From mboxrd@z Thu Jan  1 00:00:00 1970
Received: from mail4-relais-sop.national.inria.fr (mail4-relais-sop.national.inria.fr [192.134.164.105])
	by walapai.inria.fr (8.13.6/8.13.6) with ESMTP id q23IpJ0q032138
	for <caml-list@sympa-roc.inria.fr>; Sat, 3 Mar 2012 19:51:19 +0100
X-IronPort-Anti-Spam-Filtered: true
X-IronPort-Anti-Spam-Result: AvgCAKlnUk/UGyoBkWdsb2JhbABDgxauO4J8IgEBAQEJCwsHFAMkgX0BAQQBMgFFBgsLIRYPCQMCAQIBRRMIAQGHfgm3FY07gyIElT6FYY0V
X-IronPort-AV: E=Sophos;i="4.73,526,1325458800"; 
   d="scan'208";a="134183711"
Received: from smtp1-g21.free.fr ([212.27.42.1])
  by mail4-smtp-sop.national.inria.fr with ESMTP; 03 Mar 2012 19:51:13 +0100
Received: from [192.168.0.3] (bur91-2-82-231-161-160.fbx.proxad.net [82.231.161.160])
	(Authenticated sender: tiphaine.turpin)
	by smtp1-g21.free.fr (Postfix) with ESMTPA id 432E594018A
	for <caml-list@inria.fr>; Sat,  3 Mar 2012 19:51:06 +0100 (CET)
Message-ID: <4F52761B.6010509@free.fr>
Date: Sat, 03 Mar 2012 20:50:51 +0100
From: Tiphaine Turpin <tiphaine.turpin@free.fr>
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:8.0) Gecko/20111221 Thunderbird/8.0
MIME-Version: 1.0
To: caml-list@inria.fr
References: <4F527560.6010409@free.fr>
In-Reply-To: <4F527560.6010409@free.fr>
X-Forwarded-Message-Id: <4F527560.6010409@free.fr>
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 8bit
Subject: Re: [Caml-list] application scope for hashtable on weak pointers


Le 03/03/2012 19:21, Philippe Strauss a écrit :
> I'm puzzled by this superb sentence in the refman:
>
> "A weak hash table is a hashed set of values.
> Each value may magically disappear from the
> set when it is not used by the rest of the program any more."
>
> I have a use case for an hashtable of values who'll needs to get aged and removed after a while,
> what can I reuse of the standard lib, under which conditions are the values collected by the GC ?
A value in a weak hash-table may be collected provided :
- it is not reachable (from the roots of the reachable heap, i.e., the
stack, C-registered pointers, etc.) except through paths having a
weak-pointer (which is the case for the weak hash-table)
- the triggering conditions for the GC are met.
A typical use-case is hash-consing (physical sharing of
structuraly-equal data, for memory-efficiency).

> my use case would be to first age out when a threshold of number of bindings is reached, also, after some wall time, flush oldest entries (LRU). 
Maybe you can have a "LRU-sorted" (normal) array of not-too-old data and
a weak array (or hash-table) of aged data that the GC may collect, if no
one else points to it, but which you can still use until the next collection

Tiphaine