• dsimcha (5/5) Nov 23 2010 I was browsing through druntime's lifetime.d for an unrelated reason and...
• Jonathan M Davis (5/10) Nov 23 2010 Well, if it's copied for the purpose of appending, the elements need to ...
• Max Samukha (8/18) Nov 24 2010 The appended elements would need to be postblitted. Also, the original
• Steven Schveighoffer (12/21) Nov 24 2010 Array appending does not respect any of the new copy construction

dsimcha <dsimcha yahoo.com> writes:

I was browsing through druntime's lifetime.d for an unrelated reason and I
noticed that, when an array is copied for the purpose of appending to it, the
postblits are not called on the new array if it is an array of structs.  Is
there something I'm missing, or is this a bug that somehow noone has noticed
up to this point?

Jonathan M Davis <jmdavisProg gmx.com> writes:

On Tuesday 23 November 2010 20:43:34 dsimcha wrote:
 I was browsing through druntime's lifetime.d for an unrelated reason and I
 noticed that, when an array is copied for the purpose of appending to it,
 the postblits are not called on the new array if it is an array of
 structs.  Is there something I'm missing, or is this a bug that somehow
 noone has noticed up to this point?

Well, if it's copied for the purpose of appending, the elements need to be 
moved, not copied, don't they? And if they're moved, there's no need to call
the 
postblit constructor.

- Jonathan M Davis

Max Samukha <spambox d-coding.com> writes:

On 11/24/2010 07:06 AM, Jonathan M Davis wrote:
 On Tuesday 23 November 2010 20:43:34 dsimcha wrote:
 I was browsing through druntime's lifetime.d for an unrelated reason and I
 noticed that, when an array is copied for the purpose of appending to it,
 the postblits are not called on the new array if it is an array of
 structs.  Is there something I'm missing, or is this a bug that somehow
 noone has noticed up to this point?

 Well, if it's copied for the purpose of appending, the elements need to be
 moved, not copied, don't they? And if they're moved, there's no need to call
the
 postblit constructor.

The appended elements would need to be postblitted. Also, the original 
elements would need to be postblitted in case the append triggers a 
reallocation. They cannot be simply moved because there may be slices 
pointing to the original array.

Anyway, in the general case we cannot store structs with dtors/postblits 
in dynamic arrays (append requires GC and GC cannot destroy structs), 
which means you will have to use a special container for that purpose.

 - Jonathan M Davis

"Steven Schveighoffer" <schveiguy yahoo.com> writes:

On Tue, 23 Nov 2010 23:43:34 -0500, dsimcha <dsimcha yahoo.com> wrote:

 I was browsing through druntime's lifetime.d for an unrelated reason and  
 I
 noticed that, when an array is copied for the purpose of appending to  
 it, the
 postblits are not called on the new array if it is an array of structs.   
 Is
 there something I'm missing, or is this a bug that somehow noone has  
 noticed
 up to this point?

Array appending does not respect any of the new copy construction  
semantics.  It basically does a memcpy and that's it.  It doesn't even  
respect shared (shared data is memcpy'd just like any other data),  
although it does use a lock to protect shared appends.

I see there is a postblit function in the typeinfo, so it should be  
possible to do the right thing.

One thing I'm unsure of is how postblit works on const or immutable  
structs, anyone know?  There is no typechecking in the array functions,  
everything is a void[].

File a bug on it please.

-Steve