I'm leaning to splitting identifier from the classification.

A pure monotonically increasing integer would ensure the non-reuse.

And the classification scheme could evolve seperately.

Using numbers for the classification scheme would require a translation
table. So why not just use alphanumerics.

At a minimum I can see, the originator, the severity, and some class.

I'm leaning to having the class tie in with lexical warnings.

<chaim>

>>>>> "AS" == Ariel Scolnicov <[EMAIL PROTECTED]> writes:

>> =item As a floating point number

AS> Since you're anyway not using a _floating_ point, here are 2
AS> alternatives:

AS> 1. Error codes are integers, but with an implied decimal point (fixed
AS>    point).  E.g. 1234567 represents error 4567 in class 123.

AS> 2. Error codes as digit strings with decimal point.  So "123.4567"
AS>    (_not_ 123.4567) represents error 4567.  People will still try
AS>    stupid floating-point math tricks to get at the suberror code, but
AS>    at least we'll know they didn't read the bit in the documentation
AS>    where it will specifically warn _not_ to do this.

-- 
Chaim Frenkel                                        Nonlinear Knowledge, Inc.
[EMAIL PROTECTED]                                               +1-718-236-0183

Reply via email to