From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from smtp36.i.mail.ru (smtp36.i.mail.ru [94.100.177.96]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by dev.tarantool.org (Postfix) with ESMTPS id 949564696C3 for ; Mon, 20 Apr 2020 18:29:32 +0300 (MSK) Date: Mon, 20 Apr 2020 18:29:31 +0300 From: Kirill Yukhin Message-ID: <20200420152931.e7au2ldtiwpqp2dt@tarantool.org> References: <20200420142237.jrnslig6pusym6nd@tarantool.org> <20200420143524.GA27460@tarantool.org> MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Disposition: inline Content-Transfer-Encoding: 8bit In-Reply-To: <20200420143524.GA27460@tarantool.org> Subject: Re: [Tarantool-patches] [PATCH v2 0/2] Stacked diagnostic area follow-ups List-Id: Tarantool development patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Nikita Pettik Cc: tarantool-patches@dev.tarantool.org, v.shpilevoy@tarantool.org On 20 апр 14:35, Nikita Pettik wrote: > On 20 Apr 17:22, Kirill Yukhin wrote: > > Hello, > > > > On 17 апр 23:16, Nikita Pettik wrote: > > > Branch: https://github.com/tarantool/tarantool/tree/np/gh-4887-ref-error-on-prev > > > Issue: https://github.com/tarantool/tarantool/issues/4887 > > > > > > Changes in v2: > > > > > > - modified test so that now it uses weak references to check that > > > gc collected error objects (i.e. there's no memory leaks); > > > - added overflow check in error_ref() so that after 2^32 calls > > > of box.error.last() or error:prev() error object won't contain > > > broken reference counter. > > > > > > Nikita Pettik (2): > > > box/error: don't allow overflow of error ref counter > > > box/error: ref error.prev while accessing it > > > > I've checked your patchset into master. > > > > However, looks like it'd be better to just replace int32 to > > int64 and avoid problems if GC64. > > Could you please specify which problems you are talking about? > Current fix simply prevents from overflow, so it does not > introduce any other problems. Meanwhile users who ref single > error more than 2^32 times are likely doing smth wrong so > Lua error notifying them aboit it is a good practice IMHO. I'd avoid work 'likely' when it is possible. Vlad's approach allows to do that: it will work always, wherever user doing something right or wrong, it will just work. BTW, we are close to start off adopotion of GC64 (w/ ujit or on vanilla luajit), so ref error 2 billions times doesn't look impossible. -- Regards, Kirill Yukhin