ANN: GNAT 3.15p binary for OS/2 available

ANN: GNAT 3.15p binary for OS/2 available

[Dave Parsons]

Following the release of GNAT 3.15p by ACT last week, I have
now compiled a binary for OS/2 V4 / eCS.

This is again based upon EMX 9d with GCC 2.8.1.
See Arnaud Charlet's post and the readme for more details.

It is currently available as

http://dwparsons.bei.t-online.de/gnat-3.15p-os2-bin-20021124.zip

and it is expected that it will be mirrorred by cs.nyu.edu in
ftp://cs.nyu.edu/pub/gnat/3.15p/contrib/os2

David Parsons

Re: ANN: GNAT 3.15p binary for OS/2 available

[tmoran]

If it uses the same run-time library s-osprim.adb and s-taprop.adb, then
it has the same bug: Monotonic_Clock isn't monotonic, so a "delay" can
be much longer than it should be.  I believe moving the line
   Base_Monotonic_Clock := Base_Clock;
from being the last statement in s-osprim.adb, to instead be the last
statement in the routine Get_Base_Time, will solve the problem.
  I sent a bug report to report@gnat.com - is that in general sufficient
to inform those who might need to know for non-ACT ports?

Re: ANN: GNAT 3.15p binary for OS/2 available

[Dave Parsons]

On Sun, 24 Nov 2002 20:07:02 UTC, tmoran@acm.org wrote:

> If it uses the same run-time library s-osprim.adb and s-taprop.adb, then
> it has the same bug: Monotonic_Clock isn't monotonic, so a "delay" can
> be much longer than it should be.  I believe moving the line
>    Base_Monotonic_Clock := Base_Clock;
> from being the last statement in s-osprim.adb, to instead be the last
> statement in the routine Get_Base_Time, will solve the problem.
>   I sent a bug report to report@gnat.com - is that in general sufficient
> to inform those who might need to know for non-ACT ports?

I can only find 'Base_Monotonic_Clock := Base_Clock;' in 5wosprim.adb
which is for NT and not used for OS/2. So, OS/2 would appear at first
glance not to have that bug but do you have a number for the bug report
so I can check it out in a bit more detail.

Dave

Re: ANN: GNAT 3.15p binary for OS/2 available

[tmoran]

> I can only find 'Base_Monotonic_Clock :=3D Base_Clock;' in 5wosprim.adb
> which is for NT and not used for OS/2. So, OS/2 would appear at first
> glance not to have that bug
  Excellent!  The problem lay in attempting to keep the two Windows clocks
in synch, and one monotonically increasing.  If one task called Clock, and
that caused a resynch, then Base_Ticks, but not Base_Monotonic_Clock,
would be reset, thus causing a backward movement of Monotonic_Clock, and a
consequent increase in any other task's outstanding "delay"s.  Perhaps OS2
doesn't have such a mess.  The symptom was that in a multitasking program
where one task called Clock, and one of the clocks was adjusted manually or
automatically, another task's "delay" would be markedly lengthened.
> do you have a number for the bug report
  I just sent it in Saturday.

Re: ANN: GNAT 3.15p binary for OS/2 available

[Dave Parsons]

On Mon, 25 Nov 2002 08:08:22 UTC, tmoran@acm.org wrote:

> > I can only find 'Base_Monotonic_Clock :=3D Base_Clock;' in 5wosprim.adb
> > which is for NT and not used for OS/2. So, OS/2 would appear at first
> > glance not to have that bug
>   Excellent!  The problem lay in attempting to keep the two Windows clocks
> in synch, and one monotonically increasing.  If one task called Clock, and
> that caused a resynch, then Base_Ticks, but not Base_Monotonic_Clock,
> would be reset, thus causing a backward movement of Monotonic_Clock, and a
> consequent increase in any other task's outstanding "delay"s.  Perhaps OS2
> doesn't have such a mess.  The symptom was that in a multitasking program
> where one task called Clock, and one of the clocks was adjusted manually or
> automatically, another task's "delay" would be markedly lengthened.
> > do you have a number for the bug report
>   I just sent it in Saturday.

Ok thanks for the explanation. I'll have a look at the OS/2 implementation
in a bit more detail to see if it could suffer from a similar problem.

Dave