<div dir="ltr"><div class="gmail_extra"><br><div class="gmail_quote">On Mon, Feb 8, 2016 at 7:39 PM, Xinliang David Li <span dir="ltr"><<a href="mailto:davidxl@google.com" target="_blank">davidxl@google.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">I took a look at the problem. The implicitly defaulted operators<br>
should not be instrumented as specified -- I actually I just added the<br>
new test case for that (checking profile counter not generated) right<br>
after my previous reply and it still passes with this patch. The<br>
reason is that those operators have 'implicit' bit set, and profile<br>
instrumentation in FE is implemented in two stages: 1) counter<br>
assignment; 2) actual transformation. For methods with implicit bit<br>
set, step 1) is skipped as designed, so step 2) simply becomes a<br>
no-op.<br>
<br>
In short, the test case still needs explicit '=default', and the<br>
implicit case is covered elsewhere.<br></blockquote><div><br></div><div>OK, thanks for the explanation!<br><br>Why is that the case, though - why would an implicitly default function be any different from a profile (& profile-guided-optimization) perspective from an explicitly defaulted one?<br></div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<br>
thanks,<br>
<br>
David<br>
<div class="HOEnZb"><div class="h5"><br>
On Mon, Feb 8, 2016 at 5:23 PM, David Blaikie <<a href="mailto:dblaikie@gmail.com">dblaikie@gmail.com</a>> wrote:<br>
><br>
><br>
> On Mon, Feb 8, 2016 at 5:05 PM, Xinliang David Li <<a href="mailto:davidxl@google.com">davidxl@google.com</a>><br>
> wrote:<br>
>><br>
>> ha! somehow I kept thinking you are referring to implicit declared ctors.<br>
><br>
><br>
> Ah, glad we figured out the disconnect - thanks for bearing with me!<br>
><br>
>><br>
>><br>
>> From your test case, it is seems that the implicit copy/move op is<br>
>> also broken and is fixed by this patch too. That means a missing test<br>
>> case to me. Will update the case when verified.<br>
><br>
><br>
> Again, this is a case where I'd probably just simplify the test, as I asked<br>
> earlier in the thread (I asked if it mattered if the op was explicitly or<br>
> implicitly defaulted (& your response: "> Is the fix/codepath specifically<br>
> about explicitly defaulted ops?<br>
><br>
> yes -- explicitly defaulted. There are some test coverage already for<br>
> implicitly declared ctors (but not assignment op -- probably worth<br>
> adding some testing too).")<br>
><br>
> So I'd just simplify the test by removing the "= default" - I don't think<br>
> there's value in testing both the explicit default and implicit default if<br>
> it's just the default-y-ness that's relevant here. Otherwise we could end up<br>
> testing all sorts of ways of writing/interacting with dtors which wouldn't<br>
> be relevant to the code/fix/etc.<br>
><br>
> This seems like the obvious test for the behavior:<br>
><br>
> struct foo {<br>
> // non-trivial ops<br>
> foo &operator=(const foo&);<br>
> foo &operator=(foo&&);<br>
> };<br>
><br>
> struct bar {<br>
> foo f; // or derive bar from foo, but I think the member version is<br>
> simpler<br>
> };<br>
><br>
> // force emission of bar's implicit special members, one way or another:<br>
> bar &(bar::*x)(const bar&) = &bar::operator=;<br>
> bar &(bar::*x)(bar&&) = &bar::operator=;<br>
><br>
> (or just call them as you had in your test case - but that produces more<br>
> code, etc in the module, extra functions/profile counters/etc)<br>
><br>
> - Dave<br>
><br>
>><br>
>><br>
>> thanks,<br>
>><br>
>> David<br>
>><br>
>><br>
>> On Mon, Feb 8, 2016 at 4:58 PM, David Blaikie <<a href="mailto:dblaikie@gmail.com">dblaikie@gmail.com</a>> wrote:<br>
>> ><br>
>> ><br>
>> > On Mon, Feb 8, 2016 at 4:31 PM, Xinliang David Li <<a href="mailto:davidxl@google.com">davidxl@google.com</a>><br>
>> > wrote:<br>
>> >><br>
>> >> On Mon, Feb 8, 2016 at 4:05 PM, David Blaikie <<a href="mailto:dblaikie@gmail.com">dblaikie@gmail.com</a>><br>
>> >> wrote:<br>
>> >> ><br>
>> >> ><br>
>> >> > On Mon, Feb 8, 2016 at 3:58 PM, Xinliang David Li<br>
>> >> > <<a href="mailto:davidxl@google.com">davidxl@google.com</a>><br>
>> >> > wrote:<br>
>> >> >><br>
>> >> >> To be clear, you are suggesting breaking the test into two (one for<br>
>> >> >> copy, and one for move) ? I am totally fine with that.<br>
>> >> ><br>
>> >> ><br>
>> >> > Nah, no need to split the test case - we try to keep the number of<br>
>> >> > test<br>
>> >> > files down (& group related tests into a single file) to reduce test<br>
>> >> > execution time (a non-trivial about of check time is spent in process<br>
>> >> > overhead).<br>
>> >> ><br>
>> >> >><br>
>> >> >> I thought you<br>
>> >> >> suggested removing the testing of move/op case because they might<br>
>> >> >> share the same code path (clang's implementation) as the copy/op.<br>
>> >> ><br>
>> >> ><br>
>> >> > I was suggesting that two cases is no big deal whether you test both<br>
>> >> > or<br>
>> >> > test<br>
>> >> > one if they're the same codepath - if there were 5/many more things<br>
>> >> > that<br>
>> >> > shared the same codepath, I'd generally suggest testing a<br>
>> >> > representative<br>
>> >> > sample (possibly just a single one) rather than testing every client<br>
>> >> > of<br>
>> >> > the<br>
>> >> > same code.<br>
>> >> ><br>
>> >> > Feel free to leave the two here as-is. (though if we're talking about<br>
>> >> > test<br>
>> >> > granularity, it's probably worth just putting these cases in the same<br>
>> >> > file/type/etc as the ctor cases you mentioned were already covered)<br>
>> >><br>
>> >> There is a balance somewhere:<br>
>> >> 1) for small test cases like this, the overhead mainly comes from test<br>
>> >> set up cost -- adding additional incremental test in the same file<br>
>> >> probably almost comes for free (in terms of cost). However<br>
>> >> 2) having too many cases grouped together also reduces the<br>
>> >> debuggability when some test fails.<br>
>> ><br>
>> ><br>
>> > Yep, for sure. In this case, testing the ctors and assignment ops in one<br>
>> > file's probably not a bad tradeoff (you can see how Clang groups its<br>
>> > tests -<br>
>> > a file per language feature in many cases, exploring the myriad ways the<br>
>> > feature can be used - this doesn't always work spectacularly (when you<br>
>> > can't<br>
>> > order the IR emission to happen mostly in the order that the source is<br>
>> > written (rather being interleaved))<br>
>> ><br>
>> > Anyway, up to you - that part isn't something I'm terribly worried about<br>
>> > either way.<br>
>> ><br>
>> >><br>
>> >><br>
>> >> ><br>
>> >> > & I'm still curious/wondering if there's a common codepath that would<br>
>> >> > provide a simpler fix/code that solved both implicit and explicitly<br>
>> >> > defaulted ops.<br>
>> >><br>
>> >> I may take a look at that when I find time -- but there is no guarantee<br>
>> >> :)<br>
>> ><br>
>> ><br>
>> > A quick test of putting "assert(false)" in<br>
>> > emitImplicitAssignmentOperatorBody and running Clang over this code:<br>
>> ><br>
>> > struct foo {<br>
>> > foo &operator=(const foo &);<br>
>> > };<br>
>> ><br>
>> > struct bar {<br>
>> > foo f;<br>
>> > };<br>
>> ><br>
>> > auto (bar::*x)(const bar&) = &bar::operator=;<br>
>> ><br>
>> > Fires the assertion - this seems to me to indicate that the codepath you<br>
>> > changed is used for both the explicitly (based on the change fixing your<br>
>> > test case) and implicitly defaulted (based on my test case) cases.<br>
>> ><br>
>> > Is it possible that you end up with duplicate counters by accident in<br>
>> > this<br>
>> > path, then? Or at least that whatever codepath was handling the<br>
>> > implicitly<br>
>> > defaulted ones is now redundant with this one?<br>
>> ><br>
>> > Actually, so far as I can tell this doesn't work for implicitly<br>
>> > defaulted<br>
>> > move ops (the above test case doesn't have an add pgocount in it) -<br>
>> > perhaps<br>
>> > I'm missing something/doing it wrong? or was just not communicating<br>
>> > clearly<br>
>> > regarding explicit versus implicitly defaulted special members.<br>
>> ><br>
>> > - Dave<br>
>> ><br>
>> >><br>
>> >><br>
>> >> thanks,<br>
>> >><br>
>> >> David<br>
>> >><br>
>> >><br>
>> >><br>
>> >> ><br>
>> >> > - Dave<br>
>> >> ><br>
>> >> >><br>
>> >> >><br>
>> >> >> thanks,<br>
>> >> >><br>
>> >> >> David<br>
>> >> >><br>
>> >> >> On Mon, Feb 8, 2016 at 3:52 PM, David Blaikie <<a href="mailto:dblaikie@gmail.com">dblaikie@gmail.com</a>><br>
>> >> >> wrote:<br>
>> >> >> ><br>
>> >> >> ><br>
>> >> >> > On Mon, Feb 8, 2016 at 3:46 PM, Xinliang David Li<br>
>> >> >> > <<a href="mailto:davidxl@google.com">davidxl@google.com</a>><br>
>> >> >> > wrote:<br>
>> >> >> >><br>
>> >> >> >> On Mon, Feb 8, 2016 at 3:35 PM, David Blaikie<br>
>> >> >> >> <<a href="mailto:dblaikie@gmail.com">dblaikie@gmail.com</a>><br>
>> >> >> >> wrote:<br>
>> >> >> >> ><br>
>> >> >> >> ><br>
>> >> >> >> > On Mon, Feb 8, 2016 at 3:21 PM, Xinliang David Li<br>
>> >> >> >> > <<a href="mailto:davidxl@google.com">davidxl@google.com</a>><br>
>> >> >> >> > wrote:<br>
>> >> >> >> >><br>
>> >> >> >> >> On Mon, Feb 8, 2016 at 3:17 PM, David Blaikie<br>
>> >> >> >> >> <<a href="mailto:dblaikie@gmail.com">dblaikie@gmail.com</a>><br>
>> >> >> >> >> wrote:<br>
>> >> >> >> >> ><br>
>> >> >> >> >> ><br>
>> >> >> >> >> > On Mon, Feb 8, 2016 at 12:07 PM, Xinliang David Li<br>
>> >> >> >> >> > <<a href="mailto:davidxl@google.com">davidxl@google.com</a>><br>
>> >> >> >> >> > wrote:<br>
>> >> >> >> >> >><br>
>> >> >> >> >> >> On Mon, Feb 8, 2016 at 11:39 AM, David Blaikie<br>
>> >> >> >> >> >> <<a href="mailto:dblaikie@gmail.com">dblaikie@gmail.com</a>><br>
>> >> >> >> >> >> wrote:<br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> > On Mon, Feb 8, 2016 at 9:25 AM, David Li via llvm-commits<br>
>> >> >> >> >> >> > <<a href="mailto:llvm-commits@lists.llvm.org">llvm-commits@lists.llvm.org</a>> wrote:<br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >> davidxl updated this revision to Diff 47217.<br>
>> >> >> >> >> >> >> davidxl added a comment.<br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >> Simplified test case suggested by Vedant.<br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >> <a href="http://reviews.llvm.org/D16947" rel="noreferrer" target="_blank">http://reviews.llvm.org/D16947</a><br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >> Files:<br>
>> >> >> >> >> >> >> lib/CodeGen/CGClass.cpp<br>
>> >> >> >> >> >> >> test/Profile/def-assignop.cpp<br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >> Index: test/Profile/def-assignop.cpp<br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >> ===================================================================<br>
>> >> >> >> >> >> >> --- test/Profile/def-assignop.cpp<br>
>> >> >> >> >> >> >> +++ test/Profile/def-assignop.cpp<br>
>> >> >> >> >> >> >> @@ -0,0 +1,32 @@<br>
>> >> >> >> >> >> >> +// RUN: %clang_cc1 -x c++ -std=c++11 %s -triple<br>
>> >> >> >> >> >> >> x86_64-unknown-linux-gnu<br>
>> >> >> >> >> >> >> -main-file-name def-assignop.cpp -o - -emit-llvm<br>
>> >> >> >> >> >> >> -fprofile-instrument=clang<br>
>> >> >> >> >> >> >> | FileCheck --check-prefix=PGOGEN %s<br>
>> >> >> >> >> >> >> +// RUN: %clang_cc1 -x c++ -std=c++11 %s -triple<br>
>> >> >> >> >> >> >> x86_64-unknown-linux-gnu<br>
>> >> >> >> >> >> >> -main-file-name def-assignop.cpp -o - -emit-llvm<br>
>> >> >> >> >> >> >> -fprofile-instrument=clang<br>
>> >> >> >> >> >> >> -fcoverage-mapping | FileCheck --check-prefix=COVMAP %s<br>
>> >> >> >> >> >> >> +<br>
>> >> >> >> >> >> >> +struct B {<br>
>> >> >> >> >> >> >> + void operator=(const B &b) {}<br>
>> >> >> >> >> >> >> + void operator=(const B &&b) {}<br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> > Probably best to make these canonical to avoid confusion:<br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> > B &operator=(const B&);<br>
>> >> >> >> >> >> > B &operator=(B&&);<br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> > (& they don't need definitions - just declarations)<br>
>> >> >> >> >> >><br>
>> >> >> >> >> >> Will change.<br>
>> >> >> >> >> >><br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> > Also, neither of these are the move /constructor/, just<br>
>> >> >> >> >> >> > the<br>
>> >> >> >> >> >> > move<br>
>> >> >> >> >> >> > operator.<br>
>> >> >> >> >> >> > Not sure if Vedant just used the wrong terminology, or<br>
>> >> >> >> >> >> > whether<br>
>> >> >> >> >> >> > it's<br>
>> >> >> >> >> >> > worth<br>
>> >> >> >> >> >> > testing the move/copy ctors too, to check that they do<br>
>> >> >> >> >> >> > the<br>
>> >> >> >> >> >> > right<br>
>> >> >> >> >> >> > thing<br>
>> >> >> >> >> >> > as<br>
>> >> >> >> >> >><br>
>> >> >> >> >> >> I added tests for copy ctors, and plan to add move ctor<br>
>> >> >> >> >> >> test<br>
>> >> >> >> >> >> soon.<br>
>> >> >> >> >> >><br>
>> >> >> >> >> >> > well. (if all of these things use the same codepath, I<br>
>> >> >> >> >> >> > don't<br>
>> >> >> >> >> >> > see a<br>
>> >> >> >> >> >> > great<br>
>> >> >> >> >> >> > benefit in having separate tests for them (but you can<br>
>> >> >> >> >> >> > add<br>
>> >> >> >> >> >> > them<br>
>> >> >> >> >> >> > here<br>
>> >> >> >> >> >> > if<br>
>> >> >> >> >> >> > you<br>
>> >> >> >> >> >> > like) - I'm just suggesting a manual verification in case<br>
>> >> >> >> >> >> > those<br>
>> >> >> >> >> >> > need<br>
>> >> >> >> >> >> > a<br>
>> >> >> >> >> >> > separate fix<br>
>> >> >> >> >> >><br>
>> >> >> >> >> >> the ctor and assignment op do not share the same path --<br>
>> >> >> >> >> >> the<br>
>> >> >> >> >> >> ctor<br>
>> >> >> >> >> >> path<br>
>> >> >> >> >> >> is working as expected without the fix -- or do you mean<br>
>> >> >> >> >> >> there<br>
>> >> >> >> >> >> is<br>
>> >> >> >> >> >> no<br>
>> >> >> >> >> >> need to cover both copy and move variants?<br>
>> >> >> >> >> ><br>
>> >> >> >> >> ><br>
>> >> >> >> >> > I wouldn't necessarily bother testing multiple instances of<br>
>> >> >> >> >> > the<br>
>> >> >> >> >> > same<br>
>> >> >> >> >> > codepath (so the copy and move ctor for example) - but 2<br>
>> >> >> >> >> > instances<br>
>> >> >> >> >> > is<br>
>> >> >> >> >> > no<br>
>> >> >> >> >> > big<br>
>> >> >> >> >> > deal (if there were several more, I might be inclined to<br>
>> >> >> >> >> > just<br>
>> >> >> >> >> > test<br>
>> >> >> >> >> > one<br>
>> >> >> >> >> > as a<br>
>> >> >> >> >> > representative sample). I don't mind either way, though. The<br>
>> >> >> >> >> > number<br>
>> >> >> >> >> > is<br>
>> >> >> >> >> > small<br>
>> >> >> >> >> > & the test cases are arguably distinct.<br>
>> >> >> >> >><br>
>> >> >> >> >> Sorry I disagree with your general statement here. I treat<br>
>> >> >> >> >> such<br>
>> >> >> >> >> test<br>
>> >> >> >> >> cases as 'black box testing' that do not know about the<br>
>> >> >> >> >> internal<br>
>> >> >> >> >> implementation (code path). It may or may not share the same<br>
>> >> >> >> >> code<br>
>> >> >> >> >> path<br>
>> >> >> >> >> today -- same is true in the future.<br>
>> >> >> >> ><br>
>> >> >> >> ><br>
>> >> >> >> > While there's merit in both approaches, practically speaking it<br>
>> >> >> >> > seems<br>
>> >> >> >> > difficult to test in that way in general - any feature could<br>
>> >> >> >> > interact<br>
>> >> >> >> > with<br>
>> >> >> >> > any other.<br>
>> >> >> >><br>
>> >> >> >> The language features are well specified -- so writing small test<br>
>> >> >> >> cases to cover them is a general accepted way of testing.<br>
>> >> >> ><br>
>> >> >> ><br>
>> >> >> > I'm not sure I follow the distinction you're drawing between the<br>
>> >> >> > middle<br>
>> >> >> > end<br>
>> >> >> > optimization tests and the features you're testing here. If the<br>
>> >> >> > features<br>
>> >> >> > are<br>
>> >> >> > relatively independent, even within the same API/feature area,<br>
>> >> >> > they're<br>
>> >> >> > generally tested independently (even two features within a single<br>
>> >> >> > middle<br>
>> >> >> > end<br>
>> >> >> > optimization - a test case is written to ensure that, say,<br>
>> >> >> > ArgumentPromotion<br>
>> >> >> > correctly handles debug info, and another that it correctly<br>
>> >> >> > handles<br>
>> >> >> > inalloca<br>
>> >> >> > (or fp80, etc - just looking at test/Transforms/ArgumentPromotion)<br>
>> >> >> > -<br>
>> >> >> > but<br>
>> >> >> > we<br>
>> >> >> > don't test the matrix of combinations of these features)<br>
>> >> >> ><br>
>> >> >> >><br>
>> >> >> >><br>
>> >> >> >> >The LLVM regression suite is far more narrowly targeted than<br>
>> >> >> >> > that<br>
>> >> >> >> > - we don't test combinations of optimizations, for example - we<br>
>> >> >> >> > test<br>
>> >> >> >> > each<br>
>> >> >> >> > optimization in isolation. The same would be true of two<br>
>> >> >> >> > independent<br>
>> >> >> >> > features on an interface such as this, I think.<br>
>> >> >> >><br>
>> >> >> >> This is a weakness of the test system -- a problem at a different<br>
>> >> >> >> dimension.<br>
>> >> >> ><br>
>> >> >> ><br>
>> >> >> > If we want to have a discussion about the LLVM community testing<br>
>> >> >> > methodology, that might be best taken up on llvm-dev (and<br>
>> >> >> > clang-dev).<br>
>> >> >> > But<br>
>> >> >> > for now, I'd ask that tests in the lit regression suite are<br>
>> >> >> > generally<br>
>> >> >> > as<br>
>> >> >> > isolated as possible and test one thing at a time.<br>
>> >> >> ><br>
>> >> >> >><br>
>> >> >> >><br>
>> >> >> >> ><br>
>> >> >> >> >><br>
>> >> >> >> >><br>
>> >> >> >> >> ><br>
>> >> >> >> >> >><br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >> +};<br>
>> >> >> >> >> >> >> +<br>
>> >> >> >> >> >> >> +struct A {<br>
>> >> >> >> >> >> >> + A &operator=(const A &) = default;<br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> > Is the fix/codepath specifically about explicitly<br>
>> >> >> >> >> >> > defaulted<br>
>> >> >> >> >> >> > ops?<br>
>> >> >> >> >> >><br>
>> >> >> >> >> >> yes -- explicitly defaulted. There are some test coverage<br>
>> >> >> >> >> >> already<br>
>> >> >> >> >> >> for<br>
>> >> >> >> >> >> implicitly declared ctors (but not assignment op --<br>
>> >> >> >> >> >> probably<br>
>> >> >> >> >> >> worth<br>
>> >> >> >> >> >> adding some testing too).<br>
>> >> >> >> >> ><br>
>> >> >> >> >> ><br>
>> >> >> >> >> > Hmm - are you sure there's no common codepath that would<br>
>> >> >> >> >> > cover<br>
>> >> >> >> >> > the<br>
>> >> >> >> >> > explicitly defaulted or implicitly defaulted ops together in<br>
>> >> >> >> >> > one<br>
>> >> >> >> >> > go?<br>
>> >> >> >> >><br>
>> >> >> >> >> Sorry I am not sure what you mean here.<br>
>> >> >> >> > Is there some part of Clang that is responsible for generating<br>
>> >> >> >> > both<br>
>> >> >> >> > implicitly defaulted and explicitly defaulted move/copy ops<br>
>> >> >> >> > that<br>
>> >> >> >> > could<br>
>> >> >> >> > handle this case, rather than apparently handling the implicit<br>
>> >> >> >> > and<br>
>> >> >> >> > explicit<br>
>> >> >> >> > cases separately (it seems they're being handled separately if<br>
>> >> >> >> > the<br>
>> >> >> >> > implicit<br>
>> >> >> >> > case worked before and you added code (rather than moving code)<br>
>> >> >> >> > to<br>
>> >> >> >> > fix<br>
>> >> >> >> > the<br>
>> >> >> >> > explicit case - it sounds like we now have two bits of code,<br>
>> >> >> >> > one<br>
>> >> >> >> > for<br>
>> >> >> >> > implicit and one for explicit - perhaps there's a single bit of<br>
>> >> >> >> > code<br>
>> >> >> >> > that we<br>
>> >> >> >> > could write that would handle both?)<br>
>> >> >> >><br>
>> >> >> >> The codegen paths are different -- otherwise as you commented,<br>
>> >> >> >> the<br>
>> >> >> >> implicit case would have been broken too.<br>
>> >> >> >><br>
>> >> >> >> Refactoring FE code to handle both is probably beyond the scope<br>
>> >> >> >> of<br>
>> >> >> >> this fix. Having a good test case here will exactly help avoid<br>
>> >> >> >> regression if that happens in the future.<br>
>> >> >> >><br>
>> >> >> >> David<br>
>> >> >> >><br>
>> >> >> >> ><br>
>> >> >> >> > - David<br>
>> >> >> >> ><br>
>> >> >> >> >><br>
>> >> >> >> >><br>
>> >> >> >> >> David<br>
>> >> >> >> >><br>
>> >> >> >> >> ><br>
>> >> >> >> >> >><br>
>> >> >> >> >> >><br>
>> >> >> >> >> >> > Or just any<br>
>> >> >> >> >> >> > compiler-generated ones? (you could drop these lines if<br>
>> >> >> >> >> >> > it's<br>
>> >> >> >> >> >> > about<br>
>> >> >> >> >> >> > any<br>
>> >> >> >> >> >> > compiler-generated ones, might be simpler/more obvious<br>
>> >> >> >> >> >> > that<br>
>> >> >> >> >> >> > it's<br>
>> >> >> >> >> >> > not<br>
>> >> >> >> >> >> > about<br>
>> >> >> >> >> >> > the "= default" feature)<br>
>> >> >> >> >> >><br>
>> >> >> >> >> >> Other compiler generated ones are handled differently.<br>
>> >> >> >> >> >><br>
>> >> >> >> >> >> thanks,<br>
>> >> >> >> >> >><br>
>> >> >> >> >> >> David<br>
>> >> >> >> >> >><br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >> + // PGOGEN: define {{.*}}@_ZN1AaSERKS_(<br>
>> >> >> >> >> >> >> + // PGOGEN: %pgocount = load {{.*}}<br>
>> >> >> >> >> >> >> @__profc__ZN1AaSERKS_<br>
>> >> >> >> >> >> >> + // PGOGEN: {{.*}}add{{.*}}%pgocount, 1<br>
>> >> >> >> >> >> >> + // PGOGEN: store{{.*}}@__profc__ZN1AaSERKS_<br>
>> >> >> >> >> >> >> + A &operator=(A &&) = default;<br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >> + // PGOGEN: define {{.*}}@_ZN1AaSEOS_<br>
>> >> >> >> >> >> >> + // PGOGEN: %pgocount = load {{.*}}<br>
>> >> >> >> >> >> >> @__profc__ZN1AaSEOS_<br>
>> >> >> >> >> >> >> + // PGOGEN: {{.*}}add{{.*}}%pgocount, 1<br>
>> >> >> >> >> >> >> + // PGOGEN: store{{.*}}@__profc__ZN1AaSEOS_<br>
>> >> >> >> >> >> >> +<br>
>> >> >> >> >> >> >> + // Check that coverage mapping includes 6 function<br>
>> >> >> >> >> >> >> records<br>
>> >> >> >> >> >> >> including<br>
>> >> >> >> >> >> >> the<br>
>> >> >> >> >> >> >> + // defaulted copy and move operators: A::operator=<br>
>> >> >> >> >> >> >> + // COVMAP: @__llvm_coverage_mapping = {{.*}} { { i32,<br>
>> >> >> >> >> >> >> i32,<br>
>> >> >> >> >> >> >> i32,<br>
>> >> >> >> >> >> >> i32<br>
>> >> >> >> >> >> >> },<br>
>> >> >> >> >> >> >> [5 x <{{.*}}>],<br>
>> >> >> >> >> >> >> + B b;<br>
>> >> >> >> >> >> >> +};<br>
>> >> >> >> >> >> >> +<br>
>> >> >> >> >> >> >> +int main() {<br>
>> >> >> >> >> >> >> + A a1, a2;<br>
>> >> >> >> >> >> >> + a1 = a2;<br>
>> >> >> >> >> >> >> + a2 = static_cast<A &&>(a1);<br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> > An option, though not necessarily better, would be to<br>
>> >> >> >> >> >> > just<br>
>> >> >> >> >> >> > take<br>
>> >> >> >> >> >> > the<br>
>> >> >> >> >> >> > address<br>
>> >> >> >> >> >> > of the special members:<br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> > auto (B::*x)(const B&) = &bar::operator=;<br>
>> >> >> >> >> >> > auto (B::*x)(B&&) = &bar::operator=;<br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> > In short, what I'm picturing, in total:<br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> > struct A {<br>
>> >> >> >> >> >> > A &operator=(const A&);<br>
>> >> >> >> >> >> > A &operator=(A&&);<br>
>> >> >> >> >> >> > };<br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> > struct B {<br>
>> >> >> >> >> >> > A a;<br>
>> >> >> >> >> >> > };<br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> > auto (B::*x)(const B&) = &B::operator=;<br>
>> >> >> >> >> >> > auto (B::*x)(B&&) = &B::operator=;<br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> > Also, this test should probably be in clang, since it's a<br>
>> >> >> >> >> >> > clang<br>
>> >> >> >> >> >> > code<br>
>> >> >> >> >> >> > change/fix.<br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >> + return 0;<br>
>> >> >> >> >> >> >> +}<br>
>> >> >> >> >> >> >> Index: lib/CodeGen/CGClass.cpp<br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >> ===================================================================<br>
>> >> >> >> >> >> >> --- lib/CodeGen/CGClass.cpp<br>
>> >> >> >> >> >> >> +++ lib/CodeGen/CGClass.cpp<br>
>> >> >> >> >> >> >> @@ -1608,6 +1608,7 @@<br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >> LexicalScope Scope(*this, RootCS->getSourceRange());<br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >> + incrementProfileCounter(RootCS);<br>
>> >> >> >> >> >> >> AssignmentMemcpyizer AM(*this, AssignOp, Args);<br>
>> >> >> >> >> >> >> for (auto *I : RootCS->body())<br>
>> >> >> >> >> >> >> AM.emitAssignment(I);<br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >> _______________________________________________<br>
>> >> >> >> >> >> >> llvm-commits mailing list<br>
>> >> >> >> >> >> >> <a href="mailto:llvm-commits@lists.llvm.org">llvm-commits@lists.llvm.org</a><br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> >> <a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits</a><br>
>> >> >> >> >> >> >><br>
>> >> >> >> >> >> ><br>
>> >> >> >> >> ><br>
>> >> >> >> >> ><br>
>> >> >> >> ><br>
>> >> >> >> ><br>
>> >> >> ><br>
>> >> >> ><br>
>> >> ><br>
>> >> ><br>
>> ><br>
>> ><br>
><br>
><br>
</div></div></blockquote></div><br></div></div>