http://d.puremagic.com/issues/show_bug.cgi?id=4358
--- Comment #8 from Leandro Lucarella <llu...@gmail.com> 2010-06-22 10:03:22 PDT --- (In reply to comment #7) > std.file.read returns a void[], but it's allocated manually via gc.malloc(), > and I think the no-pointers thing was done right. And what is the rationale for not returning ubyte[]? > The correct solution to this problem would be to apply the patch in bug 3463 > (precise heap scanning), and then probably investigate precise stack scanning. You can't have full precise scanning in D, because of unions and other low-level constructs, so the problem of false pointers will be still there even with that patch (of course the chances of having a false pointer will be much lower). > I believe the patch in bug 2927 (no interior pointers attribute) would not > roll > well with Andrei's safety ideology, so let's forget that. That's stupid, it's not unsafer than NO_SCAN or casting a pointer to int, or whatever construct you like. That's obviously a construct that not everybody will use. So let's not forget that (and putting words in somebody else's mouth is not a good argument either =). > Note that enhancement 2927 could be simulated by using a wrapper object, that > behaves like an array, but actually uses C malloc to allocate memory. When the > wrapper gets collected, its finalizer can free() the memory. Or using > reference > counting, like (I believe) Andrei's TightArray (or what was it named) does. That's true, is a little more convoluted than having NO_INTERIOR and is not as easy to use by the compiler itself, but could be another option. -- Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email ------- You are receiving this mail because: -------
