--- a/codeDoc/policies.txt	Fri May 02 16:20:35 2008 +0100
+++ b/codeDoc/policies.txt	Fri May 02 17:38:43 2008 +0100
@@ -44,6 +44,9 @@
 
 8   Translating strings
 
+9 Floating point types 
+
+
 
 
 --- Coding conventions ---
@@ -145,3 +148,8 @@
 User output (internationalization support): i18n.I18nTranslation is designed to fetch a full string and optionally an associated descripion given an identifier. The identifier may also be a code symbol, and should be brief, in English, and give at least some idea of its meaning, since if no translation is available the identifier will be output. For example:  "example message" or "exampleMessage", not "An example of a message". For any new entry created, at least one full entry should be added to the i18n database for some form of English so that:
     (a) there is something available to translate to other locales
     (b) English developers can understand what it means
+
+
+
+--- Floating point types ---
+The issue is whether to use float, double or real as the basic floating point type. In some cases which types are appropriate for use is set by the prototypes of external methods being used. In other cases, which type to use is left entirely to the user's discretion. Since even modern 32-bit CPUs support native 80-bit floating point operations, there seems little point in using floats. Since doubles are more widely supported than reals and provide sufficient precision for most purposes, double seems to be the ideal type to use. Actually some basic tests with 32-bit dmd on my athlon-64 (under 64-bit linux) show floats to be slower to use than doubles but faster than reals, so this appears to be a good choice for now.
\ No newline at end of file