<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Tue, Dec 15, 2015 at 12:55 PM, Justin Bogner <span dir="ltr"><<a href="mailto:mail@justinbogner.com" target="_blank">mail@justinbogner.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div class=""><div class="h5">Akira Hatanaka <<a href="mailto:ahatanak@gmail.com">ahatanak@gmail.com</a>> writes:<br>
> On Mon, Dec 14, 2015 at 10:39 AM, Justin Bogner <<a href="mailto:mail@justinbogner.com">mail@justinbogner.com</a>><br>
> wrote:<br>
><br>
>> Akira Hatanaka via cfe-commits <<a href="mailto:cfe-commits@lists.llvm.org">cfe-commits@lists.llvm.org</a>> writes:<br>
>> > ahatanak created this revision.<br>
>> > ahatanak added a subscriber: cfe-commits.<br>
>> ><br>
>> > This patch fixes a crash that occurs when __kindof is incorrectly used<br>
>> > in the type parameter list of an interface. The crash occurs because<br>
>> > ObjCTypeParamList::back() is called in checkTypeParamListConsistency<br>
>> > on an empty list:<br>
>> ><br>
>> > 00762 diagLoc =<br>
>> S.getLocForEndOfToken(newTypeParams->back()->getLocEnd());<br>
>> ><br>
>> > <a href="http://reviews.llvm.org/D15463" rel="noreferrer" target="_blank">http://reviews.llvm.org/D15463</a><br>
>> ><br>
>> > Files:<br>
>> > lib/Parse/ParseObjc.cpp<br>
>> > test/SemaObjC/kindof.m<br>
>> ><br>
>> > Index: test/SemaObjC/kindof.m<br>
>> > ===================================================================<br>
>> > --- test/SemaObjC/kindof.m<br>
>> > +++ test/SemaObjC/kindof.m<br>
>> > @@ -302,3 +302,13 @@<br>
>> > void processCopyable(__typeof(getSomeCopyable()) string);<br>
>> > processCopyable(0); // expected-warning{{null passed to a callee that<br>
>> requires a non-null argument}}<br>
>> > }<br>
>> > +<br>
>> > +// __kinddof cannot be used in parameter list.<br>
>> > +@interface Array1<T> : NSObject<br>
>> > +@end<br>
>> > +<br>
>> > +@interface I1 : NSObject<br>
>> > +@end<br>
>> > +<br>
>> > +@interface Array1<__kindof I1*>(extensions) // //<br>
>> expected-error{{expected type parameter name}}<br>
>> > +@end<br>
>> > Index: lib/Parse/ParseObjc.cpp<br>
>> > ===================================================================<br>
>> > --- lib/Parse/ParseObjc.cpp<br>
>> > +++ lib/Parse/ParseObjc.cpp<br>
>> > @@ -603,7 +603,7 @@<br>
>> > // whether there are any protocol references.<br>
>> > lAngleLoc = SourceLocation();<br>
>> > rAngleLoc = SourceLocation();<br>
>> > - return list;<br>
>> > + return invalid ? nullptr : list;<br>
>><br>
>> This looks a bit suspicious to me. Since `invalid` is set *way* earlier<br>
>> in the function, it seems like we should be able to return earlier if<br>
>> this is correct, and not even call actOnObjCTypeParamList. OTOH, are<br>
>> there cases where `invalid == true` but list is non-empty? If so, are we<br>
>> doing the right thing when that happens?<br>
>><br>
>><br>
> I'm assuming you meant it should return earlier if this is *incorrect* (but<br>
> let me know if I misunderstood your comment).<br>
<br>
</div></div>I meant, "If `return invalid ? nullptr : list` is correct, then I<br>
suspect returning nullptr before calling actOnObjCTypeParamList makes<br>
more sense.<br>
<span class=""><br>
> The only downside to returning nullptr before actOnObjCTypeParamList is<br>
> called in line 595 is that it won't print the diagnostics that are printed<br>
> in actOnObjCTypeParamList. For example, if the following interface were<br>
> compiled,<br>
><br>
> @interface I1<T, T, __kindof I2*> NSObject<br>
> @end<br>
><br>
> actOnObjCTypeParamList could tell the user that parameter "T" was<br>
> redeclared in addition to printing the error message about __kindof. So if<br>
> we return before it's called, we won't see that error message.<br>
><br>
> When the interface declaration of category extensions1 in the following<br>
> piece of code is compiled, invalid is set to true and the list is not empty<br>
> (it contains the first two parameters).<br>
><br>
> @interface I1<T1, T2, T3> NSObject<br>
> @end<br>
><br>
> @interface I1<T, T, __kindof I2*> (extensions1)<br>
> @end<br>
><br>
</span>> *test3.m:6:23: **error: **expected type parameter name*<br>
<span class="">><br>
> @interface I6<T1, T2, __kindof I2*> (extensions1)<br>
><br>
</span>> * ^*<br>
><br>
> *test3.m:6:21: **error: **category has too few type parameters (expected 3,<br>
> have 2)*<br>
<span class="">><br>
> @interface I6<T1, T2, __kindof I2*> (extensions1)<br>
> It looks like it's doing the right thing, but the second diagnostic doesn't<br>
> seem necessary to me.<br>
<br>
</span>I'm not sure I follow exactly which errors are being reported with your<br>
patch, so maybe this is already working, but what we basically want is<br>
to only emit errors that are helpful.<br>
<br>
That is, if we get the `"T" is a redeclaration` error as well as the<br>
__kindof error, that's nice, but if we get the `too few type parameters`<br>
error that's going to be confusing and bad - the user didn't provide too<br>
few type parameters, but our previous error caused us to see too few.<br>
<br>
Basically, we don't ever want to report errors that are the result of<br>
previous errors, but as long as we don't end up doing that then<br>
gathering further errors is totally reasonable. Make sense?<br>
<br></blockquote><div><br></div><div>Yes, that makes perfect sense.</div><div><br></div><div>The following shows the error messages I see when I compile test case test3.m:</div><div>
<p class=""><span class=""></span></p>$ cat test3.m <br><br>#import <Foundation/Foundation.h><br>@interface I2 : NSObject<br>@end<br><br>@interface I7<T0, T1, T2> : NSObject<br>@end<br><br>@interface I7<T, T, __kindof I2*>(ext2)<br>@end<br><br>@interface I6<T> : NSObject<br>@end<br><br>@interface I6<__kindof I2*>(ext1)<br>@end<p class=""><br></p><p class="">### This patch: it prints the error messages we want to see.<br></p></div>test3.m:9:21: error: expected type parameter name<br>@interface I7<T, T, __kindof I2*>(ext2)<br> ^</div><div class="gmail_quote"><br>test3.m:9:18: error: redeclaration of type parameter 'T'<br>@interface I7<T, T, __kindof I2*>(ext2)<br> ~ ^<br><br>test3.m:15:15: error: expected type parameter name<br>@interface I6<__kindof I2*>(ext1)<br> ^</div><div class="gmail_quote"><br>3 errors generated.<div><br></div><div>### If it returns nullptr before actOnObjCTypeParamList is called: it doesn't print the "redeclaration of type parameter" error.</div><div><br></div>test3.m:9:21: error: expected type parameter name<br>@interface I7<T, T, __kindof I2*>(ext2)<br> ^<br><br>test3.m:15:15: error: expected type parameter name<br>@interface I6<__kindof I2*>(ext1)<br> ^<br><br>2 errors generated.</div><div class="gmail_quote"><br><div>### Trunk: It prints the "too few type parameters" message and it crashes.</div><div> </div>test3.m:9:21: error: expected type parameter name<br>@interface I7<T, T, __kindof I2*>(ext2)<br> ^<br><br>test3.m:9:18: error: redeclaration of type parameter 'T'<br>@interface I7<T, T, __kindof I2*>(ext2)<br> ~ ^<br><br>test3.m:9:19: error: category has too few type parameters (expected 3, have 2)<br>@interface I7<T, T, __kindof I2*>(ext2)<br> ^<br><br>test3.m:15:15: error: expected type parameter name<br>@interface I6<__kindof I2*>(ext1)<br> ^<br><br>clang: error: unable to execute command: Segmentation fault: 11</div><div class="gmail_quote"><br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex">
><br>
>> }<br>
>> ><br>
>> > /// Parse an objc-type-parameter-list.<br>
>> ><br>
>> ><br>
>><br>
</blockquote></div><br></div></div>