220 18298 article Path: news.gmane.org!not-for-mail From: "'Jeffrey Yasskin' via ISO C++ Standard - Future Proposals" Newsgroups: gmane.comp.lang.c++.isocpp.proposals Subject: Re: Parsing Numbers Date: Fri, 29 May 2015 09:31:54 -0700 Lines: 118 Approved: news@gmane.org Message-ID: References: <555CDDEF.8000002@gmx.net> <555ECB8E.9010606@gmx.net> <55680FE0.9050608@gmx.net> <5568237D.7010802@gmx.net> <5568481E.3060709@gmx.net> Reply-To: std-proposals@isocpp.org NNTP-Posting-Host: plane.gmane.org Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 X-Trace: ger.gmane.org 1432917144 14032 80.91.229.3 (29 May 2015 16:32:24 GMT) X-Complaints-To: usenet@ger.gmane.org NNTP-Posting-Date: Fri, 29 May 2015 16:32:24 +0000 (UTC) To: std-proposals@isocpp.org Original-X-From: std-proposals+bncBDDM34EO6QDRBEFJUKVQKGQEVWJJREI@isocpp.org Fri May 29 18:32:23 2015 Return-path: Envelope-to: gclcip-std-proposals@m.gmane.org Original-Received: from mail-ie0-f200.google.com ([209.85.223.200]) by plane.gmane.org with esmtp (Exim 4.69) (envelope-from ) id 1YyNCk-0003VR-ID for gclcip-std-proposals@m.gmane.org; Fri, 29 May 2015 18:32:18 +0200 Original-Received: by iepj10 with SMTP id j10sf124765385iep.0 for ; Fri, 29 May 2015 09:32:17 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20130820; h=x-gm-message-state:mime-version:in-reply-to:references:from:date :message-id:subject:to:content-type:x-original-sender :x-original-authentication-results:reply-to:precedence:mailing-list :list-id:list-post:list-help:list-archive:list-subscribe :list-unsubscribe; bh=M/nEuwnCOx9U8JaeeS5/7qwdOikHvsj2ummKTJ4rFqM=; b=cKq7al4IH0HGUaslq8uL9K+rBfOqPpNtOKe8HEzjh+QYWlkd5lYIF/vjc0svV2EVXu c6j8sXHRxAyECLZeFLTLl8gloyDuyAaqASxYr7EwQhCDVAy568Ohvc1uA9jBWVqWqmGy vgBUDoc5UiVpkelZemCvolJOIbyUPlMGJO9vdFi10V1HatHKdsEHjhNsnxcW9WPDvJ7h hoALn6ARBfcGwa4LRuao1BiT1Gt3AA+waZaWISGl+whXhPQuPeBismE8kkpA/e6/gx4h IGRYWrhXnRyKIiUqKnLppbeIWS/ae66UdaLoR6HmrG6VceO6qqEhgRghPKdjkljIifVm VNew== X-Gm-Message-State: ALoCoQl2sASuf/QPpeR58fVJVpeKBg/Qc1a/nZMeGeq+sE62HKNB/9R7BQWrdWOZPWnnlxy5em5B X-Received: by 10.182.61.66 with SMTP id n2mr10844702obr.20.1432917137597; Fri, 29 May 2015 09:32:17 -0700 (PDT) X-BeenThere: std-proposals@isocpp.org Original-Received: by 10.107.156.143 with SMTP id f137ls1003942ioe.60.gmail; Fri, 29 May 2015 09:32:16 -0700 (PDT) X-Received: by 10.107.156.71 with SMTP id f68mr11072366ioe.36.1432917136543; Fri, 29 May 2015 09:32:16 -0700 (PDT) Original-Received: from mail-ie0-x235.google.com (mail-ie0-x235.google.com. [2607:f8b0:4001:c03::235]) by mx.google.com with ESMTPS id kb9si1930882igb.48.2015.05.29.09.32.16 for (version=TLSv1.2 cipher=ECDHE-RSA-AES128-GCM-SHA256 bits=128/128); Fri, 29 May 2015 09:32:16 -0700 (PDT) Received-SPF: pass (google.com: domain of jyasskin@google.com designates 2607:f8b0:4001:c03::235 as permitted sender) client-ip=2607:f8b0:4001:c03::235; Original-Received: by iepj10 with SMTP id j10so66664121iep.3 for ; Fri, 29 May 2015 09:32:16 -0700 (PDT) X-Received: by 10.107.10.151 with SMTP id 23mr9145676iok.89.1432917136130; Fri, 29 May 2015 09:32:16 -0700 (PDT) Original-Received: by 10.64.73.2 with HTTP; Fri, 29 May 2015 09:31:54 -0700 (PDT) In-Reply-To: <5568481E.3060709@gmx.net> X-Original-Sender: jyasskin@google.com X-Original-Authentication-Results: mx.google.com; spf=pass (google.com: domain of jyasskin@google.com designates 2607:f8b0:4001:c03::235 as permitted sender) smtp.mail=jyasskin@google.com; dkim=pass header.i=@google.com; dmarc=pass (p=REJECT dis=NONE) header.from=google.com Precedence: list Mailing-list: list std-proposals@isocpp.org; contact std-proposals+owners@isocpp.org List-ID: X-Google-Group-Id: 399137483710 List-Post: , List-Help: , List-Archive: List-Subscribe: , List-Unsubscribe: , X-Original-From: Jeffrey Yasskin Xref: news.gmane.org gmane.comp.lang.c++.isocpp.proposals:18298 Archived-At: FWIW, I think Jens' arguments are going to be convincing to the committee, so it's probably a good idea to follow them in the first iteration of this paper. On Fri, May 29, 2015 at 4:06 AM, Jens Maurer wrote: > On 05/29/2015 12:12 PM, Olaf van der Spek wrote: >> 2015-05-29 10:29 GMT+02:00 Jens Maurer : >>>> Can't the string_view one be easily generalized too? >>> >>> Which way? Have a range_view that takes a pair of arbitrary >>> iterators of type It? Sure. This feels even farther away >>> from established STL precedence. >> >> No, I mean adding an overload taking two or three iterators.. > > I'm looking for the most basic abstraction to be standardized. > > We've standardized quite a few non-basic abstractions, e.g. > iostreams or num_get<>(), which leaves a performance gap between > "what the standard provides" and "what the user could write himself > when not needing the bells + whistles". I don't want to have a > gap discussion ever again in the area we're talking about. > > I have no sustained objection to standardizing various additional > overloads in addition to standardizing the basic abstraction, > if people feel like it. (I'm not really in favor, because it > blows up the std.lib interface, which makes it a little harder to > learn each time we do that.) > >> I guess we'll include it anyway due to the aliasing concerns. >> >> File read() and write() also update the file pointer in a hidden way.. >> is doing so really that bad? > > File I/O has hidden state, yes. Is that bad? Maybe. We also > have pread and pwrite for those that don't like the hidden state. > >> If we get unified call syntax we even might be able to write is.parse() > > Yes, if string_view is the first parameter (which it currently isn't). > >>> Let's assume the "parse" function is inlined. Ideally, string_view's >>> components should be kept in registers and never hit memory. >>> Now, the first "parse" call changes the value of the member of "s" >>> holding s.begin() (let's call it s.first). This write must hit >>> memory, because s.first might point to itself or to s.end under >>> the aliasing rules. >>> >>> In the second parse call, the "auto it = is.begin()" and >>> "is.end()" must read from memory afresh, because those values >>> might have been changed by the write to s.first of the first >>> parse call. >> >> is.begin will almost surely have been changed. > > Yes, sorry. > >>> End result: string_view's components are not kept in registers. >> >> True, but how big of a problem is one or two reads from L1 cache? > > That might not be the only effect. Memory writes to an arbitrary > location (from the viewpoint of the compiler) might inhibit > code motion / scheduling between the two calls to parse, too, > producing more pipeline stalls etc. > >>> When s.first and s.end are separate local variables, it's obvious >>> to the compiler that an externally-supplied value cannot possibly >>> point to them. >> >> Hmm, I'm inclined to call this a quality of implementation issue. > > As I said earlier, I failed to achieve the QoI for the "output" > case and a similar interface structure. > >>>>> and can be made to look parallel with output. >>>>> Output doesn't work >>>>> with string_view at all, because its elements are const. >>>> >>>> True, but IMO we should punish input for that. >>> >>> When we're done, we should have a set of elementary parse() >>> functions in the standard, and a corresponding set of output() >>> functions. Considering some in isolation is fine up to a point, >>> but we shouldn't lose track of the big picture. >> >> Output / format functions are a whole different story. I'm not sure >> it's worth it to worry about that for this proposal. > > My use case is a (say) JSON input / output library. > I don't want this in the standard right now, but I do want to have > the basic building blocks available so that I can write my own > in C++ whose performance blows everything else out of the water while > using standard library facilities for the basic building blocks. > > The status quo is that I won't even try, because I can't get rid of > locale-dependent parsing / output formatting of numbers when using > the standard library. > > Thanks, > Jens > > -- > > --- > You received this message because you are subscribed to the Google Groups "ISO C++ Standard - Future Proposals" group. > To unsubscribe from this group and stop receiving emails from it, send an email to std-proposals+unsubscribe@isocpp.org. > To post to this group, send email to std-proposals@isocpp.org. > Visit this group at http://groups.google.com/a/isocpp.org/group/std-proposals/. -- --- You received this message because you are subscribed to the Google Groups "ISO C++ Standard - Future Proposals" group. To unsubscribe from this group and stop receiving emails from it, send an email to std-proposals+unsubscribe@isocpp.org. To post to this group, send email to std-proposals@isocpp.org. Visit this group at http://groups.google.com/a/isocpp.org/group/std-proposals/. .