Thanks for the info Justin. (Sorry I'm slow getting back to you; I was away last week).
If I work at implementing this can I bug you for help and ask you to review the work as I go along? I'm sure I'll need some advice on how to do things properly. Lastly - how do we define a "Safe Conversion"? Is this something we can define or is this well defined somewhere. I'm thinking that converting from integer to double is probably safe but not so for double to integer. I'm sure the other types have similar properties. Thanks, Emily Justin Deoliveira wrote: > Hi Emily and Jody, > > We have run up against this issue before, and it illuminates a pretty > fundamental bug in our filter code. The following very basic test case > for instance fails: > > Literal d = ff.literal( 1.1 ); > Literal i = ff.literal( 1 ); > > Filter f1 = ff.greater( d, i ); > assertTrue( f1.evaluate( null ) ); > > Filter f2 = ff.less( i, d ); > assertTrue( f2.evaluate( null ) ); > > When we ran up against this last time the plan (which we never > implemented) was to have a way to specify that only "safe" conversions > should be executed (via a hint). That way the logic in > BinaryComparisonAbstract becomes: > > 1. try safe conversion from class1 to class 2 > 2. try safe conversion from class2 to class 1 > 3. fall back to current logic of unsafe conversions > > The converters api already supports hints, so all that is needed is to > a) create the hint for safe conversions and b) to respect that hint in > the NumericConverterFactory, and any other converter factories that > potentially perform conversions that lose information. > > -Justin > > Jody Garnett wrote: >> On Wed, Jun 17, 2009 at 8:38 AM, Emily Gouge<ego...@refractions.net> >> wrote: >>> Hmm I'm just not sure your extra code is needed. There is already a >>> converter to convert from Strings to Numbers - what I really need is to >>> ensure that doubles are not converted to integers. >> >> That sounds good. >> >>> I wonder if we just need to call the convert function with Number >>> instead of >>> double/integer etc. This will convert strings if necessary. I'm >>> thinking >>> of this (passes all of the test cases you provided previously): >>> >>> Object o = Converters.convert(v2, (v1 instanceof Number) ? Number.class >>> : v1.getClass()); >>> >>> instead of >>> >>> Object o = Converters.convert(v2, v1.getClass()); >> >> Both those fail the test when: >> - v1 is a number and v2 is a string >> - both v1 or v2 are strings >> >> I think you need a two pass check >> 1. Check if you can convert the values to Number; if so do a comparison >> 2. Continue with normal comparison as per before (try and make them >> the same Compariable; and then compare) >> >> ------------------------------------------------------------------------------ >> >> >> Crystal Reports - New Free Runtime and 30 Day Trial >> Check out the new simplified licensing option that enables unlimited >> royalty-free distribution of the report engine for externally facing >> server and web deployment. >> http://p.sf.net/sfu/businessobjects >> _______________________________________________ >> Geotools-devel mailing list >> Geotools-devel@lists.sourceforge.net >> https://lists.sourceforge.net/lists/listinfo/geotools-devel > > ------------------------------------------------------------------------------ _______________________________________________ Geotools-devel mailing list Geotools-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/geotools-devel
