From: Warren <ve3wwg@gmail.com>
Subject: Re: Generic Embedded List Nodes
Date: Mon, 20 Jun 2016 19:28:07 -0700 (PDT)
Date: 2016-06-20T19:28:07-07:00 [thread overview]
Message-ID: <6100770a-774c-41a7-b1d9-498f80426835@googlegroups.com> (raw)
In-Reply-To: <4da8ad8e-f6b4-46b3-b81d-b255c030a45c@googlegroups.com>
On Monday, 20 June 2016 15:47:12 UTC-4, Shark8 wrote:
> On Saturday, June 18, 2016 at 4:52:06 PM UTC-6, Warren wrote:
> > Getting back to Ada after a hiatus, I currently have a need to build a
> > generic package to implement "embedded list node" lists. The advantage
> > is high performance in a server setting to avoid underlying malloc/free
> > calls.
> >
> > The idea is that the list node resides within the structure/tagged type,
> > and acts as a doubly linked list node when in a list. The Emb_Node can
> > be used as a list head when itself (or as part of another structure). This
> > kind of thing is done in the Linux kernel, for example.
> >
> > The Emb_Node and its operations are trivial. The problem occurs when
> > you traverse a linked list of Emb_Nodes (or its derived type). With
> > a given node, I need to then access the object that _contains_ it. In
> > C/C++ you do some offset calculation from the node address back to
> > the owning struct/class.
> >
> > My idea (in Ada) was to save some kind of access value in a type
> > derived from Emb_List. But that is where the trouble starts.
> >
> > package Emb_List is
> >
> > type Emb_Node is tagged private;
> >
> > procedure Insert_Head(Head: access Emb_Node; Node: access Emb_Node);
> > procedure Unlink(Node: access Emb_Node);
> >
> > private
> >
> > type Emb_Node is tagged
> > record
> > Next: access Emb_Node; -- Ptr to next node in list (if any)
> > Prev: access Emb_Node; -- Ptr to prev node in list (if any)
> > end record;
> >
> > end Emb_List;
> >
> > The generic extension (below) was intended to hold the reference to the
> > containing object, which is where the trouble starts. I was hoping this
> > would work, but running into "missing full declaration for private
> > extension".
> >
> > generic
> > type Object_Type is tagged private;
> > package Emb_List.Nodes is
> >
> > type Node_Type(Obj: access Object_Type) is new Emb_Node;
> >
> > function Object(Node: Node_Type) return Object_Type;
> >
> > end Emb_List.Nodes;
> >
> >
> > The other end of the challenge will be to have the Node_Type save the
> > owning object access value at the time it is created:
> >
> > type Something_New_Type is tagged
> > record
> > Stuff: Natural;
> > Timeout_Node: Node_Type(some access to this record);
> > ...
> > end record;
> >
> > I am trying to avoid using 'Address for this, since there is likely a
> > way to do this "right".
> >
> > Warren.
>
> Hm, have you tried seeing if any Ada.Containers.* fits your needs? You could perhaps use Indefinite_Holder to hold your item or Indefinite_Vector to hold the entire list.
If you read upstream, I am looking for an _extremely_ high performance set of operations, requiring only:
- insert head
- traverse (head to tail only)
- delete from list
The implementation I have used on a C++ server works extremely well (inspired by the Linux kernel code), and costs almost nothing. The three operations above are inlined and amount to a few pointer related operations.
Even a few microseconds of overhead drops your transaction throughput considerably, so using something like Ada containers (or STL in C++) is definitely out for this application. Also any use of malloc/free where possible must be avoided because these are performance killers (even when using jremalloc).
Warren
next prev parent reply other threads:[~2016-06-21 2:28 UTC|newest]
Thread overview: 52+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-06-18 22:52 Generic Embedded List Nodes Warren
2016-06-18 23:40 ` Jeffrey R. Carter
2016-06-19 2:15 ` Warren
2016-06-19 3:04 ` Jeffrey R. Carter
2016-06-19 2:14 ` Jeremiah
2016-06-19 2:21 ` Warren
2016-06-19 2:50 ` Warren
2016-06-19 4:45 ` Simon Wright
2016-06-19 18:27 ` Warren
2016-06-19 19:04 ` Dmitry A. Kazakov
2016-06-19 20:13 ` Warren
2016-06-19 20:35 ` Dmitry A. Kazakov
2016-06-20 2:42 ` Warren
2016-06-20 7:25 ` Dmitry A. Kazakov
2016-06-20 12:26 ` Warren
2016-06-20 19:33 ` Niklas Holsti
2016-06-21 2:20 ` Warren
2016-06-21 5:52 ` Niklas Holsti
2016-06-21 7:15 ` Dmitry A. Kazakov
2016-06-21 18:54 ` Niklas Holsti
2016-06-21 19:54 ` Dmitry A. Kazakov
2016-06-21 10:31 ` Warren
2016-06-21 17:13 ` Jeffrey R. Carter
2016-06-21 18:56 ` Niklas Holsti
2016-06-21 20:13 ` Warren
2016-06-21 21:38 ` Niklas Holsti
2016-06-23 2:12 ` Warren
2016-06-23 8:19 ` Niklas Holsti
2016-06-23 12:37 ` Warren
2016-06-23 15:36 ` Niklas Holsti
2016-06-24 1:55 ` Warren
2016-06-24 12:49 ` Warren
2016-06-25 5:50 ` Niklas Holsti
2016-06-26 1:36 ` Warren
2016-07-01 13:49 ` Warren
2016-07-01 16:28 ` Warren
2016-06-24 20:25 ` Warren
2016-06-22 13:01 ` G.B.
2016-06-23 2:30 ` Warren
2016-06-20 6:08 ` Niklas Holsti
2016-06-20 12:20 ` Warren
2016-06-20 19:47 ` Shark8
2016-06-21 2:28 ` Warren [this message]
2016-06-21 7:21 ` Dmitry A. Kazakov
2016-06-21 10:32 ` Warren
2016-06-21 11:56 ` Dmitry A. Kazakov
2016-06-21 13:39 ` Warren
2016-06-21 14:04 ` Dmitry A. Kazakov
2016-06-23 0:37 ` Randy Brukardt
2016-06-23 2:25 ` Warren
2016-07-01 19:50 ` brbarkstrom
2016-07-02 1:55 ` Warren
replies disabled
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox