From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Original-To: caml-list@sympa.inria.fr Delivered-To: caml-list@sympa.inria.fr Received: from mail3-relais-sop.national.inria.fr (mail3-relais-sop.national.inria.fr [192.134.164.104]) by sympa.inria.fr (Postfix) with ESMTPS id 381B6800B6 for ; Sun, 1 Jan 2017 18:58:50 +0100 (CET) From: Pierre Weis X-IronPort-AV: E=Sophos;i="5.33,432,1477954800"; d="scan'208";a="206623918" Received: from yquem.paris.inria.fr (HELO yquem.inria.fr) ([128.93.101.33]) by mail3-relais-sop.national.inria.fr with ESMTP; 01 Jan 2017 18:58:49 +0100 Received: by yquem.inria.fr (Postfix, from userid 24253) id 962132016; Sun, 1 Jan 2017 18:58:49 +0100 (CET) Date: Sun, 1 Jan 2017 18:58:49 +0100 To: Ivan Gotovchits Cc: Gabriel Scherer , caml-list , Pierre Weis , Richard.Bonichon@gmail.com Message-ID: <20170101175849.GA21907@yquem.inria.fr> References: MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.21 (2010-09-15) Subject: Re: [Caml-list] Deprecation of tabulation boxes Hi Ivan, hi Gabriel, Sorry for the late answer: I generally have no time to read the Caml-list... As Gabriel said, I deprecated the tabulation boxes after a discussion about their not so smooth integration with other features of the Format module. And as you mentioned it, the discussion can be summerize as ``we do not understand the feature and do not use it'', and so probably nobody use it. I should have been more careful and wait to check if some users in the field were using tabulation boxes in their code. The problem about tabulation boxes is that they cannot be easily mixed with other boxes; on the other hand, if you do know the fact, they are perfectly usable to do exactly what you are doing with tabulation boxes, i.e. basic and simple tabular printing. That's exactly why those boxes have been introduced in Format in the first place: to print text file using lines with fixed fields separated with spaces or tabs (basic tab-tab-return line format). However, the code for tabulation boxes is still there and still up and running. So reverting its status is easy and should not break anything: it could be done anytime. During last year, Richard Bonichon and I wrote a paper about Format, its features, usage, internals and principles, to be presented at the next JFLA'2017 conference. During this work, we had long brain storming sessions about the new features I'm planning to introduce in the module and in particular a new tabular printing feature that would subsume tabulation boxes, since tabular printing would be fully integrated to the regular box management (and would have a proper format string extension to express tabular printing). This is still paper work, but we are confident to get a draft of the code during 2017. So, now that we know that some people indeed use the tabulation box feature, I propose to revert its status. When the new tabular printing feature will be there, people could easily port their code to the new feature and we could definitely deprecate the old tabulation boxes. PS: I wish you the best for this new year :) -- Pierre Weis INRIA Paris, http://bat8.inria.fr/~weis/ > Well, the discussion literally says, we don't understand tabulation, and it > looks like nobody is using it, so let's just throw it away :) > > We're using tabulation, and they are quite useful. Yep, I agree, that the > interface for setting the marks is kind of awkward, > it would be nice if there would be a `pp_setup_tabs : _ formatter -> int > list -> unit` function, that would push into the stack a new tabular box > with > the specified tabulations. > > What concerning your solution with the alignment, it is less general and > doesn't work in our case. First of all, the pretty printing functions, that > are printing > into the columns are not specified with `%s`, but with `%a` (e.g., address, > memory string, assembly string). Second, the printing functions are not > actually > defined in the same module. The tabs are initialized in the frontend, that > defines it based on the architecture (address size, maximum length of > instruction, etc), > and pretty printers are registered separately, and they just rely on a > fact, that we have three columns, the first is for address, the second is > for memory, then assembly, etc). > This design simplifies the actual instruction printers, by consolidating > the common code in the formatter setup procedure. > > So, it is still not clear to me, why the tabulations are wrong. If they do > complicate the code base and raise the support cost, then it is > understandable, why you would like to remove it > from the standard library. But in this case, it would be nice, to move this > code out as a separate library, as just removing a feature, that was in the > language for years (I would say even forever, > if we will start the history from caml-light), without providing any > substitution is... not nice :) > > > [1]: http://caml.inria.fr/pub/docs/manual-caml-light/node15.5.html > > > > On Mon, Dec 19, 2016 at 1:20 PM, Gabriel Scherer > wrote: > > > You may be interested in the discussion > > https://github.com/ocaml/ocaml/pull/229 > > which discussed a few ways in which the proposed tabulation interface > > may be inconvenient. It is after this discussion that Pierre Weis > > decided to deprecate tabulation boxes -- I believe that the reason is > > that tabulation and formatting never mixed very well. > > > > Note that if you can decide in advance a maximal size for a given > > "column" of your formatted output, you can use the left or > > right-justification features of formatting conversions to print > > aligned text: > > > > let data = [("x", "foo"); ("loop", "bar")] > > > > let () = > > print_newline (); > > data |> List.iter (fun (lab, instr) -> Printf.printf "%5s: %s\n" lab > > instr) > > (* > > x: foo > > loop: bar > > *) > > > > let () = > > print_newline (); > > data |> List.iter (fun (lab, instr) -> Printf.printf "%-5s: %s\n" lab > > instr) > > (* > > x : foo > > loop : bar > > *) > > > > let () = > > let len = List.fold_left (fun m (lab, _) -> max m (String.length > > lab)) 0 data in > > print_newline (); > > data |> List.iter (fun (lab, instr) -> Printf.printf "%*s: %s\n" len > > lab instr) > > (* > > x: foo > > loop: bar > > *) > > > > On Mon, Dec 19, 2016 at 12:59 PM, Ivan Gotovchits wrote: > > > Greetings, > > > > > > The tabulation boxes are marked as deprecated since 4.03.0. I've tried to > > > google for > > > any reasons that justify the removal but found only a note by Pierre > > Weis in > > > the Matis issue tracker[1]: > > > > > > > > >> The proposed printf-like syntax is fine, but tabulation boxes are now > > >> deprecated. > > >> Indeed, tabulation boxes interaction with other pretty-printing boxes > > have > > >> never been sorted out and tabulation boxes usage is orthogonal to the > > rest > > >> of the Format module. > > >> If considered useful, tabulation boxes could be implemented out of the > > >> Format module. > > > > > > > > > First of all the tabulation boxes can't be implemented outside of the > > format > > > module since the tab stops are actually stored in the stack of tabulation > > > boxes. If this data field would be removed from the formatter we will > > need > > > to pass an extra argument to all pretty-printers that use the tabulation > > > break, or use some global variable. Neither solution can be considered > > > acceptable. > > > > > > Speaking about the usefulness. The tabulation boxes are useful for > > printing > > > assembly outputs. And since compiler writing is sort of an application > > area > > > for OCaml, it shouldn't be considered as a rare case. It is also very > > useful > > > for printing Fortran code, that can be considered an assembler for the > > > numeric computing. It also just allows printing nicely formatted texts, > > that > > > it the main purpose of the Format library. As an example, tabulation > > boxes > > > are used in BAP and CIL frameworks. > > > > > > To summarize, the deprecation will eventually make few project > > > non-compilable. And there is no clear substitution for the deprecated > > > feature. > > > > > > Given that, I would like to hear the justifications for the deprecation > > of > > > tabulation boxes and suggested workarounds. > > > > > > One possible workaround, that I could see, is making the `formatter` type > > > extensible with existential boxes or, more generally, with existential > > > attributes. In that case, we will indeed be able to implement tabulation > > > boxes outside of the format module. > > > > > > Best wishes, > > > Ivan Gotovchits > > > > > > [1]: https://caml.inria.fr/mantis/view.php?id=4665 > >