Skip to content

[BZ-2434] ACE_Time_Value operations near limit/saturation fail on 64-bit #2434

Open
Open
[BZ-2434] ACE_Time_Value operations near limit/saturation fail on 64-bit#2434
Labels
component:ACE Corepriority:P3product:ACEseverity:normal

Description

@jwillemsen
jwillemsen
opened on Mar 28, 2026
Field Value
Bugzilla ID 2434
Reporter @shuston
Assigned to DOC Center Support List (internal)
Product ACE
Component ACE Core
Version 5.6.2
Platform / OS All / All
Priority P3
Severity normal
Status REOPENED
Resolution
Created 2006-03-03 16:11:31 -0600
Depends on #2563

Originally posted by @shuston on 2006-03-03 16:11:31 -0600

ACE_Time_Value test fails on 64-bit systems (other than Windows, which was
explicitly worked around for other reasons). The operations that manipulate
values, such as operator*, will fail for values near or at the range of a 64-bit
integer.
The root cause is that the operations are done using "double" floating point
operations and a 'double' on most of these systems is 64 bits and, thus, has a
value range less than that of a 64-bit integer.

This report is to capture thinking on this subject so it can be fixed.

Some notes extracted from email regarding this subject:

From J.T. Conklin:

I think the way to do this is to do this in multiprecision integer
arithmetic. The simplest way to do this would be to add/import an
arbitrary precision integer library, but that would probably bloat
the footprint to an undesirable degree. The alternative is to do
it in line, shouldn't take more than a evening with Seminumerical
Algorithms to work it out.

From Steve Huston:

FYI, there's a GNU MP library that can do what we need, I believe. It's LGPL
though (of course).

AIX and Solaris have a mp(3) library that can do mp integer math, but I haven't
found any floating point that works with it.

Rich Seibel also had input that ACE_Time_Value should be split into a relative
and an absolute time class; Carlos O'Ryan suggested looking at how boost does
this for ideas.

Activity

  1. jwillemsen
    added
    severity:normal
    priority:P3
    product:ACE
    component:ACE Core
    on Mar 28, 2026

  2. jwillemsen commented on Mar 28, 2026

    @jwillemsen
    jwillemsen
    on Mar 28, 2026
    MemberAuthor

    Originally posted by @jwillemsen on 2006-06-21 14:33:58 -0500 (Bugzilla comment 1)

    added depends

  3. jwillemsen commented on Mar 28, 2026

    @jwillemsen
    jwillemsen
    on Mar 28, 2026
    MemberAuthor

    Originally posted by @jwillemsen on 2006-08-09 09:14:37 -0500 (Bugzilla comment 2)

    accept for tao-support

  4. jwillemsen commented on Mar 28, 2026

    @jwillemsen
    jwillemsen
    on Mar 28, 2026
    MemberAuthor

    Originally posted by @jwillemsen on 2007-01-08 14:20:37 -0600 (Bugzilla comment 3)

    *** Bug #2563 has been marked as a duplicate of this bug. ***

  5. jwillemsen commented on Mar 28, 2026

    @jwillemsen
    jwillemsen
    on Mar 28, 2026
    MemberAuthor

    Originally posted by @jwillemsen on 2007-03-07 15:32:18 -0600 (Bugzilla comment 4)

    Ossama committed a few fixes yesterday and today, we should watch the test stats
    and see if things are improved

  6. jwillemsen commented on Mar 28, 2026

    @jwillemsen
    jwillemsen
    on Mar 28, 2026
    MemberAuthor

    Originally posted by @jwillemsen on 2008-02-18 09:04:06 -0600 (Bugzilla comment 5)

    not resolved

    Thu Mar 9 09:44:14 UTC 2007 Ossama Othman <ossama_othman at symantec dot com>

        * ace/Time_Value.cpp (operator *=):

      Reverted bulk of the changes.  New implementation will be
      revisited as time permits.

    Thu Mar 9 07:44:14 UTC 2007 Ossama Othman ossama_othman@symantec.com

        * tests/Time_Value_Test.cpp (run_main):

      Simplified assertions.  Rather than test equality of sec() and
      usec() components separately, just test equality of the
      ACE_Time_Value objects as a whole.

    Thu Mar 8 18:06:14 UTC 2007 Ossama Othman <ossama_othman at symantec dot com>

        * ace/Time_Value.cpp (operator *=):

      Removed vestigial time value saturation code that does not work
      correctly with latest implementation.

    Wed Mar 7 17:32:51 UTC 2007 Ossama Othman <ossama_othman at symantec dot com>

        * ace/Time_Value.cpp (operator*=):

      Altered implementation to avoid converting the microseconds
      component of the ACE_Time_Value to a fraction of a second.  In
      the case were the floating point type's mantissa width is the
      same as the width of the time_t (seconds) value, a maxed out
      ACE_Time_Value seconds component would utilize all bits in the
      mantissa, leaving no room for a fraction.  The microseconds
      component would end up not contributing to the multiplication.
      Multiply the seconds and microseconds components separately
      instead.  This also saves us a floating point division, which
      can be expensive.

    Tue Mar 6 23:06:15 UTC 2007 Ossama Othman <ossama_othman at symantec dot com>

        * ace/Time_Value.cpp (operator*=):

      Use long double for internal computations if double is not large
      enough to retain precision of the time_t/suseconds_t values.
      This will only provide improved behavior with compilers that
      support double-extended precision floating numbers, i.e. >= 79
      bits.

      Added comments to explain what is being done in the
      implementation.

  7. jwillemsen commented on Mar 28, 2026

    @jwillemsen
    jwillemsen
    on Mar 28, 2026
    MemberAuthor

    Originally posted by @jwillemsen on 2010-01-21 02:20:05 -0600 (Bugzilla comment 6)

    Thu Jan 21 08:13:30 UTC 2010 Johnny Willemsen jwillemsen@remedy.nl

        * tests/Time_Value_Test.cpp:
    * tests/tests.mpc:
    * tests/run_test.lst:
    * tests/Bug_2434_Regression_Test.cpp:
      Moved the 64bit overflow test which fails on 64bit systems to
      its own file and marked that as not fixed

  8. jwillemsen commented on Mar 28, 2026

    @jwillemsen
    jwillemsen
    on Mar 28, 2026
    MemberAuthor

    Originally posted by Olivier Langlois on 2010-01-27 10:42:18 -0600 (Bugzilla comment 7)

    Created attachment 1245
    svn diff from the trunk to fix the bug.

    Field Value
    Attachment ID 1245
    Filename bug2434_bug3818.patch
    Description svn diff from the trunk to fix the bug.
    Uploaded by Olivier Langlois
    Date 2010-01-27 10:42:18 -0600
    Size 13350 bytes
    Patch yes
    Obsolete no
    Download Download from repository

  9. jwillemsen commented on Mar 28, 2026

    @jwillemsen
    jwillemsen
    on Mar 28, 2026
    MemberAuthor

    Originally posted by Douglas C. Schmidt on 2010-02-21 15:35:35 -0600 (Bugzilla comment 8)

    Fixed, thanks!

  10. jwillemsen commented on Mar 28, 2026

    @jwillemsen
    jwillemsen
    on Mar 28, 2026
    MemberAuthor

    Originally posted by @shuston on 2010-02-23 16:03:03 -0600 (Bugzilla comment 9)

    Still broken; Bug_2434_Regression_Test and Time_Value_Test are still failing.

  11. jwillemsen commented on Mar 28, 2026

    @jwillemsen
    jwillemsen
    on Mar 28, 2026
    MemberAuthor

    Originally posted by @jwillemsen on 2010-02-25 04:32:28 -0600 (Bugzilla comment 10)

    seems to work on a lot of 64bit systems, but 64bit windows is failing

  12. jwillemsen commented on Oct 1, 2026

    @jwillemsen
    jwillemsen
    on Oct 1, 2026
    MemberAuthor

    DOCGroup/ACE_TAO#2598

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    component:ACE Corepriority:P3product:ACEseverity:normal

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions