(Original thread, and test program, can be found here at
http://permalink.gmane.org/gmane.comp.gcc.help/26289)
A few days ago, Kevin posted an interesting problem, whose ultimate
solution I find rather confusing... Here's the results of
__builtin_types_compatible_p on some various types; I'm hoping someone
that knows gcc's internals can comment?
> char*, char* => true
> char*, const char* => false
The doc says that qualifiers are ignored, but in this case that isn't
happening? (However, we WANT this behavior, so this is a complaint
against the doc, not the code generation!)
> char[], char* => false
Why are these not compatible?
> char[], const char[] => true
Ok, here the qualifier is being ignored?
> typeof("hello"), char* => false
> typeof("hello"), char[] => true
> typeof("hello"), const char* => false
> typeof("hello"), const char[] => true
This seems... odd. Given the above I can understand how the 'const' is
getting lost, but this seems less helpful than it could be.
> typeof(&(*"hello")), char* => true
> typeof(&(*"hello")), char[] => false
> typeof(&(*"hello")), const char* => false
> typeof(&(*"hello")), const char[] => false
I think I understand why gcc ate the 'const' here, but it's still a
little scary :-).
> typeof(&(("hello")[0])), char* => false
> typeof(&(("hello")[0])), char[] => false
> typeof(&(("hello")[0])), const char* => true
> typeof(&(("hello")[0])), const char[] => false
...and we finally arrive at a useful incantation that turns
pointer-or-array into pointer without losing the qualifier. But why was
this necessary?
--
Matthew
Please do not quote my e-mail address unobfuscated in message bodies.
--
When in doubt, give it a good swift kick.
