220 32548 <5b87a9c6-b15d-465d-9d42-1620842e4f86@isocpp.org> article Path: news.gmane.org!.POSTED!not-for-mail From: ma.kalbfuss@web.de Newsgroups: gmane.comp.lang.c++.isocpp.proposals Subject: Re: relaxing rules for ternary operator. Allow incompatible types. Date: Sun, 21 May 2017 07:07:54 -0700 (PDT) Lines: 220 Approved: news@gmane.org Message-ID: <5b87a9c6-b15d-465d-9d42-1620842e4f86@isocpp.org> References: <1b5ee8eb-53df-4e98-af2f-829c7bc2e5b2@isocpp.org> <9064929.QEV8q21eIZ@tjmaciei-mobl1> <99350a9e-0c6b-468d-9761-f2b2b052275e@isocpp.org> <795341bd-d042-49c9-9eba-6b8ac6ca8a5e@isocpp.org> Reply-To: std-proposals@isocpp.org NNTP-Posting-Host: blaine.gmane.org Mime-Version: 1.0 Content-Type: multipart/mixed; boundary="----=_Part_2118_1299156135.1495375674459" X-Trace: blaine.gmane.org 1495375675 6460 195.159.176.226 (21 May 2017 14:07:55 GMT) X-Complaints-To: usenet@blaine.gmane.org NNTP-Posting-Date: Sun, 21 May 2017 14:07:55 +0000 (UTC) Cc: ma.kalbfuss@web.de To: ISO C++ Standard - Future Proposals Original-X-From: std-proposals+bncBC7YTI72XQOBBO56Q3EQKGQEGDM3M4I@isocpp.org Sun May 21 16:07:51 2017 Return-path: Envelope-to: gclcip-std-proposals@m.gmane.org Original-Received: from mail-pg0-f69.google.com ([74.125.83.69]) by blaine.gmane.org with esmtp (Exim 4.84_2) (envelope-from ) id 1dCRWQ-0001a1-Mh for gclcip-std-proposals@m.gmane.org; Sun, 21 May 2017 16:07:51 +0200 Original-Received: by mail-pg0-f69.google.com with SMTP id l125sf74783422pga.4 for ; Sun, 21 May 2017 07:07:56 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=isocpp-org.20150623.gappssmtp.com; s=20150623; h=date:from:to:cc:message-id:in-reply-to:references:subject :mime-version:x-original-sender:reply-to:precedence:mailing-list :list-id:x-spam-checked-in-group:list-post:list-help:list-archive :list-subscribe:list-unsubscribe; bh=tuhBg83H2RvGEWeIQHRxNSqRhxHY5i6NisxxGKAVdVw=; b=v5hFx+EL9VVNfWOLDA0LZ5AIbLbxv2iFzicDnii134q53uvd9GCUqnO1FpoWysLvNT rhPPfnPpNTNvR0AyJVhM4TIFkd8yJ1Oz/kq6uA/tho00N5ToqXDW7+enTFLCb6Hxldk+ tFMn/gCEKFA8YkjKJmdPfsne8QI9qHhORdOFeUFeVrqpHJJ6m1w8XDchF7IZ2J4pMwYd uPqZc4c7qjFVWUM6LLNKnoouTYipmsNUvHEvLuv+g1dlcF6/BSpK0n4HPO0Dyt5JpoR8 f7Y2oQrENAj/7JDH2987elo7tNbDq7rvJezRa05CoYFl1P54flgmnT3Pgmb9tUdQa5Fw hX2w== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:date:from:to:cc:message-id:in-reply-to :references:subject:mime-version:x-original-sender:reply-to :precedence:mailing-list:list-id:x-spam-checked-in-group:list-post :list-help:list-archive:list-subscribe:list-unsubscribe; bh=tuhBg83H2RvGEWeIQHRxNSqRhxHY5i6NisxxGKAVdVw=; b=gNormlGCCwRcFN5dkD0dDWSWMmm/ZbAqPWGGsTULgimHyUd4EN1BGsOscAHhyJVhP3 L+qOUT0BckF8HPE/PmzJdYc3gNUB4u3E1NlxJ+c+uo04UzTV8sKm4fkAIV9vSyQo3ZKQ Rk0V5eq3ojfoyMj/INNfDDbbodhaBUgCSq4s1yswKaovDl4+ro2gFf/GvpcZuJ1D6ww2 rCO/3q1zkUP3LCk/qhTw30f6xgXzVTXNlGLyIwguFE9i8CdmEchUJTFWH6sxynkgWmNM okm2gnqkx19X3yBOD/VzIbT/YbVbzOfUGyRknoYN/j2vCRU2iPBFXETIJuF4Wv4Dp8kF aqog== X-Gm-Message-State: AODbwcD4uCBW6N6dtWjw+GHrPKGcrEMj/IB7AoxPPxw82k8Qf/+UMuy5 Rb4MPvqYSLgjF9MK X-Received: by 10.99.117.84 with SMTP id f20mr8567729pgn.149.1495375675838; Sun, 21 May 2017 07:07:55 -0700 (PDT) X-BeenThere: std-proposals@isocpp.org Original-Received: by 10.157.32.134 with SMTP id x6ls2233119ota.47.gmail; Sun, 21 May 2017 07:07:54 -0700 (PDT) X-Received: by 10.157.40.242 with SMTP id s105mr390007ota.5.1495375674928; Sun, 21 May 2017 07:07:54 -0700 (PDT) In-Reply-To: <795341bd-d042-49c9-9eba-6b8ac6ca8a5e@isocpp.org> X-Original-Sender: ma.kalbfuss@web.de 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: , Xref: news.gmane.org gmane.comp.lang.c++.isocpp.proposals:32548 Archived-At: ------=_Part_2118_1299156135.1495375674459 Content-Type: multipart/alternative; boundary="----=_Part_2119_2006947193.1495375674459" ------=_Part_2119_2006947193.1495375674459 Content-Type: text/plain; charset="UTF-8" Am Sonntag, 21. Mai 2017 15:00:03 UTC+2 schrieb Mingxin Wang: > > You may as well consider the following situations: > > *Loops* > > It is difficult for compilers to generate code for loops with the feature > you proposed, e.g.: > > void foo(int n) { > /* A: Do something before the loop */ > for (int i = 0; i < n; ++i) { > /* B: Do something */ > auto x = runtime_condition() ? A{} : B{}; > /* C: Do some other things with x */ > } > /* D: Do something after the loop */ > } > *Code Snippet 1* > > As far as I am concerned, here is the equivalent of the code that the > compiler would automatically generate for the code above (with your method): > > template > void bar(int i, int n, T x) { > /* C: Do some other things with x */ > ++i; > if (i < n) { > /* B: Do something */ > if (runtime_condition()) { > bar(i, n, A{}); > } else { > bar(i, n, B{}); > } > } > } > > void foo(int n) { > /* A: Do something before the loop */ > if (0 < n) { > /* B: Do something */ > if (runtime_condition()) { > bar(1, n, A{}); > } else { > bar(1, n, B{}); > } > } > /* D: Do something after the loop */ > } > *Code Snippet 2* > > It is too complicated! Moreover, as the recursive call of the "function > template bar" is not able to be optimized, it will possibly increase the > runtime overhead. > > *Size* > > Compilers may generate a lot of redundant code when the feature is used > extensively, e.g.: > > auto x0 = runtime_condition_0() ? A0{} : B0{}; > auto x1 = runtime_condition_1() ? A1{} : B1{}; > // ... > auto xn = runtime_condition_n() ? An{} : Bn{}; > *Code Snippet 3* > > As you introduced, the compiler is supposed to generate code for every > path! For the code above, it requires O(2^n) size for code generation, > which is NOT acceptable in most cases. > > In fact, the challenges above can be solved gracefully with the "proxies" > and the "wrappers". > > As the condition is determined at runtime, the concrete type cannot be > deduced at compile time. The only reason for us to save the returned value > is that *they have similar expressions and semantics*. We can simply > extract the common expressions of the types which are used in the context > to a "proxy", and save the returned value with a "wrapper", e.g. Code > Snippet 1 can be reconstructed as is shown below (with better performance > and smaller size): > > /* Declares the common expressions required in the context */ > proxy P { > void f_0(); > double f_1(int, int); > }; > > void foo(int n) { > /* A: Do something before the loop */ > for (int i = 0; i < n; ++i) { > /* B: Do something */ > Wrapper

x(runtime_condition() ? Wrapper

(A{}) : Wrapper

(B{})); > P p = x.get_proxy(); > /* C: Do some other things with p */ > } > /* D: Do something after the loop */ > } > *Code Snippet 4* > > Mingxin Wang > -- 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. To view this discussion on the web visit https://groups.google.com/a/isocpp.org/d/msgid/std-proposals/5b87a9c6-b15d-465d-9d42-1620842e4f86%40isocpp.org. ------=_Part_2119_2006947193.1495375674459 Content-Type: text/html; charset="UTF-8" Content-Transfer-Encoding: quoted-printable

Am Sonntag, 21. Mai 2017 15:00:03 UTC+2 schrieb Mingxin Wa= ng:
You may as well consider the following situations:=

<= div style=3D"font-family:georgia,serif">Loops

It is difficult for compilers to generate code for = loops with the feature you proposed, e.g.:

void foo(int n) {
=C2=A0 /* A:= Do something before the loop */
=C2=A0 for (int i =3D 0; i < = n; ++i) {
=C2=A0 =C2=A0 /* B: Do something */
=C2=A0 = =C2=A0 auto x =3D runtime_condition() ? A{} : B{};
=C2=A0 =C2=A0 = /* C: Do some other things with x */
=C2=A0 }
=C2=A0 /*= D: Do something after the loop */
}
Code Snippet 1

As far as I am concerned, here is the equivalent of the code that the c= ompiler would automatically generate for the code above (with your method):=

template <cl= ass T>
void bar(int i, int n, T x) {
=C2=A0 /* C: Do= some other things with x */
=C2=A0 ++i;
=C2=A0 if (i &= lt; n) {
=C2=A0 =C2=A0 /* B: Do something */
=C2=A0 =C2= =A0 if (runtime_condition()) {
=C2=A0 =C2=A0 =C2=A0 bar(i, n, A{}= );
=C2=A0 =C2=A0 } else {
=C2=A0 =C2=A0 =C2=A0 bar(i, n= , B{});
=C2=A0 =C2=A0 }
=C2=A0 }
}
=
void foo(int n) {
=C2=A0 /* A: Do something before= the loop */
=C2=A0 if (0 < n) {
=C2=A0 =C2=A0 /* B:= Do something */
=C2=A0 =C2=A0 if (runtime_condition()) {
=C2=A0 =C2=A0 =C2=A0 bar(1, n, A{});
=C2=A0 =C2=A0 } else {
=C2=A0 =C2=A0 =C2=A0 bar(1, n, B{});
=C2=A0 =C2=A0 }
=C2=A0 }
=C2=A0 /* D: Do something after the loop */
<= div>}
Code Snippet 2

It = is too complicated! Moreover, as the recursive call of the "function t= emplate bar" is not able to be optimized, it will possibly increase th= e runtime overhead.

Size

Compilers may generate a lot of redundant code when the fea= ture is used extensively, e.g.:

auto x0 =3D runtime_condition_0() ? A0{} : = B0{};
auto x1 =3D runtime_condition_1() ? A1{} : B1{};
= // ...
auto xn =3D runtime_condition_n() ? An{} : Bn{};
Code Snippet 3
=

As you introduced, the compiler is supposed to generate= code for every path! For the code above, it requires O(2^n) size for code = generation, which is NOT acceptable in most cases.

In fact, the challenges above can be solved gracefully with the "prox= ies" and the "wrappers".

As the con= dition is determined at runtime, the concrete type cannot be deduced at com= pile time. The only reason for us to save the returned value is that the= y have similar expressions and semantics. We can simply extract the com= mon expressions of the types which are used in the context to a "proxy= ", and save the returned value with a "wrapper", e.g. Code S= nippet 1 can be reconstructed as is shown below (with better performance an= d smaller size):

<= /font>
/* Declares the common expressions required in the contex= t */
proxy P {
=C2=A0 void f_0();
=C2=A0 doub= le f_1(int, int);
};

void foo(int n) {
=C2=A0 /* A: Do something before the loop */
=C2=A0 for = (int i =3D 0; i < n; ++i) {
=C2=A0 =C2=A0 /* B: Do something *= /
=C2=A0 =C2=A0 Wrapper<P> x(runtime_condition() ? Wrapper&= lt;P>(A{}) : Wrapper<P>(B{}));
=C2=A0 =C2=A0 P p =3D x.g= et_proxy();
=C2=A0 =C2=A0 /* C: Do some other things with p */
=C2=A0 }
=C2=A0 /* D: Do something after the loop */
}
Code Snippet 4

Mingxin Wang=

--
You received this message because you are subscribed to the Google Groups &= quot;ISO C++ Standard - Future Proposals" group.
To unsubscribe from this group and stop receiving emails from it, send an e= mail to std-proposa= ls+unsubscribe@isocpp.org.
To post to this group, send email to std-proposals@isocpp.org.
To view this discussion on the web visit https://groups.google.= com/a/isocpp.org/d/msgid/std-proposals/5b87a9c6-b15d-465d-9d42-1620842e4f86= %40isocpp.org.
------=_Part_2119_2006947193.1495375674459-- ------=_Part_2118_1299156135.1495375674459-- .