Re: Internal type conversion in Epics

[mrk@aps.anl.gov (Marty Kraimer)]

> From ajf@ast.cam.ac.uk Wed Jul 19 09:09 CDT 1995
> Date: Wed, 19 Jul 1995 15:08:01 +0100 (BST)
> From: Andy Foster <ajf@ast.cam.ac.uk>
> X-Sender: ajf@orc
> To: Epics Questions <tech-talk@aps.anl.gov>
> Subject: Internal type conversion in Epics
> Mime-Version: 1.0
> Content-Type> : > TEXT/PLAIN> ; > charset=US-ASCII> 
> Content-Length: 1090
> 
> 
> Marty,
> 
> Thanks for your reply concerning the accuracy of
> converting a Double to a String within EPICS.
> I have done some further investigation and found that
> by setting PREC to 7 in the incoming record (the record
> which is receiving the double), I do indeed get an accuracy
> of 7 decimal places. However, changing PREC to 15 (which you
> might want for the accuracy of a double), 
> I discovered that the double goes through with an
> accuracy of only 5 decimal places!
> 
> The problem lies within the code for "cvtDoubleToString" inside
> libCom/cvtFast.c. Line 150 of cvtFast.c shows that the value will
> be converted to a string as follows, whenever PREC > 8.
> 
> sprintf(pstr_value,"%12.5e\0",flt_value);
> 
> I would say that "%12.5e" format is not what we want in
> this case. Are there any plans to change this to something
> which greater reflects the accuracy expected of a double?

I agree. I will add this to list of bugs in 3.12.1. I think that format should
be  "%22.15e". This change could break code that calls cvtDoubleToString
with insufficiant space for the string. We can only give a warning when
change is made.

Marty Kraimer