Re: Time_Span divide question

Re: Time_Span divide question

[Mark D. McKinney]

>  D1 := 0.495;
>  D2 := 0.500;
>  Put_Line ("Duration divide result: " & Integer'Image(Integer(D1/D2)));
   D1 / D2 = 0.990
   Integer(D1/D2) = 1
>
>  TS1 := To_Time_Span (D1);
>  TS2 := To_Time_Span (D2);
>  Put_Line ("Time_Span divide result: " & Integer'Image(TS1/TS2));

To_Time_Span(D1) = Integer(D1) = 0
To_Time_Span(D2) = Integer(D2) = 1
TS1 / TS2 = 0

Yes they should be different.

Careful where rounding occurs.

Re: Time_Span divide question

[Tucker Taft]

NoJunkMailPlease wrote:

: Should these print the same value?

:   D1 := 0.495;
:   D2 := 0.500;
:   Put_Line ("Duration divide result: " & Integer'Image(Integer(D1/D2)));

The conversion form fixed point to integer requires rounding,
so this will certainly give 1.

:   TS1 := To_Time_Span (D1);
:   TS2 := To_Time_Span (D2);
:   Put_Line ("Time_Span divide result: " & Integer'Image(TS1/TS2));

Integer division is truncating, and hence so is Time_Span divide, so 
this will give zero.

: It is not clear to me what the RM requires of the divide operator for
: two Time_Span types returning an integer.  It says they should act the
: same as integer operators.

So they are required to truncate, like integer divide.

: The compiler I'm using prints 1 for the Duration divide and zero for
: the Time_Span divide.  Correct or not correct?

Correct.

: I suspect that is correct, but why is there no divide returning a Time_Span?

There is.  But it represents the result of dividing a time span by
an integer.  Dividing a time span by a time span is clearly producing
a unitless, non-time value.  (E.g. seconds/seconds = unitless).

: What about "rem"?

This can be computed by multiplying the quotient by the divisor and
subtracting it from the dividend.  In retrospect, it would probably have 
made sense to include it and "mod" just for completeness, but probably because
we were comparing with the operators on Duration, there was no
dicussion of "rem" or "mod."

: I guess to obtain the same functionality you have to add 0.5 to the
: numerator.  Any comments on why it was designed this way?

Presuming you want rounding, then you will have to add 0.5 times the
divisor before dividing.  However, one of the primary goals of Time_Span
was to get away from the rounding and general vagueries of fixed-point
divide, so that the programmer has full control.  You can think of
Time_Span as simply a count of ticks, and treat it accordingly.

-Tucker Taft   stt@inmet.com   http://www.inmet.com/~stt/
Intermetrics, Inc.  Burlington, MA  USA

Re: Time_Span divide question

[Stephen Leake]

Mark D. McKinney wrote:
> 
> >  D1 := 0.495;
> >  D2 := 0.500;
> >  Put_Line ("Duration divide result: " & Integer'Image(Integer(D1/D2)));
>    D1 / D2 = 0.990
>    Integer(D1/D2) = 1
> >
> >  TS1 := To_Time_Span (D1);
> >  TS2 := To_Time_Span (D2);
> >  Put_Line ("Time_Span divide result: " & Integer'Image(TS1/TS2));
> 
> To_Time_Span(D1) = Integer(D1) = 0
> To_Time_Span(D2) = Integer(D2) = 1

Um, no. Time_Span is an integer type, but it is scaled, so that 1 is one
Time_Unit, which is a small fraction of seconds. In GNAT, Time_Unit is
one nanosecond. So To_Time_Span (foo) is NOT the same as Integer (foo).

> TS1 / TS2 = 0
> 
> Yes they should be different.
> 
> Careful where rounding occurs.

Here's a compilable test:

with Text_IO; use Text_IO;
with Ada.Real_Time; use Ada.Real_Time;
with Unchecked_Conversion;
with Interfaces;
procedure Test
is
   D1, D2 : Duration;
   TS1, TS2 : Time_Span;

   function To_Int_64 is new Unchecked_Conversion
      (Source => Time_Span,
       Target => Interfaces.Integer_64);
begin
   D1 := 0.495;
   D2 := 0.500;
   Put_Line ("Duration divide result: " &
Integer'Image(Integer(D1/D2)));

   TS1 := To_Time_Span (D1);
   TS2 := To_Time_Span (D2);
   Put_Line (" Int_64 (TS1) => " & Interfaces.Integer_64'Image
(To_Int_64 (TS1)));
   Put_Line (" Int_64 (TS2) => " & Interfaces.Integer_64'Image
(To_Int_64 (TS2)));

   Put_Line ("Time_Span divide result: " & Integer'Image(TS1/TS2));

end Test;

Note that your compiler may have a different representation for
Time_Span.

Running the code on GNAT 3.09, under Windows 95 on a Pentium, I get:

Duration divide result:  0
 Int_64 (TS1) =>  495000000
 Int_64 (TS2) =>  500000000
Time_Span divide result:  0

Integer division truncates, in order to satisfy the rules in RM 4.5.5 (1
.. 5). Fixed point division goes up or down; see RM 4.5.5(21)
-- 
- Stephe

Re: Time_Span divide question

[Robert A Duff]

In article <337BD41B.2FE1@mail.us.net>,
Mark D. McKinney <mckmark@mail.us.net> wrote:
>>  D1 := 0.495;
>>  D2 := 0.500;

>To_Time_Span(D1) = Integer(D1) = 0
>To_Time_Span(D2) = Integer(D2) = 1

No, that's not right.  Time_Span is an integer count of Time_Span_Units,
which is less than or equal to 20 microseconds.  E.g. if Time_Span_Unit
= 0.000_010, then To_Time_Span(D2) will be approx 50_000
Time_Span_Units, which is half a second.

- Bob

Re: Time_Span divide question

[Tucker Taft]

Stephen Leake (Stephen.Leake@gsfc.nasa.gov) wrote:

: ...
: Here's a compilable test:

: with Text_IO; use Text_IO;
: with Ada.Real_Time; use Ada.Real_Time;
: with Unchecked_Conversion;
: with Interfaces;
: procedure Test
: is
:    D1, D2 : Duration;
:    TS1, TS2 : Time_Span;

:    function To_Int_64 is new Unchecked_Conversion
:       (Source => Time_Span,
:        Target => Interfaces.Integer_64);
: begin
:    D1 := 0.495;
:    D2 := 0.500;
:    Put_Line ("Duration divide result: " &
: Integer'Image(Integer(D1/D2)));

:    TS1 := To_Time_Span (D1);
:    TS2 := To_Time_Span (D2);
:    Put_Line (" Int_64 (TS1) => " & Interfaces.Integer_64'Image
: (To_Int_64 (TS1)));
:    Put_Line (" Int_64 (TS2) => " & Interfaces.Integer_64'Image
: (To_Int_64 (TS2)));

:    Put_Line ("Time_Span divide result: " & Integer'Image(TS1/TS2));

: end Test;

: Note that your compiler may have a different representation for
: Time_Span.

: Running the code on GNAT 3.09, under Windows 95 on a Pentium, I get:

: Duration divide result:  0

That seems wrong.  Unless the 'small for Duration is extremely large (i.e.
>= 1.0), D1/D2 should be just slightly less than 1 as a universal_fixed 
value (the result of fixed-fixed division), and when converted to Integer, 
it should be rounded (per RM95 4.6(33)).

:  Int_64 (TS1) =>  495000000
:  Int_64 (TS2) =>  500000000
: Time_Span divide result:  0

This one looks right.

: Integer division truncates, in order to satisfy the rules in RM 4.5.5 (1
: .. 5). Fixed point division goes up or down; see RM 4.5.5(21)

That is a bit misleading.  Fixed-point division goes up or down
when converting the universal_fixed result of division to an ordinary
fixed-point value (G.2.3(10)).  However, when converting to an 
integer value, it should be rounded to the nearest integer (away 
from zero if a tie).  See G.2.3(18) and 4.6(33).

: -- 
: - Stephe

-Tucker Taft   stt@inmet.com   http://www.inmet.com/~stt/
Intermetrics, Inc.  Burlington, MA  USA

Re: Time_Span divide question

[Robert Dewar]

Tuck says

<<: Duration divide result:  0

That seems wrong.  Unless the 'small for Duration is extremely large (i.e.
>= 1.0), D1/D2 should be just slightly less than 1 as a universal_fixed
value (the result of fixed-fixed division), and when converted to Integer,
it should be rounded (per RM95 4.6(33)).>>

That's definitely wrong. Using the latest version of GNAT on the program,
the output is:

Duration divide result:  1
 Int_64 (TS1) =>  495000000
 Int_64 (TS2) =>  500000000
Time_Span divide result:  0

Re: Time_Span divide question

[Wayne]

Tucker Taft writes:
>
>: Should these print the same value?
>
>:   D1 := 0.495;
>:   D2 := 0.500;
>:   Put_Line ("Duration divide result: " & Integer'Image(Integer(D1/D2)));
>
>The conversion form fixed point to integer requires rounding,
>so this will certainly give 1.

I agree.  Someone told me that GNAT generates a zero.  Must be a bug in
GNAT (I think it was the NT one).

>: I suspect that is correct, but why is there no divide returning a Time_Span?
>
>There is.  But it represents the result of dividing a time span by
>an integer.  Dividing a time span by a time span is clearly producing
>a unitless, non-time value.  (E.g. seconds/seconds = unitless).

A Duration can be divided by a Duration.  Sometimes you want a unitless
result.  I want to divide a time by a period to get the number of cycles.

>: What about "rem"?
>
>This can be computed by multiplying the quotient by the divisor and
>subtracting it from the dividend.  In retrospect, it would probably have 
>made sense to include it and "mod" just for completeness, but probably because
>we were comparing with the operators on Duration, there was no
>dicussion of "rem" or "mod."

They should be added to the next version of Ada.Real_Time, don't you think?
This is very minor, but would be nice for completeness.

Re: Time_Span divide question

[Kevin Radke]

In article <dewar.864099351@merv>, Robert Dewar <dewar@merv.cs.nyu.edu> wrote:
>Tuck says
>
><<: Duration divide result:  0
>
>That seems wrong.  Unless the 'small for Duration is extremely large (i.e.
>>= 1.0), D1/D2 should be just slightly less than 1 as a universal_fixed
>value (the result of fixed-fixed division), and when converted to Integer,
>it should be rounded (per RM95 4.6(33)).>>
>
>That's definitely wrong. Using the latest version of GNAT on the program,
>the output is:
>
>Duration divide result:  1
> Int_64 (TS1) =>  495000000
> Int_64 (TS2) =>  500000000
>Time_Span divide result:  0


Hmmm... 3.07 on NT was definitely broken then since it
gave 0.

Kevin

Re: Time_Span divide question

[Robert Dewar]

Kevin wrote, responding to me

<<>That's definitely wrong. Using the latest version of GNAT on the program,
>the output is:
>
>Duration divide result:  1
> Int_64 (TS1) =>  495000000
> Int_64 (TS2) =>  500000000
>Time_Span divide result:  0


Hmmm... 3.07 on NT was definitely broken then since it
gave 0.

Kevin>>


First, there is no 3.07 on NT, perhaps you are talking about 3.09 on NT
or the 3.07 on DOS (perhaps running on NT?) Anyway, that may well be.
The correct version above is what I got from GNAT 3.10, which is the
current version available to customers. We expect a code freeze for
version 3.10 quite soon, and a public release of 3.10 will occur some
time in the future.

Robert Dewar
Ada Core Technologies

Re: Time_Span divide question

[Kevin Radke]

In article <dewar.864583354@merv>, Robert Dewar <dewar@merv.cs.nyu.edu> wrote:
>Kevin wrote, responding to me
>
><<>That's definitely wrong. Using the latest version of GNAT on the program,
>>the output is:
>>
>>Duration divide result:  1
>> Int_64 (TS1) =>  495000000
>> Int_64 (TS2) =>  500000000
>>Time_Span divide result:  0
>
>
>Hmmm... 3.07 on NT was definitely broken then since it
>gave 0.
>
>Kevin>>
>
>
>First, there is no 3.07 on NT, perhaps you are talking about 3.09 on NT
>or the 3.07 on DOS (perhaps running on NT?) Anyway, that may well be.
>The correct version above is what I got from GNAT 3.10, which is the
>current version available to customers. We expect a code freeze for
>version 3.10 quite soon, and a public release of 3.10 will occur some
>time in the future.

Alas, Robert is correct (as usual).  It seems I was using 3.09 on
NT (The last public release on NT), and it is the one that returns
the incorrect results.  I look forward for 3.10...

Kevin