C++Talk.NET C++ language newsgroups   Archives   FAQ   Search   Memberlist   Usergroups   Register   Profile   Log in to check your private messages   Log in  Reading objects from an array of char Goto page 1, 2  Next          C++Talk.NET Forum Index -> C++ Language (Moderated) View previous topic :: View next topic   Author Message glenlow@pixelglow.com Guest Posted: Tue Dec 14, 2004 8:33 pm    Post subject: Reading objects from an array of char Hi All, Consider the following: T t; char buffer [sizeof (T)]; *reinterpret_cast <T*> (buffer) = t; T t2 = *reinterpret_cast <T*> (buffer); I don't think the last statement is kosher with respect to the aliasing rules since the declared object type is char [N] and not T; on the other hand, I never write in chars, only T's, so is the stored object type really a T? Cheers, Glen Low, Pixelglow Software www.pixelglow.com [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Ali Çehreli Guest Posted: Wed Dec 15, 2004 1:38 am    Post subject: Re: Reading objects from an array of char <glenlow (AT) pixelglow (DOT) com> wrote Quote: T t; char buffer [sizeof (T)]; *reinterpret_cast <T*> (buffer) = t; Undefined behavior above. You are calling operator= on an unitialized piece of memory. buffer does not contain a T because no constructor has constructed a T on that memory. I think you should use placement new to construct the object on buffer. Quote: T t2 = *reinterpret_cast <T*> (buffer); That's fine only if buffer contains a valid T; and that's not the case with the code posted. Quote: I don't think the last statement is kosher with respect to the aliasing rules since the declared object type is char [N] and not T; on the other hand, I never write in chars, only T's, so is the stored object type really a T? Do you mean alignment issues; as in, probably a T could not have been placed on buffer? Though my gut feeling is that it should be ok (judging from how container classes like std::vector work with any type), I need a definitive answer as well. Ali [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Jack Klein Guest Posted: Wed Dec 15, 2004 4:39 am    Post subject: Re: Reading objects from an array of char On 14 Dec 2004 15:33:02 -0500, [email]glenlow (AT) pixelglow (DOT) com[/email] wrote in comp.lang.c++.moderated: Quote: Hi All, Consider the following: T t; char buffer [sizeof (T)]; *reinterpret_cast <T*> (buffer) = t; T t2 = *reinterpret_cast <T*> (buffer); I don't think the last statement is kosher with respect to the aliasing rules since the declared object type is char [N] and not T; on the other hand, I never write in chars, only T's, so is the stored object type really a T? Cheers, Glen Low, Pixelglow Software www.pixelglow.com Depends on two things. First what T is, if it has a non-trivial constructor or a user-defined assignment operator, there's a good chance it will fail. Second is alignment. Say T is, for example, a double, on an implementation where sizeof(double) is 8 and doubles must be located at addresses evenly divisible by 8. If 'buffer' does not have the correct alignment for T, undefined behavior occurs. Placement new could take care of the first, if the object requires construction to be valid, but not the second. In C++, as in C, memory is memory is memory. What counts is not really the object type, but the type of lvalue used. The block of memory (which is what an object is) has no type, values stored in memory have types imposed by the type of the lvalue used to store them there. If you have a pointer to a large enough array of characters that is suitably aligned, you may copy a POD object to it or construct any sort of object in it. At that point is may no longer be legitimate to access it as an array of signed characters, but it is valid to access via an lvalue of the type of the actual value it contains. There is no difference between an ordinary array of characters and a block of bytes allocated with malloc() of the same size, other than a possible alignment issue. -- Jack Klein Home: http://JK-Technology.Com FAQs for comp.lang.c http://www.eskimo.com/~scs/C-faq/top.html comp.lang.c++ http://www.parashift.com/c++-faq-lite/ alt.comp.lang.learn.c-c++ http://www.contrib.andrew.cmu.edu/~ajo/docs/FAQ-acllc.html [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top glenlow@pixelglow.com Guest Posted: Wed Dec 15, 2004 4:42 am    Post subject: Re: Reading objects from an array of char Quote: T t; char buffer [sizeof (T)]; *reinterpret_cast <T*> (buffer) = t; Undefined behavior above. You are calling operator= on an unitialized piece of memory. buffer does not contain a T because no constructor has constructed a T on that memory. I think you should use placement new to construct the object on buffer. Apologies, you're right, it should be: new (buffer) (t); Quote: I don't think the last statement is kosher with respect to the aliasing rules since the declared object type is char [N] and not T; on the other hand, I never write in chars, only T's, so is the stored object type really a T? Do you mean alignment issues; as in, probably a T could not have been placed on buffer? Now that you mention it, alignment would probably defeat this, irrespective of whether aliasing would work. I know malloc and new are supposed to yield memory suitable for any type to put in, alignment-wise. Does a simple (auto) char buffer have this guarantee? The aliasing issue is also problematic. The C99 standard (and probably the C++ standard if you interpret it compatibly) seems to think that (heap?) allocated memory "becomes" the first type you put in e.g. by memcpy or assignment. On the other hand, auto memory with a declared type is that declared type, so reading another type from it might violate the aliasing rules. Cheers, Glen Low, Pixelglow Software www.pixelglow.com [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Ben Hutchings Guest Posted: Wed Dec 15, 2004 4:47 am    Post subject: Re: Reading objects from an array of char [email]glenlow (AT) pixelglow (DOT) com[/email] wrote: Quote: Hi All, Consider the following: T t; char buffer [sizeof (T)]; *reinterpret_cast <T*> (buffer) = t; T t2 = *reinterpret_cast <T*> (buffer); I don't think the last statement is kosher with respect to the aliasing rules since the declared object type is char [N] and not T; on the other hand, I never write in chars, only T's, so is the stored object type really a T? According to standard section 3.8 paragraphs 1-2, buffer contains an object of every POD type for which it is suitably aligned and sized (but obviously not a valid object!). Assuming you restrict T to be a POD type and ensure correct alignment of buffer, then the stored object type is indeed a T. -- Ben Hutchings compatible: Gracefully accepts erroneous data from any source [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Thomas Mang Guest Posted: Wed Dec 15, 2004 4:47 pm    Post subject: Re: Reading objects from an array of char <glenlow (AT) pixelglow (DOT) com> schrieb im Newsbeitrag news:1103075881.288182.281390 (AT) z14g2000cwz (DOT) googlegroups.com... Quote: T t; char buffer [sizeof (T)]; *reinterpret_cast <T*> (buffer) = t; Undefined behavior above. You are calling operator= on an unitialized piece of memory. buffer does not contain a T because no constructor has constructed a T on that memory. I think you should use placement new to construct the object on buffer. Apologies, you're right, it should be: new (buffer) (t); I don't think the last statement is kosher with respect to the aliasing rules since the declared object type is char [N] and not T; on the other hand, I never write in chars, only T's, so is the stored object type really a T? Do you mean alignment issues; as in, probably a T could not have been placed on buffer? Now that you mention it, alignment would probably defeat this, irrespective of whether aliasing would work. I know malloc and new are supposed to yield memory suitable for any type to put in, alignment-wise. Does a simple (auto) char buffer have this guarantee? No. Boost has something that finds out alignment, but I think it is only "guessing", at least for some platforms.... Thomas [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top kanze@gabi-soft.fr Guest Posted: Wed Dec 15, 2004 6:19 pm    Post subject: Re: Reading objects from an array of char [email]glenlow (AT) pixelglow (DOT) com[/email] wrote: Quote: Now that you mention it, alignment would probably defeat this, Alignement definitly defeats it. Quote: irrespective of whether aliasing would work. I know malloc and new are supposed to yield memory suitable for any type to put in, alignment-wise. Does a simple (auto) char buffer have this guarantee? Not at all. In fact, all of the compilers I use will allocate a char[] at the first address which happens to be free, regardless of alignment. Quote: The aliasing issue is also problematic. The C99 standard (and probably the C++ standard if you interpret it compatibly) seems to think that (heap?) allocated memory "becomes" the first type you put in e.g. by memcpy or assignment. On the other hand, auto memory with a declared type is that declared type, so reading another type from it might violate the aliasing rules. You can get away with a lot in char[] buffers, provided you have adequat alignment. The problem is with declared types: if you declare a T as a local variable, the destructor for T will be called when the variable leaves scope. If you get funny and start doing things like var.~T(), you're likely to have problems when the destructor gets called. -- James Kanze GABI Software http://www.gabi-soft.fr Conseils en informatique orientée objet/ Beratung in objektorientierter Datenverarbeitung 9 place Sémard, 78210 St.-Cyr-l'École, France, +33 (0)1 30 23 00 34 [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Branimir Maksimovic Guest Posted: Thu Dec 16, 2004 1:38 pm    Post subject: Re: Reading objects from an array of char [email]glenlow (AT) pixelglow (DOT) com[/email] wrote: Quote: T t; char buffer [sizeof (T)]; new (buffer) (t); Now that you mention it, alignment would probably defeat this, irrespective of whether aliasing would work. I know malloc and new are supposed to yield memory suitable for any type to put in, alignment-wise. Does a simple (auto) char buffer have this guarantee? Nope. char buffers have alignment requirement of single char, and that is weakiest alignment requirement possible. If you have alloca function available, just use that instead of char buffer, it guarantees alignment for any type as malloc does. alloca can be implemented as single assembly instruction that adjusts stack pointer which means: it never returns NULL , and func(alloca()) is no no. That's why usage of alloca should be very very restrictive, but replacing auto char buffers with it, in order to use allocated memory as storage for any type is meaningfull and valid purpose. Quote: The aliasing issue is also problematic. The C99 standard (and probably the C++ standard if you interpret it compatibly) seems to think that (heap?) allocated memory "becomes" the first type you put in e.g. by memcpy or assignment. On the other hand, auto memory with a declared type is that declared type, so reading another type from it might violate the aliasing rules. With mentioned function your concern disappears. Greetings, Bane. [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top glenlow@pixelglow.com Guest Posted: Fri Dec 17, 2004 11:15 am    Post subject: Re: Reading objects from an array of char I know of alloca and while it helps with the specific code I submitted, it won't help with the general case of allocating char buffers e.g. embedded within an object. Cheers, Glen Low, Pixelglow Software www.pixelglow.com [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Branimir Maksimovic Guest Posted: Fri Dec 17, 2004 5:31 pm    Post subject: Re: Reading objects from an array of char [email]glenlow (AT) pixelglow (DOT) com[/email] wrote: Quote: I know of alloca and while it helps with the specific code I submitted, it won't help with the general case of allocating char buffers e.g. embedded within an object. Oh, I see, then use union eg: struct{ union{ max_allignment_req_t allignment; char buf[42];}; int a; }; Greetings, Bane. [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top kanze@gabi-soft.fr Guest Posted: Tue Dec 21, 2004 12:57 am    Post subject: Re: Reading objects from an array of char Branimir Maksimovic wrote: Quote: glenlow (AT) pixelglow (DOT) com wrote: I know of alloca and while it helps with the specific code I submitted, it won't help with the general case of allocating char buffers e.g. embedded within an object. Oh, I see, then use union eg: struct{ union{ max_allignment_req_t allignment; char buf[42];}; int a; }; And how do you define max_allignment_req_t? I currently use a union of five or six different types, but there's nothing in the standard which guarantees it (although I've yet to have it fail in practice). -- James Kanze GABI Software http://www.gabi-soft.fr Conseils en informatique orientée objet/ Beratung in objektorientierter Datenverarbeitung 9 place Sémard, 78210 St.-Cyr-l'École, France, +33 (0)1 30 23 00 34 [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Branimir Maksimovic Guest Posted: Wed Dec 22, 2004 2:52 am    Post subject: Re: Reading objects from an array of char [email]kanze (AT) gabi-soft (DOT) fr[/email] wrote: Quote: Branimir Maksimovic wrote: [email]glenlow (AT) pixelglow (DOT) com[/email] wrote: I know of alloca and while it helps with the specific code I submitted, it won't help with the general case of allocating char buffers e.g. embedded within an object. Oh, I see, then use union eg: struct{ union{ max_allignment_req_t allignment; char buf[42];}; int a; }; And how do you define max_allignment_req_t? From case to case. Usually it's enough to be size_t or long, but for vector types (eg SSE 128 bit types on intel) it is compiler and system specific thing, therefore typedef. I currently use a union of Quote: five or six different types, but there's nothing in the standard which guarantees it (although I've yet to have it fail in practice). It never fails, (although I should say "never say never"). Greetings, Bane. [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Allan W Guest Posted: Wed Dec 22, 2004 8:57 am    Post subject: Re: Reading objects from an array of char Quote: Branimir Maksimovic wrote: struct{ union{ max_allignment_req_t allignment; char buf[42];}; int a; }; [email]kanze (AT) gabi-soft (DOT) fr[/email] wrote: Quote: And how do you define max_allignment_req_t? I currently use a union of five or six different types, but there's nothing in the standard which guarantees it (although I've yet to have it fail in practice). Through the documentation for the platform (or trial & error, if all else fails), find the one type with the strictest alignment requirements. Put it with the other platform-specific code to make it easier to deal with when porting to other platforms... // platform_specific.h #ifndef PLATFORM_SPECIFIC_H #define PLATFORM_SPECIFIC_H #if defined(PLATFORM1) // Only for Platform 1 typedef double max_alignment_req_t; #elif defined(PLATFORM2) // Only for Platform 2 typedef long long max_alignment_req_t; // ... etc #endif #endif // PLATFORM_SPECIFIC_H [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Allan W Guest Posted: Wed Dec 22, 2004 8:58 am    Post subject: Re: Reading objects from an array of char Quote: Branimir Maksimovic wrote: struct{ union{ max_allignment_req_t allignment; char buf[42];}; int a; }; [email]kanze (AT) gabi-soft (DOT) fr[/email] wrote: Quote: And how do you define max_allignment_req_t? I currently use a union of five or six different types, but there's nothing in the standard which guarantees it (although I've yet to have it fail in practice). Through the documentation for the platform (or trial & error, if all else fails), find the one type with the strictest alignment requirements. Put it with the other platform-specific code to make it easier to deal with when porting to other platforms... // platform_specific.h #ifndef PLATFORM_SPECIFIC_H #define PLATFORM_SPECIFIC_H #if defined(PLATFORM1) // Only for Platform 1 typedef double max_alignment_req_t; #elif defined(PLATFORM2) // Only for Platform 2 typedef long long max_alignment_req_t; // ... etc #endif #endif // PLATFORM_SPECIFIC_H [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top kanze@gabi-soft.fr Guest Posted: Thu Dec 23, 2004 9:42 pm    Post subject: Re: Reading objects from an array of char Branimir Maksimovic wrote: Quote: kanze (AT) gabi-soft (DOT) fr wrote: Branimir Maksimovic wrote: [email]glenlow (AT) pixelglow (DOT) com[/email] wrote: I know of alloca and while it helps with the specific code I submitted, it won't help with the general case of allocating char buffers e.g. embedded within an object. Oh, I see, then use union eg: struct{ union{ max_allignment_req_t allignment; char buf[42];}; int a; }; And how do you define max_allignment_req_t? From case to case. Usually it's enough to be size_t or long, Except on a Sparc in 32 bit mode. Or most other 32 bit RISC architectures, I think -- very often, size_t and long are four bytes, but double requires 8 byte alignment. Quote: but for vector types (eg SSE 128 bit types on intel) it is compiler and system specific thing, therefore typedef. OK. You use a single typedef in an implementation dependant file. I rather suspect that that is superior to my current solution (the union described below) -- it allows for strange special cases, but doesn't impose excessive overhead otherwise. Quote: I currently use a union of five or six different types, but there's nothing in the standard which guarantees it (although I've yet to have it fail in practice). It never fails, (although I should say "never say never"). Given what I know of hardware design and architecture, I cannot imagine it failing. But my knowledge of hardware design and architecture are somewhat dated, and I've been surprised by other things already. -- James Kanze GABI Software http://www.gabi-soft.fr Conseils en informatique orientée objet/ Beratung in objektorientierter Datenverarbeitung 9 place Sémard, 78210 St.-Cyr-l'École, France, +33 (0)1 30 23 00 34 [ 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  Next Page 1 of 2   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.