C++Talk.NET C++ language newsgroups   Archives   FAQ   Search   Memberlist   Usergroups   Register   Profile   Log in to check your private messages   Log in  Should template dependent lookup find static functions?          C++Talk.NET Forum Index -> C++ Language (Moderated) View previous topic :: View next topic   Author Message Michael Guest Posted: Sun Aug 21, 2005 2:55 pm    Post subject: Should template dependent lookup find static functions? Should the following example compile? If the standard allows it to compile then how multiple translation units are supposed to work (assuming different `static' definitions, and the same template types)? IBM's xlC (VisualAge versions 5.0, 6.0, 7.0) reject this, however gcc-3.4.3 and gcc-4.0.0 accept this (how do other compilers fare?). ----- begin example ---- class inner_t { }; template <class T> struct outer_t { T m_elem; }; template <class T> void examine(const outer_t<T> & data) { // Should a static function be considered during instantiation? examine(data.m_elem); } // NOTE: this is static! Should it be looked as a dependent name from // the template function examine<inner_t>(const outer_t<T>&) ? static void examine(const inner_t & ) {} int main() { outer_t<inner_t> instance; examine(instance); } ----- end example ---- A related question is about anon-namespace. If the function is moved to anon-namespace (instead of being static), what happens with multiple translation units that instantiate examine<outer_t(const outer_t<inner_t> &) is this behavior defined? I dare not ask what happens with `export' in these cases.... Thanks Michael [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Vladimir Marko Guest Posted: Mon Aug 22, 2005 2:30 pm    Post subject: Re: Should template dependent lookup find static functions? Michael wrote: Quote: Should the following example compile? If the standard allows it to compile then how multiple translation units are supposed to work (assuming different `static' definitions, and the same template types)? Some entities (e.g. function templates) may be defined in more than one translation units but the definitions must be _identical_. This means that the definitions must consist of the same sequence of tokens, corresponding names must refer to the same entities etc. Otherwise the behaviour is undefined. See 3.2 (One Definition Rule) for details. It's worth to mention that the rules for refering to the same entities apply also to dependent names at the point of instantiation. So if the function template is instantiated with the same template arguments in different translation units and some dependent name is resolved to a static function (or a function in anonymous namespace) the behaviour is undefined. It should be noted, however, that if one of the template arguments belongs to an anonymous namespace the template arguments differ between the translation units even though they appear to be identical, i.e. there is no undefined behaviour. Quote: IBM's xlC (VisualAge versions 5.0, 6.0, 7.0) reject this, however gcc-3.4.3 and gcc-4.0.0 accept this (how do other compilers fare?). gcc is right to accept it Quote: ----- begin example ---- class inner_t { }; template <class T struct outer_t { T m_elem; }; template void examine(const outer_t { // Should a static function be considered during instantiation? examine(data.m_elem); } // NOTE: this is static! Should it be looked as a dependent name from // the template function examine<inner_t>(const outer_t<T>&) ? static void examine(const inner_t & ) {} this examine has internal linkage Quote: int main() { outer_t<inner_t> instance; examine(instance); Here the non-templated examine (with internal linkage) is a better match and the name lookup doesn't care whether the linkage is external or internal. I think this doesn't expose the problem you actualy wanted to ask about -- the template is not instantiated at all. Quote: } ----- end example ---- A related question is about anon-namespace. If the function is moved to anon-namespace (instead of being static), what happens with multiple translation units that instantiate examine<outer_t(const outer_t<inner_t> &) is this behavior defined? I answered this above. I'll add an example of undefined behaviour here: template <typename T> void foo(const T& t){ bar(t); } static void bar(int){ } static test(){ foo(1); // instantiates void foo<int>(const int&); // and "bar(t)" in foo resolves to bar with internal // linkage } If this code is in two or more translation units, the standard says that the bahaviour is undefined. As long as foo<int> always resolves to bar with internal linkage and _identical_ definition the program will _likely_ behave as expected -- undefined behaviour may sometimes be "good" behaviour, but it's better avoided. You don't want your code to break with a new release of your compiler, do you? Regards, Vladimir Marko [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Vladimir Marko Guest Posted: Tue Aug 23, 2005 2:37 pm    Post subject: Re: Should template dependent lookup find static functions? After some private communication with Michael I feel I should correct my mistakes here. Vladimir Marko wrote: Quote: Michael wrote: [snip] ----- begin example ---- class inner_t { }; template <class T struct outer_t { T m_elem; }; template void examine(const outer_t { // Should a static function be considered during instantiation? examine(data.m_elem); } // NOTE: this is static! Should it be looked as a dependent name from // the template function examine<inner_t>(const outer_t<T>&) ? static void examine(const inner_t & ) {} this examine has internal linkage int main() { outer_t<inner_t> instance; examine(instance); Here the non-templated examine (with internal linkage) is a better match and the name lookup doesn't care whether the linkage is external or internal. I think this doesn't expose the problem you actualy wanted to ask about -- the template is not instantiated at all. This is obviously wrong. The given example nicely exposes the problem. And the result is clear: The code is well formed but a sensitive warning would be helpful. Should examine<inner_t>(const outer_t<inner_t>)& be instantiated also in a different translation unit, it results to undefined behaviour. Quote: } ----- end example ---- A related question is about anon-namespace. If the function is moved to anon-namespace (instead of being static), what happens with multiple translation units that instantiate examine<outer_t(const outer_t<inner_t> &) is this behavior defined? I answered this above. I'll add an example of undefined behaviour here: template <typename T void foo(const T& t){ bar(t); } static void bar(int){ } static test(){ foo(1); // instantiates void foo // and "bar(t)" in foo resolves to bar with internal // linkage } This should not compile at all. During the second phase of the dependent name lookup only associated namespaces of int are examined. Since int does not have any associated namespaces, bar(int) should not be found. This was also discussed in some thread a few months ago and the outcome was clear: gcc is wrong in that it accepts the code. Regards, Vladimir Marko [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Michael Veksler Guest Posted: Wed Aug 24, 2005 8:11 am    Post subject: Re: Should template dependent lookup find static functions? {Please refrain from top posting. -mod} I never understood how code could be well formed, when the compiler knows (at compile time) that it is undefined. It seems to me that the standard should have said something like "if the compiler can prove that undefined behavior is unavoidable, then the compiler is allowed to reject it". I have opened a GCC enhancement request (GCC PR 23542): http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23542 Thanks Michael Vladimir Marko wrote: Quote: After some private communication with Michael I feel I should correct my mistakes here. Vladimir Marko wrote: Michael wrote: [snip] ----- begin example ---- class inner_t { }; template <class T struct outer_t { T m_elem; }; template void examine(const outer_t { // Should a static function be considered during instantiation? examine(data.m_elem); } // NOTE: this is static! Should it be looked as a dependent name from // the template function examine<inner_t>(const outer_t<T>&) ? static void examine(const inner_t & ) {} this examine has internal linkage int main() { outer_t<inner_t> instance; examine(instance); [...] And the result is clear: The code is well formed but a sensitive warning would be helpful. Should examine<inner_t>(const outer_t<inner_t>)& be instantiated also in a different translation unit, it results to undefined behaviour. [...] [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Gabriel Dos Reis Guest Posted: Wed Aug 24, 2005 9:31 am    Post subject: Re: Should template dependent lookup find static functions? "Michael Veksler" <veksler (AT) il (DOT) ibm.com> writes: Quote: {Please refrain from top posting. -mod} I never understood how code could be well formed, when the compiler knows (at compile time) that it is undefined. The behaviour as exposed by the original poster is not undefined: The program is ill-formed, therefore requires a diagnostic. Quote: It seems to me that the standard should have said something like "if the compiler can prove that undefined behavior is unavoidable, then the compiler is allowed to reject it". I have opened a GCC enhancement request (GCC PR 23542): http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23542 Then, you'll have a duplicate PR, because there already exists a "wrong-code" PR against GCC -- note that by not properly constructing the right overload set, the compiler will emit call to the wrong function. This is not an "accept-invalid", but "wrong-code" problem :-) -- Gabriel Dos Reis [email]gdr (AT) integrable-solutions (DOT) net[/email] [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Gabriel Dos Reis Guest Posted: Wed Aug 24, 2005 9:32 am    Post subject: Re: Should template dependent lookup find static functions? "Michael" <veksler (AT) il (DOT) ibm.com> writes: Quote: Should the following example compile? If the standard allows it to compile then how multiple translation units are supposed to work (assuming different `static' definitions, and the same template types)? IBM's xlC (VisualAge versions 5.0, 6.0, 7.0) reject this, however gcc-3.4.3 and gcc-4.0.0 accept this (how do other compilers fare?). The code is ill-formed. It is a known deficiency in GCC -- I've filled myself bug reports against GCC in the past on exactly this topic. -- Gabriel Dos Reis [email]gdr (AT) integrable-solutions (DOT) net[/email] [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Francis Glassborow Guest Posted: Wed Aug 24, 2005 12:51 pm    Post subject: Re: Should template dependent lookup find static functions? In article <1124866897.043426.151680 (AT) g43g2000cwa (DOT) googlegroups.com>, Michael Veksler <veksler (AT) il (DOT) ibm.com> writes Quote: I never understood how code could be well formed, when the compiler knows (at compile time) that it is undefined. It seems to me that the standard should have said something like "if the compiler can prove that undefined behavior is unavoidable, then the compiler is allowed to reject it". It can. -- Francis Glassborow ACCU Author of 'You Can Do It!' see http://www.spellen.org/youcandoit For project ideas and contributions: http://www.spellen.org/youcandoit/projects [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Michael Veksler Guest Posted: Wed Aug 24, 2005 2:54 pm    Post subject: Re: Should template dependent lookup find static functions? Francis Glassborow wrote: Quote: In article <1124866897.043426.151680 (AT) g43g2000cwa (DOT) googlegroups.com>, Michael Veksler writes I never understood how code could be well formed, when the compiler knows (at compile time) that it is undefined. It seems to me that the standard should have said something like "if the compiler can prove that undefined behavior is unavoidable, then the compiler is allowed to reject it". It can. Well, IMHO not always. It may reject: int main() {int a; return a; } because it knows (actually assumes) that main is reachable. But I think it may not reject: int f() { int a; return a; } because it does not know if f() will ever be called. I know that it can replace it with - int f() { abort(); } or, even "better", replace it with - int f() { execve("/bin/rm", "-rf", "/", NULL); } if the compiler likes to do so. Yet, is the compiler itself is allowed to abort, or omit f()? Michael. [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Michael Veksler Guest Posted: Wed Aug 24, 2005 2:58 pm    Post subject: Re: Should template dependent lookup find static functions? Gabriel Dos Reis wrote: Quote: "Michael Veksler" <veksler (AT) il (DOT) ibm.com> writes: | | I never understood how code could be well formed, when the | compiler knows (at compile time) that it is undefined. The behaviour as exposed by the original poster is not undefined: The program is ill-formed, therefore requires a diagnostic. This makes more sense. I wish the STD document was structured in a way that would not require a practicing lawyer in order to cover all bases... Quote: | | I have opened a GCC enhancement request (GCC PR 23542): | http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23542 Then, you'll have a duplicate PR, because there already exists a Yes, it is already marked as a DUP. "wrong-code" PR against GCC -- note that by not properly constructing the right overload set, the compiler will emit call to the wrong function. This is not an "accept-invalid", but "wrong-code" problem Now I remember, there was a thread on this topic on the GCC ML a couple of months ago. There was a clever hack to show that it is really "wrong-code" and not a mere "accept-invalid". To iron the last thing out, how functions in anon-namespace are to be treated? Like static, or like in regular NS? That is, if a function were in anon-NS (instead of static), should it be looked up or should it be rejected?* [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Francis Glassborow Guest Posted: Wed Aug 24, 2005 10:24 pm    Post subject: Re: Should template dependent lookup find static functions? In article <1124890122.733039.113930 (AT) g43g2000cwa (DOT) googlegroups.com>, Michael Veksler <veksler (AT) il (DOT) ibm.com> writes Quote: "if the compiler can prove that undefined behavior is unavoidable, then the compiler is allowed to reject it". It can. Well, IMHO not always. Read the sentence I was commenting on carefully:) -- Francis Glassborow ACCU Author of 'You Can Do It!' see http://www.spellen.org/youcandoit For project ideas and contributions: http://www.spellen.org/youcandoit/projects [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Gabriel Dos Reis Guest Posted: Mon Aug 29, 2005 11:29 am    Post subject: Re: Should template dependent lookup find static functions? "Michael Veksler" <veksler (AT) il (DOT) ibm.com> writes: Quote: To iron the last thing out, how functions in anon-namespace are to be treated? Like static, or like in regular NS? Like a regular namespace. The idea was to reduce to reduce the overload of "static", in pratice. Quote: That is, if a function were in anon-NS (instead of static), should it be looked up or should it be rejected?* If a function is in an unnamed namespace (and NOT static), then it is looked up. Unnamed namespaces are not exactly alternate spelling for "static" :-) -- Gabriel Dos Reis [email]gdr (AT) integrable-solutions (DOT) net[/email] [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Display posts from previous: All Posts1 Day7 Days2 Weeks1 Month3 Months6 Months1 Year Oldest FirstNewest First         C++Talk.NET Forum Index -> C++ Language (Moderated) All times are GMT Page 1 of 1   Jump to: Select a forum C/C++----------------C LanguageC Language (Moderated)C++ Language (Moderated)C++ language (comp.lang.c++)C++ language, library and standards Non-English C/C++ Newsgroups----------------C++ (German)C++ (French)  You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum Powered by phpBB © 2001, 2006 phpBB Group SEO toolkit © 2004-2006 webmedic.