On Thu, 2014-01-09 at 22:10 +0100, Luca Bruno wrote:
On Thu, Jan 9, 2014 at 9:58 PM, Tal Hadad <tal_hd-PkbjNfxxIARBDgjK7y7TUQ public gmane org> wrote:That would be a memory leak.No it wouldn't: If the "if" statement is false, then it should delete it in the end of the block. You might afraid the complicity of valac calculation. That could be resolved - using two C variables - one for owned and one for unowned. The owned always deleted at the end of the block, the trick is that this C variable is set to null when transferring ownership.If an owned "bar" was defined outside the block, I already said you're right and it should be null in when outside the block(e.g. class field), since there's no way to automate this safely. I know it may add a minor complicity, but it just seems rational to use an object with "unowned" access after transferring it to something else, since passing ownership privilege doesn't means forgetting a reference.Ok now this makes sense, but it's not worth in my opinion. I've never encountered such a use case personally.I suppose that most Vala programmer solve this(including me) problem by saving this variable in different unowned variable, but this is not intuitive(especially for a new language) and may raise many bugs.I don't think so, I never had such a use case.
There are a few places in libgee where the code looks like:
unowned G g_ref = g;
something = (owned)g;
do_something_else(g_ref); // This line cannot be moved above
On the other hand I prefer when it blows with segfault during tests
rather then doing potential write to random memory without extra
verbosness:
void do_something(G ref) {
something = null;
}
something = (owned)g; // It's the only reference
do_something_else(g);
g.a += 1; // Oops.
Not mentioning:
if (x) {
something = (owned)g;
}
if (y) {
// It's allowed only if y -> !x [1]
something_else = (owned)g;
}
While it's possible to get around it by Rust-like types but I don't
think it's worth for Vala as it has different goals.
PS. It might be nice on the other hand to extend semantic checker to
catch use-after-ownership-transfer at some point in the future.
[1] For those who don't know implies from logic it's !y || !x or !(y &&
x)
Attachment:
signature.asc
Description: This is a digitally signed message part