Walter Bright:
Denied. I don't see the rationale for in one case but not
others.
There is a rationale. In my opinion special cases are bad in a
language, unless:
- They always produce a compilation failure in a well defined
group of cases. (while just giving a different answer in a
vaguely defined group of cases is usually not acceptable);
- They help the programmer avoid a common bug-prone situation.
In D we have already some cases of such special rules, that break
uniformity and symmetry to help D programmers write less buggy
code.
Issue 4407 (
http://d.puremagic.com/issues/show_bug.cgi?id=4407 ) is/was one
of my fifteen most important enhancement requests (and it's a
small one). Many of those fifteen are little breaking changes.
I think the case discussed in Issue 4407 passes both conditions:
The cause is defined for class/structs. I think this is a
definite enough situation
Regarding commonality:
- I know two tools that detect this bug, the tool that has
already found some (different) bugs in the DMD source code and
the Google Web Toolkit.
- I have created several bugs because of that, and I am forced to
use a coding convention (like adding a trailing _ to argument
names in constructors) to avoid them.
- I know three D programmers that have had the same problem of
mine, Byron Heads, the OP and another person.
- The static analysis tool I was referring before has uncovered
cases of such bug in well used C++ code.
Bye,
bearophile
