Style question: returning error values from functions

[Dar Scott]

On Saturday, July 12, 2003, at 06:11 AM, Richard Gaskin wrote:

> Which approach do you prefer?  One of these?  Something else?

I use these:

1.  Reshape the semantics so there are no errors.
2.  Throw
3.  Value returned in referenced variable and success/fail (or related) 
returned.
4.  Separate domain validation functions (used before instead of after) 
with, perhaps, #2 or #10
5.  Callback (very rarely for functions; I'll ignore this)

Sometimes #1 is like your "error" prefix but not always.  For example, 
char 5 of "abc" returns a reasonable value, empty, rather than an 
error. In this approach what is reasonable depends on the function.  
Another #1 approach is to define a way to coerce parameters to the 
nearest closest value.  A general scheme can be used for numbers that 
uses rounding and bounding as needed.

The callback might reduce to changing a global error to a local error.

These are also acceptable to me:
6.  Global
7.  Function like sysError()
8.  Error returned in referenced variable in parameters
9.  Use NaN, +Inf, etc for numerical functions (like #1)
10.  Let the function throw error or return garbage; used with #4

Revolution built-in functions and operators use 1-3.  Of #1 some return 
reasonable values and some coerce parameters to the function domain.

I readily use throw in scripts for my own use.  I also have used it in 
a library I will make generally available, and I'm thinking of changing 
the scripts so that they do not throw.  Are people comfortable with 
throw?

The value of functions is that you can use them in expressions.  If you 
have to check after each use, then you lose the benefit.  That makes #2 
(throw) with perhaps #4 (data checking beforehand) more interesting.

Dar Scott