Fixes bug 5272 -- Postblit not called on copying due to array append

[schveiguy]

Added handling of postblit to all array operations.

Passes all druntime and phobos unit tests on Linux.

[michelf]

Is this the right place to fix it though? Postblit should only happen when copying and not when moving. If you do something like this:

 S[] a;
 a ~= S();

I wouldn't expect postblit to be called at all, since you're simply moving the new S to the end of the array. Is this handled correctly by this patch?

[schveiguy]

michelf, it is impossible for the runtime to tell whether this is a move or a copy.

For example, the above code could be written:

 S s;
 a ~= s;

And the same exact runtime function is called with the same exact argument types. In order to fix this, the compiler would have to be modified (maybe a bool identifying move/copy?)

But I do agree it is an issue, could you file a separate bug report?

[michelf]

I'm not sure how I should fill a bug for something that isn't a bug yet. Moving a struct by appending it to an array works as expected currently (no postblit). The behaviour will become wrong only once this request is pulled.

[schveiguy]

Well, it's only not a bug because postblit is not implemented at all. The bug is that the compiler doesn't give the runtime enough info to make the right decision. The patch fixes all cases where postblit should be called. I can file the bug if you like.

On Jun 18, 2011, at 9:59 AM, michelf reply@reply.github.com wrote:

I'm not sure how I should fill a bug for something that isn't a bug yet. Moving a struct by appending it to an array works as expected currently (no postblit). The behaviour will become wrong only once this request is pulled.

Reply to this email directly or view it on GitHub:
#29 (comment)