Re: On splitting headers

From: vesa_karvonen (vesa_karvonen_at_[hidden])
Date: 2002-02-06 10:30:24

--- In boost_at_y..., "Brey, Edward D" <EdwardDBrey_at_e...> wrote:
>From: vesa_karvonen [mailto:vesa_karvonen_at_h...]

>>When you find that you need scoped_ptr in some translation unit,
>>you include it there and recompile only the translation unit.
> 
> The problem is that you will may well compile that translation unit
> fifty times in the course of making code changes. That's fifty
> compilations of scoped_ptr, which could have been just one.

This is one of the things that I have been trying to tell people 
about. It really doesn't matter whether it takes 1 or 10 seconds to 
compile. The problem is when compiling time gets to the minute range. 
When that happens your productivity drops sharply. You start spending 
time thinking about how to avoid compiling.

I've worked on projects where a change to a header was equal to at 
least a couple of minutes of compiling and >30 minutes in the worst 
case. I will never want to go back to such a project. I will much 
rather have compiling time distributed so that regardless of which 
header I change, the time it takes to compile the project afterwards 
is (almost) minimal (less than 30 seconds).

I will never want to work in a project where I'm told to avoid 
modifying a particular header or a group of headers. If I'm presented 
with such a project, I will either leave or get permission to 
restructure the project.