C++Talk.NET C++ language newsgroups   Archives   FAQ   Search   Memberlist   Usergroups   Register   Profile   Log in to check your private messages   Log in  when inline function is not inlined Goto page 1, 2, 3  Next          C++Talk.NET Forum Index -> C++ Language (Moderated) View previous topic :: View next topic   Author Message Tony Guest Posted: Fri Jan 02, 2004 6:55 pm    Post subject: when inline function is not inlined Dear C++ Gurus, As put in many books, inline is only a suggestion to the compiler. The inline function may not be inlined. When an inline function is not inlined, it will be offlined just as a offline function. But I wonder maybe there is still a difference. //f1.h inline void f1() { } void f2(); void f3(); //f2.cpp #include "f1.h" void f2() { f1(); } //f3.cpp #include "f1.h" void f3() { f1(); } //main.cpp #include "f1.h" int main() { f2(); f3(); } If f1() is complex, the inline may be ignored. But no matter how complex f1() is, the above code can always run. But if I remove "inline" from f1(), it will have duplicate defintions in f2.cpp and f3.cpp. The linker will fail. So if "inline" is used, even if it is not inlined, it is still different from not having "inline". What's the exact difference ? Where does the compile put f1() defintion when "inline" is used ? Best regards Tony [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top galathaea Guest Posted: Sat Jan 03, 2004 4:08 am    Post subject: Re: when inline function is not inlined "Tony" wrote: : As put in many books, inline is only a suggestion to the compiler. : The inline function may not be inlined. When an inline function is not : inlined, : it will be offlined just as a offline function. But I wonder maybe : there is still a difference. : : : //f1.h : inline void f1() { } : void f2(); : void f3(); : : //f2.cpp : #include "f1.h" : void f2() { f1(); } : : //f3.cpp : #include "f1.h" : void f3() { f1(); } : : //main.cpp : #include "f1.h" : int main() : { : f2(); : f3(); : } : : If f1() is complex, the inline may be ignored. : But no matter how complex f1() is, the above code can always run. : : But if I remove "inline" from f1(), it will have duplicate defintions : in f2.cpp and f3.cpp. The linker will fail. : : So if "inline" is used, even if it is not inlined, it is still : different from : not having "inline". What's the exact difference ? Where does the : compile put f1() defintion when "inline" is used ? Its not exactly _where_ it puts the function, but what additional information is attached to the function that the linker sees. In particular, it is the linkage specification attached to the function in each translation unit. You are correct, inline does more than just suggest the in place expansion / substitution of the code. The use of the keyword inline is useful anytime one might have violations of the one definition rule due to code in the header, and this is particularly true for parametrised functions and algorithmics. -- -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- galathaea: prankster, fablist, magician, liar [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Michael Tiomkin Guest Posted: Sat Jan 03, 2004 4:08 am    Post subject: Re: when inline function is not inlined It seems that you use a Windows linker that doesn't allow multiple definitions in object files. Many inline functions usually appear in .h files and therefore can appear in many object files. As a result, a WIndows compiler has to use a "weak definition" for inlined functions in order to allow overriding of multiple appearances of the same item. Unix linkers usually allow multiple definitions, and then there is no need in special definition for inlined functions that aren't inlined. In this case, when you want to enforce the "single definition" rule, you need to use a special compiler option to treat inline functions as weak defs and a special linker option against multiple appearances in the objects. [email]tongru_huo (AT) yahoo (DOT) com[/email] (Tony) wrote in message news:<1f87458d.0401020847.71074931 (AT) posting (DOT) google.com>... Quote: As put in many books, inline is only a suggestion to the compiler. The inline function may not be inlined. When an inline function is not inlined, it will be offlined just as a offline function. But I wonder maybe there is still a difference. //f1.h inline void f1() { } void f2(); void f3(); //f2.cpp #include "f1.h" void f2() { f1(); } //f3.cpp #include "f1.h" void f3() { f1(); } //main.cpp #include "f1.h" int main() { f2(); f3(); } If f1() is complex, the inline may be ignored. But no matter how complex f1() is, the above code can always run. But if I remove "inline" from f1(), it will have duplicate defintions in f2.cpp and f3.cpp. The linker will fail. So if "inline" is used, even if it is not inlined, it is still different from not having "inline". What's the exact difference ? Where does the compile put f1() defintion when "inline" is used ? Best regards Tony [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Stefan Rupp Guest Posted: Sat Jan 03, 2004 4:09 am    Post subject: Re: when inline function is not inlined Good evening, Tony schrieb: Quote: //f1.h inline void f1() { } void f2(); void f3(); Put include guards around all declarations in your headers: #if !defined(F1_H_) #define F1_H_ // your C++ header goes here #endif Quote: //f2.cpp #include "f1.h" void f2() { f1(); } //f3.cpp #include "f1.h" void f3() { f1(); } //main.cpp #include "f1.h" int main() { f2(); f3(); } [...] But if I remove "inline" from f1(), it will have duplicate defintions in f2.cpp and f3.cpp. The linker will fail. Include guards do not solve *this* problem. If you simply omit the "include" keyword, the header contains a defnition of function f1 (i.e. its implementation), rather than just its declaration (i.e. its prototype). You will need to move the definition of function f1 to a separate translation unit. The "include" keyword tells the compiler more than just "this may be used inlined". If the compiler decides not to inline the code, it must not create code for each and every occurance of the function, otherwise you would see the effect you reported. However, what exactly the various compiler do in this case, is beyond my knowledge. Some create weak symbols. Quote: Where does the compile put f1() defintion when "inline" is used ? The question is related to the question for the address for an inlined function (here &f1). Stefan [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Rajaram S. Gaunker Guest Posted: Sat Jan 03, 2004 9:50 am    Post subject: Re: when inline function is not inlined Quote: As put in many books, inline is only a suggestion to the compiler. The inline function may not be inlined. U r rite. Quote: //f1.h inline void f1() { } void f2(); void f3(); //f2.cpp #include "f1.h" void f2() { f1(); } //f3.cpp #include "f1.h" void f3() { f1(); } //main.cpp #include "f1.h" int main() { f2(); f3(); } If f1() is complex, the inline may be ignored. But no matter how complex f1() is, the above code can always run. yes Quote: But if I remove "inline" from f1(), it will have duplicate defintions in f2.cpp and f3.cpp. The linker will fail. linker wont fail as inline has two meanings 1) Request to inline the function. 2) function has internal linkage. if function is not inlined then it has internal linkage means it is linked only within the translation unit, hence there will be no clash at link time in abave case with inlining failed. correct if wrong Regards Rajaram [ 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: Sat Jan 03, 2004 3:02 pm    Post subject: Re: when inline function is not inlined In message <fVnJb.4779$HT.948 (AT) newssvr27 (DOT) news.prodigy.com>, galathaea <galathaea (AT) excite (DOT) com> writes Quote: Its not exactly _where_ it puts the function, but what additional information is attached to the function that the linker sees. In particular, it is the linkage specification attached to the function in each translation unit. You are correct, inline does more than just suggest the in place expansion / substitution of the code. The use of the keyword inline is useful anytime one might have violations of the one definition rule due to code in the header, and this is particularly true for parametrised functions and algorithmics. But you better be careful that you do not break the one definition rule. I.e. you need to ensure that all the definitions the linker sees are the same else you may get surprises when it discards some of them. -- Francis Glassborow ACCU Author of 'You Can Do It!' see http://www.spellen.org/youcandoit or http://www.robinton.demon.co.uk Happy Xmas, Hanukkah, Yuletide, Winter/Summer Solstice to all. [ 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: Sat Jan 03, 2004 3:03 pm    Post subject: Re: when inline function is not inlined In message <ef72c6cb.0401021459.4438a837 (AT) posting (DOT) google.com>, Michael Tiomkin <tmk (AT) netvision (DOT) net.il> writes Quote: It seems that you use a Windows linker that doesn't allow multiple definitions in object files. Many inline functions usually appear in .h files and therefore can appear in many object files. As a result, a WIndows compiler has to use a "weak definition" for inlined functions in order to allow overriding of multiple appearances of the same item. Unix linkers usually allow multiple definitions, and then there is no need in special definition for inlined functions that aren't inlined. In this case, when you want to enforce the "single definition" rule, you need to use a special compiler option to treat inline functions as weak defs and a special linker option against multiple appearances in the objects. Are you saying that Unix systems support code bloat? -- Francis Glassborow ACCU Author of 'You Can Do It!' see http://www.spellen.org/youcandoit or http://www.robinton.demon.co.uk Happy Xmas, Hanukkah, Yuletide, Winter/Summer Solstice to all. [ 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: Sat Jan 03, 2004 3:05 pm    Post subject: Re: when inline function is not inlined In message <e3c5fdc7.0401022130.73e41aae (AT) posting (DOT) google.com>, Rajaram S. Gaunker <rajaram (AT) emuzed (DOT) com> writes Quote: linker wont fail as inline has two meanings 1) Request to inline the function. 2) function has internal linkage. Not really true any longer (see 7.1.2). It is far from clear to me exactly what inline now does with regard to linkage of functions. What the Standard does say is that an inline function must have exactly the same definition in all TUs where it is used and that a function with external linkage that is declared inline in one TU must be declared as inline in all TUs where it appears. What really muddies the water is that through most of this sub-clause it refers to functions with external linkage (or else makes no specification. However right at the end it places two requirements on functions declared as 'extern inline'. It seems to me that those requirements (re static variables and string literals) should apply to all inline functions with external linkage however that linkage is acquired. -- Francis Glassborow ACCU Author of 'You Can Do It!' see http://www.spellen.org/youcandoit or http://www.robinton.demon.co.uk Happy Xmas, Hanukkah, Yuletide, Winter/Summer Solstice to all. [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Michael Tiomkin Guest Posted: Sun Jan 04, 2004 2:17 am    Post subject: Re: when inline function is not inlined Francis Glassborow <francis (AT) robinton (DOT) demon.co.uk> wrote Quote: In message <ef72c6cb.0401021459.4438a837 (AT) posting (DOT) google.com>, Michael Tiomkin <tmk (AT) netvision (DOT) net.il> writes It seems that you use a Windows linker that doesn't allow multiple definitions in object files. Many inline functions usually appear in .h files and therefore can appear in many object files. As a result, a WIndows compiler has to use a "weak definition" for inlined functions in order to allow overriding of multiple appearances of the same item. Unix linkers usually allow multiple definitions, and then there is no need in special definition for inlined functions that aren't inlined. In this case, when you want to enforce the "single definition" rule, you need to use a special compiler option to treat inline functions as weak defs and a special linker option against multiple appearances in the objects. Are you saying that Unix systems support code bloat? Not at all. The difference is in object formats and common practice. In Unix tradition, the executables can easily be relinked, e.g. in "mv new-exe old-exe.o; cc -o new-exe 1st.o 2nd.o old-exe.o", where the new items from 1st.o and 2nd.o are replacing their previous versions from the same compilation units. In this case you do have multiple definitions, but the linker chooses one of them. On Windows, the executable format (PE) is different from the object format (COFF), and an executable is not supposed to be freely updated, allowing the linker to enforce the "single definition" rule. As a result, if you use two different functions with the same name and signature, you'll get linker error in Windows, but on Unix you might have the linker silently replace the functions with one of the versions. It's not code bloat, but rather more error-prone environment. This is the reason why you might want to enforce the "signle definition" rule on Unix with the right compiler/linker options. Many people compare Unix with Swiss Army knife - we should be careful using it. [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Max Zinal Guest Posted: Sun Jan 04, 2004 2:23 am    Post subject: Re: when inline function is not inlined Tony wrote: Quote: So if "inline" is used, even if it is not inlined, it is still different from not having "inline". What's the exact difference ? Where does the compile put f1() defintion when "inline" is used ? If it is inlined, it is inlined - no need to put its definition anywhere but to the exact point of call. If the compiler refuses to make the function inline, it makes it static. So the linker does not complain on duplicate definitions, but you still have code to that function in *any* translation unit from which the function is called. I am not a guru in the Holy Standard, so I am not sure that language specification requires such behaviour. Instead, I'm pretty shure that is the way the things really happen (and *should* happen) in practice. Thanks, Max Zinal. [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top James Dennett Guest Posted: Sun Jan 04, 2004 11:18 am    Post subject: Re: when inline function is not inlined Max Zinal wrote: Quote: Tony wrote: So if "inline" is used, even if it is not inlined, it is still different from not having "inline". What's the exact difference ? Where does the compile put f1() defintion when "inline" is used ? If it is inlined, it is inlined - no need to put its definition anywhere but to the exact point of call. If the compiler refuses to make the function inline, it makes it static. So the linker does not complain on duplicate definitions, but you still have code to that function in *any* translation unit from which the function is called. I am not a guru in the Holy Standard, so I am not sure that language specification requires such behaviour. Instead, I'm pretty shure that is the way the things really happen (and *should* happen) in practice. Not quite -- if you take the address of an inline function, you must get the same result in each translation unit, for example, and that wouldn't happen if the compiler just treated the inline function as a static function. Similar issues arise with uniqueness of static variables within inline functions, and this is one are in which the rules of C++ and those of C have diverged. -- James. [ 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: Sun Jan 04, 2004 4:12 pm    Post subject: Re: when inline function is not inlined In message <1073148357.537454 (AT) smtp (DOT) tvcom.ru>, Max Zinal <Zlat0 (AT) mail (DOT) ru> writes Quote: Tony wrote: So if "inline" is used, even if it is not inlined, it is still different from not having "inline". What's the exact difference ? Where does the compile put f1() defintion when "inline" is used ? If it is inlined, it is inlined - no need to put its definition anywhere but to the exact point of call. If the compiler refuses to make the function inline, it makes it static. So the linker does not complain on duplicate definitions, but you still have code to that function in *any* translation unit from which the function is called. I am not a guru in the Holy Standard, so I am not sure that language specification requires such behaviour. Instead, I'm pretty shure that is the way the things really happen (and *should* happen) in practice. That used to be true in pre-Standard days but I am far from sure that it is the case today (see 7.1.2). Apart from issues of code bloat through multiple instantiations of the same definition (though the requirement that the definitions be the same in all TUs allows the linker to remove duplicates -- something it must not do for real 'static' functions) there is also the issue of string literals and static block scope variables. -- Francis Glassborow ACCU Author of 'You Can Do It!' see http://www.spellen.org/youcandoit or http://www.robinton.demon.co.uk Happy Xmas, Hanukkah, Yuletide, Winter/Summer Solstice to all. [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top James Kanze Guest Posted: Sun Jan 04, 2004 4:26 pm    Post subject: Re: when inline function is not inlined [email]tongru_huo (AT) yahoo (DOT) com[/email] (Tony) writes: Quote: As put in many books, inline is only a suggestion to the compiler. The inline function may not be inlined. When an inline function is not inlined, it will be offlined just as a offline function. But I wonder maybe there is still a difference. //f1.h inline void f1() { } void f2(); void f3(); //f2.cpp #include "f1.h" void f2() { f1(); } //f3.cpp #include "f1.h" void f3() { f1(); } //main.cpp #include "f1.h" int main() { f2(); f3(); } If f1() is complex, the inline may be ignored. But no matter how complex f1() is, the above code can always run. But if I remove "inline" from f1(), it will have duplicate defintions in f2.cpp and f3.cpp. The linker will fail. So if "inline" is used, even if it is not inlined, it is still different from not having "inline". What's the exact difference ? Where does the compile put f1() defintion when "inline" is used ? You've just stated the only formal difference -- inline allows (and in fact requires) multiple definitions, where as without inline, they are illegal. Practically, the intent of inline is clear, and any compiler who doesn't respect it except for exceptional cases has a serious quality problem. Because what constitutes an exceptional case depends at least a little on how the compiler is implemented, the standard doesn't really have a way of specifying this. But that shouldn't be taken as a license for the compiler implementer to ignore inline (unless the compiler really can do better than the programmer -- not often the case today). I rarely use inline, but when I do, it is because I need it -- the compiler had better not ignore it. -- James Kanze mailto:kanze (AT) gabi-soft (DOT) fr Conseils en informatique orientée objet/ Beratung in objektorientierter Datenverarbeitung 11 rue de Rambouillet, 78460 Chevreuse, France +33 1 41 89 80 93 [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top James Kanze Guest Posted: Sun Jan 04, 2004 9:40 pm    Post subject: Re: when inline function is not inlined [email]tmk (AT) netvision (DOT) net.il[/email] (Michael Tiomkin) writes: Quote: It seems that you use a Windows linker that doesn't allow multiple definitions in object files. Many inline functions usually appear in .h files and therefore can appear in many object files. As a result, a WIndows compiler has to use a "weak definition" for inlined functions in order to allow overriding of multiple appearances of the same item. Unix linkers usually allow multiple definitions, and then there is no need in special definition for inlined functions that aren't inlined. I've never seen a Unix linker which allowed multiple definitions. The GNU linker for Linux (version 2.13.90.0.1 certainly doesn't, nor does the Sun linker on my Sparc (/usr/ccs/bin/ld, version 3.0). Quote: In this case, when you want to enforce the "single definition" rule, you need to use a special compiler option to treat inline functions as weak defs and a special linker option against multiple appearances in the objects. That's exactly how it is done under Unix, as well. -- James Kanze mailto:kanze (AT) gabi-soft (DOT) fr Conseils en informatique orientée objet/ Beratung in objektorientierter Datenverarbeitung 11 rue de Rambouillet, 78460 Chevreuse, France +33 1 41 89 80 93 [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top James Kanze Guest Posted: Sun Jan 04, 2004 9:40 pm    Post subject: Re: when inline function is not inlined [email]tmk (AT) netvision (DOT) net.il[/email] (Michael Tiomkin) writes: Quote: Francis Glassborow <francis (AT) robinton (DOT) demon.co.uk> wrote in message news:<TUG1j3DI2o9$EwvP (AT) robinton (DOT) demon.co.uk>... In message <ef72c6cb.0401021459.4438a837 (AT) posting (DOT) google.com>, Michael Tiomkin <tmk (AT) netvision (DOT) net.il> writes It seems that you use a Windows linker that doesn't allow multiple definitions in object files. Many inline functions usually appear in .h files and therefore can appear in many object files. As a result, a WIndows compiler has to use a "weak definition" for inlined functions in order to allow overriding of multiple appearances of the same item. Unix linkers usually allow multiple definitions, and then there is no need in special definition for inlined functions that aren't inlined. In this case, when you want to enforce the "single definition" rule, you need to use a special compiler option to treat inline functions as weak defs and a special linker option against multiple appearances in the objects. Are you saying that Unix systems support code bloat? All systems support code bloat:-). Quote: Not at all. The difference is in object formats and common practice. In Unix tradition, the executables can easily be relinked, e.g. in "mv new-exe old-exe.o; cc -o new-exe 1st.o 2nd.o old-exe.o", where the new items from 1st.o and 2nd.o are replacing their previous versions from the same compilation units. In this case you do have multiple definitions, but the linker chooses one of them. On Windows, the executable format (PE) is different from the object format (COFF), and an executable is not supposed to be freely updated, allowing the linker to enforce the "single definition" rule. I don't know what the current situation is, but the earliest Intel compilers I used (circa 1979) certainly supported relinking an already linked module. I don't remember being able to do this under early Unix. (I don't remember trying, either, so it may have been possible. But the a.out files and .o files definitely had a different magic number, and some differences in the format. In 1985, at least.) But my impression is that what you are describing is incremental linking. Again, I don't know what the current situation under Windows is, but it is a relatively recent feature (say in the last five or ten years) under Unix -- it certainly didn't exist when I started using Unix. Incremental linking is a separate feature, and does not require that the executable and the object files have the same format; for it to work correctly, in fact, the linker has to know which file is the old executable file, and treat it differently. Quote: As a result, if you use two different functions with the same name and signature, you'll get linker error in Windows, but on Unix you might have the linker silently replace the functions with one of the versions. Could you cite which Unix version does this. As I said, I've never seen it, and the two current Unix I have access to (Linux and Solaris) both give me a definite error when I have a function defined in two different translation units. Or perhaps you are thinking of what happens with dynamic linking. I think that there are differences there -- at least with dlopen under Unix, if a dynamic object contains a definition for a symbol already defined, that definition can be ignored. I don't know very much about DLL's under Windows, but I seem to recall hearing that you can't export the same symbol in two different DLL's. (There is a fundamental difference in philosophy between the Unix and Windows work here. Under Unix, you decide what symbols interest you in the dynamic object when you load it. Under Windows, you seem to have to specify this information when you create the dynamic object.) -- James Kanze mailto:kanze (AT) gabi-soft (DOT) fr Conseils en informatique orientée objet/ Beratung in objektorientierter Datenverarbeitung 11 rue de Rambouillet, 78460 Chevreuse, France +33 1 41 89 80 93 [ 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 Goto page 1, 2, 3  Next Page 1 of 3   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.