Paul Pluzhnikov wrote: > "aaragon" <[EMAIL PROTECTED]> writes: > > > I'm using the g++ compiler, and I run into a problem when I try to > > separate the declaration in the .h file from the implementation in the > > .cpp file. It seems that when I do this, I have linking problems (only > > when I use templates in the declarations). Does anyone know why is > > this? > > Because gcc doesn't support external templates (few (if any) other > compilers do). > > Some compilers play tricks: when you '#include "foo.h"' which > declares templates, they "automagically" look for foo.cpp, and > include it in the current compilation unit, saving resulting > instantiated methods in a template "repository". > > All such schemes (in my experience) are buggy and eventually cause > extreme pain, when partial rebuilds fail to link, because repository > is not updated correctly under some conditions. > > On ELF platforms none of this trickery is necessary anyway, because > template bodies are put into linkonce sections, so final executable > contains only one copy of the instantiated method (i.e. you don't > get code bloat). > > > I ended up compiling only the .cpp files (which have an #include > > "respective_file.h" only) and putting all the implementation details in > > the .h files. > > That's the most portable approach. > > Another alternative is to manually manage instantiation via explicit > instantiation requests. >
What do you mean by "explicit instantiation requests"? > Cheers, > -- > In order to understand recursion you must first understand recursion. > Remove /-nsp/ for email. _______________________________________________ help-gplusplus mailing list help-gplusplus@gnu.org http://lists.gnu.org/mailman/listinfo/help-gplusplus
