(j3.2006) (SC22WG5.5736) Units of measure

John Reid John.Reid
Wed Jun 29 07:20:33 EDT 2016


Here is a slightly changed draft N2112. A diff file is attached.

Van asks

1. Was the sentiment expressed in the penultimate paragraph that of a 
significant fraction of those you asked for their reasons for opposing 
the project?

2. Did you ask whether my offer to remove the promise to incorporate the
specification into a future revision of the standard made a difference
in their positions?

For all those that attended the London meeting, I would appreciate your 
thoughts on this.

I think I should perhaps add a paragraph on 2. I think the sentiment was 
that it would obviate the whole point of a TS - to define a feature that 
WG5 intended eventually to include in the standard.



John Reid wrote:
> Dear all,
> I was required by the London meeting to provide a document describing
> the reasons why WG5 did not apply for a new project to develop a TS on
> Units of Measure in Fortran when requested to do so at its 2013 meeting.
> Here is a draft. I would like to thank Dan for writing a first draft at
> the Boulder meeting. This was shown to Van and the subgroup heads and
> their comments were taken into account.
> The obvious reason for the decision was that 7 people voted "no". This
> paper has to explore why they voted so. Have I got this right? Comments,
> please.
> Cheers,
> John.
> _______________________________________________
> J3 mailing list
> J3 at mailman.j3-fortran.org
> http://mailman.j3-fortran.org/mailman/listinfo/j3
-------------- next part --------------
<                                          ISO/IEC JTC1/SC22/WG5 N2112-1
>                                          ISO/IEC JTC1/SC22/WG5 N2112-2
< error, there did not appear to be wide-spread demand for this feature.  
> error, and other cases where such a feature would have been useful were 
> mentioned, there did not appear to be wide-spread demand for this feature.  
-------------- next part --------------
                                         ISO/IEC JTC1/SC22/WG5 N2112-2

      WG5 decision on a new work item on a TS on Units of Measure

                      John Reid, 19 June 2016

The UNITS proposal that was brought to WG5 at the meeting in 2013, see
N1969 and N1970, was intended to provide assistance to programmers who 
must manage sets of physical units used in engineering analysis programs.  
The specific case cited was the loss of a spacecraft that was intended 
to explore Mars.  The basic issue was the use of different sets of units 
(in this case, Imperial versus SI) in different programs run by different 
organizations. This discrepancy was not noticed until too late and 
resulted in the loss of the spacecraft.

While the sequence of events above describes a particular case where a 
units checking package might have assisted with earlier notice of the 
error, and other cases where such a feature would have been useful were 
mentioned, there did not appear to be wide-spread demand for this feature.  
No supplier of compilers reported significant user requests for a 
units package.

The estimates of compiler suppliers of the cost-to-implement this
feature were high. It was felt that implementing a standardized units 
proposal would take resources away from other efforts, including
fixing bugs, improving efficiency, and implementing other features.

Of particular concern was the long delay between the adoption of the
Fortran 2003 standard and its implementation. It is still not fully
implemented by several compilers. Many members of the committee have 
come to recognize that too many features were accepted into this 
standard. They are determined to ensure that a very high threshold be 
set for additions to future standards.

The straw vote among the members present was yes 3 - no 7 - abstain 1,
see N1977. It was therefore decided not to apply for a new work item. 
The reasons for the no votes were predominantly associated with the 
delay in the implementation of Fortran 2003, the high cost of 
implementation of this feature, and the low level of requests for it 
from applications programmers. 

More information about the J3 mailing list