[fpc-devel] Circular references and forward declarations

[Jonas Maebe]

On 06 Jan 2010, at 13:04, Florian Klaempfl wrote:

> Jonas Maebe schrieb:
>>
>> Another reason is probably to speed up the compilation:
>> * (re)compiling huge source files can be slow and/or require lots of
>> memory, depending on the used compiler (and debug information or
>> optimization settings)
>
> For single class c++ files, imo most of the time is spent into reading
> the huge headers which are often even not needed and a complete mess
> because nobody has an overview which classes are used and which not.

It depends. Since these compilers only see whatever is in the current  
source file (and its header files), putting more code in the same  
source file can significantly slow down interprocedural optimizations  
(as soon as one algorithm with quadratic complexity is involved). And  
inlining can significantly increase the complexity of single routines  
as well, making stuff such as register allocation much slower :)

At least compiling Apple's Mac OS X linker is fairly slow, even though  
it's only about 1MB of code. The reason is that they put virtually all  
of the classes into the header files, and then include those together  
in the main cpp file. See http://www.opensource.apple.com/source/ld64/ld64-95.2.12/src/ld/ 
. Pretty much the entire linker is implemented in  
MachOReaderRelocatable.hpp and MachOWriterExecutable.hpp (note that  
I'm not claiming that this is how typical C++ programs are structured,  
it's only to illustrate that compiling one huge file can be quite slow).


Jonas