• Per =?UTF-8?B?Tm9yZGzDtnc=?= (8/8) Oct 16 2020 Why is `Appender`'s store `Data` put directly as
• Steven Schveighoffer (3/17) Oct 16 2020 Appender is ref counted IIRC.
• Paul Backus (3/5) Oct 16 2020 It's not; it uses the GC.
• Steven Schveighoffer (8/14) Oct 17 2020 Oh yeah. In fact, it was me who did that (in 2010!).

Per =?UTF-8?B?Tm9yZGzDtnc=?= <per.nordlow gmail.com> writes:

Why is `Appender`'s store `Data` put directly as

     `private Data* _data;`

instead of

     `private Data _data;`

?

Removing the pointer indirection would give better locality.

If it's about optimizing for empty `Appender`s then a `Appender*` 
should be used in those cases instead.

Steven Schveighoffer <schveiguy gmail.com> writes:

On 10/16/20 5:40 PM, Per Nordlöw wrote:
 Why is `Appender`'s store `Data` put directly as
 
      `private Data* _data;`
 
 instead of
 
      `private Data _data;`
 
 ?
 
 Removing the pointer indirection would give better locality.
 
 If it's about optimizing for empty `Appender`s then a `Appender*` should 
 be used in those cases instead.

Appender is ref counted IIRC.

-Steve

Paul Backus <snarwin gmail.com> writes:

On Saturday, 17 October 2020 at 00:06:31 UTC, Steven 
Schveighoffer wrote:
 Appender is ref counted IIRC.

 -Steve

It's not; it uses the GC.

Steven Schveighoffer <schveiguy gmail.com> writes:

On 10/17/20 12:00 AM, Paul Backus wrote:
 On Saturday, 17 October 2020 at 00:06:31 UTC, Steven Schveighoffer wrote:
 Appender is ref counted IIRC.

 
 It's not; it uses the GC.

Oh yeah. In fact, it was me who did that (in 2010!).

My point should have been that the appender is a pImpl to avoid memory 
corruption. If you have multiple copies of an appender, and each has its 
own idea of what the capacity is, then you will get corruption.

See the original bug report here: 
https://issues.dlang.org/show_bug.cgi?id=4681

-Steve