROOT/bin/MakeProjectCreator/config/default.features
does not exist
AREA/CLASS/EXAMPLE AFFECTED:
ACE_Time_Value
DOES THE PROBLEM AFFECT:
EXECUTION
SYNOPSIS:
ACE_Time_Value: set(double) hangs when setting large values (e.g. DBL_MAX) on 64 bit platform
DESCRIPTION:
Through an error in our code an ACE_Time_Value was set to DBL_MAX which resulted in the program hanging in
ACE_Time_Value::normalize(). I looked at the code and found out that the cast from double to t_time went wrong
which results in a very large usec value (64 bit). Then normalize() is called and normalize() uses a loop to adjust the
sec and usec values. For large usec values this needs very long time.
Problematic usec values can also be set directly by the constructor or a set method.
REPEAT BY:
In a 64 Bit environment, where suseconds_t is defined as 64 Bit value.
cout << sizeof(suseconds_t) << endl; // 8
ACE_Time_Value a(0, LONG_MAX); // hangs
ACE_Time_Value b;
b.set(0, LONG_MAX); // hangs
ACE_Time_Value c;
c.set(DBL_MAX); //hangs
(For repeating the problem I used extreme values but this also happens for other large numbers)
Best regards
Markus Hossner
SAMPLE FIX/WORKAROUND:
ACE_Time_Value::set (double d): ace/Time_Value.inl
- Handle big positiv and negativ double values by limiting them to possible values for sec and usec
- Correct round for negativ double values: -10.5 => -10 -500000 (not -10 -499999)
- no normalize needed cause result is already normalized
ACE_INLINE void
ACE_Time_Value::set (double d)
{
// ACE_OS_TRACE ("ACE_Time_Value::set");
if (d < ACE_Numeric_Limits<time_t>::min())
{
this->tv_.tv_sec = ACE_Numeric_Limits<time_t>::min();
this->tv_.tv_usec = -ACE_ONE_SECOND_IN_USECS + 1;
}
else if (d > ACE_Numeric_Limits<time_t>::max())
{
this->tv_.tv_sec = ACE_Numeric_Limits<time_t>::max();
this->tv_.tv_usec = ACE_ONE_SECOND_IN_USECS - 1;
}
else
{
time_t l = (time_t) d;
this->tv_.tv_sec = l;
this->tv_.tv_usec = (suseconds_t) ((d - (double) l) * ACE_ONE_SECOND_IN_USECS + (d < 0 ? -0.5 : 0.5));
}
}
ACE_Time_Value::normalize (bool saturate): ace/Time_Value.cpp
- Better performance by not using a loop for adjusting time values
void
ACE_Time_Value::normalize (bool saturate)
{
// ACE_OS_TRACE ("ACE_Time_Value::normalize");
if (this->tv_.tv_usec >= ACE_ONE_SECOND_IN_USECS ||
this->tv_.tv_usec <= -ACE_ONE_SECOND_IN_USECS)
{
time_t sec = abs(this->tv_.tv_usec) / ACE_ONE_SECOND_IN_USECS * (this->tv_.tv_usec > 0 ? 1 : -1);
suseconds_t usec = this->tv_.tv_usec - sec * ACE_ONE_SECOND_IN_USECS;
if (saturate && this->tv_.tv_sec > 0 && sec > 0 &&
ACE_Numeric_Limits<time_t>::max() - this->tv_.tv_sec < sec)
{
this->tv_.tv_sec = ACE_Numeric_Limits<time_t>::max();
this->tv_.tv_usec = ACE_ONE_SECOND_IN_USECS - 1;
}
else if (saturate && this->tv_.tv_sec < 0 && sec < 0 &&
ACE_Numeric_Limits<time_t>::min() - this->tv_.tv_sec > sec)
{
this->tv_.tv_sec = ACE_Numeric_Limits<time_t>::min();
this->tv_.tv_usec = -ACE_ONE_SECOND_IN_USECS + 1;
}
else
{
this->tv_.tv_sec += sec;
this->tv_.tv_usec = usec;
}
}
if (this->tv_.tv_sec >= 1 && this->tv_.tv_usec < 0)
{
--this->tv_.tv_sec;
this->tv_.tv_usec += ACE_ONE_SECOND_IN_USECS;
}
// tv_sec in qnxnto is unsigned
#if !defined ( __QNX__)
else if (this->tv_.tv_sec < 0 && this->tv_.tv_usec > 0)
{
++this->tv_.tv_sec;
this->tv_.tv_usec -= ACE_ONE_SECOND_IN_USECS;
}
#endif /* __QNX__ */
}
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.isis.vanderbilt.edu/pipermail/ace-bugs/attachments/20160222/71cc43f3/attachment.html>
More information about the ace-bugs
mailing list