Prev: a compiler error in template function default argument value
Next: FRUCTOSE unit test framework status change and article correction
From: Vidar Hasfjord on 10 Apr 2008 06:22
Hi, I'm in the process of redesigning a library for handling
traditional error values in C and legacy C++ code. I would very much
appreciate comments on the merits and possible implementation problems
with the approach outlined below.

* Rationale and overview

Error values without exception handling breaks composability of
expressions.
For example, the expression (f(x) + g(y)) must be broken up into
statements that check the return value of each call.

Error-value-to-exception translation wrappers solve this, but they are
often verbose (see Enforcements by Alexandrescu/Marginean, Dr.
Dobb's). Also, such wrappers often combine the detection and
translation of an error value with handling and reporting it (e.g.
passing along error messages).

The proposed idiom is a lighter "footnote" approach in which error
detection and translation are treated orthogonal to handling and
reporting. Simply, operator % is used as an error value filter. For
example:

int r = E%Foo (E%Bar (a) + b);

Here "E%Foo" reads as "Error-checked Foo", where E is an instance of
an error checker and Foo is a function, method or functor returning
some error code. E may have the following declaration elsewhere:

ErrorTranslation::NotNegativeChecker E;

Textually E functions similarly to a footnote in ordinary text. It
says "look up E to find out more about what's going on here".

* Details

For detecting error codes in expressions without adding verbose
wrapping a binary operator with high precedent is used; % rank at the
top. Unfortunately there are unary operators of higher precedent and
binary operators of same precedent. This would normally require the
uses of parenthesis to specify the correct meaning. For example:

E%Foo (a) * (E%Bar (b))

But proxies can be used circumvent the normal evaluation order, hence
giving the illusion that E% is binding tighter than surrounding
operators.

E%Foo (a) * E%Bar (b) // same results as above

Multiplication and division are the only binary operators of equal
precedent to %. An expression (E%f(x) * E%g(y)) is evaluated as (((E%
(f(x))) * E) % (g(y))) where the expression ((...) * E) returns a
proxy object. The proxy object carries the value (...) and the
multiplication forward through to the next check and applies the
multiplication after checking.

For unary operators of higher precedent, such as negation, (-f(x))
turns into (-E%f(x)) which is evaluated ((-E) % (f(x))). The
expression (-E) returns a proxy object. The proxy carries the negation
forward to the next check and applies it after checking.

Proxies compose; e.g. a negation applied to a proxy returns a new
proxy object that includes the negation.

Undesirable applications, such as assignment and copying, of checkers
and proxies are prohibited.

Any comments appreciated.

Regards,
Vidar Hasfjord

--
[ See http://www.gotw.ca/resources/clcm.htm for info about ]
[ comp.lang.c++.moderated. First time posters: Do this! ]

 | 
Pages: 1
Prev: a compiler error in template function default argument value
Next: FRUCTOSE unit test framework status change and article correction