On 30 Jul 2001, Jean-Marc Lasgouttes wrote:
> BTW, do you have an idea about the gains (object size, compile times)
> obtained by your major chirurgical session?
As it is now, I don't expect any mayor changes in neither object size
nor compile times from a clean install. It's still too much of a tangled
web for it to pay off yet.
However, I've started chopping down the mess of unnecessary includes,
and over time, this will hopefully result in improved compile times. You
simply have to start somewhere.
The changes I have done are all conservative: I did not change anything
to the pimpl-idiom, and in general, I avoid code changes. Yes, I change
a lot of files, but all changes are simple in nature.
This is done in order to reduce the impact. Unfortunately, it seemed
that we had a bit of an impact anyway. I'm sorry for that, but once
these issues are sorted out, we should be monotonously better off than
earlier.
And for now, hopefully, the reduced dependencies will payoff when you have
to change some key files. For instance, changing LyXFont.h would almost
certainly require an almost complete recompilation of all the source
code a few days ago. Now, it might be down to 75%.
Similarly for other things.
In conclusion: Compile times and object size should mostly be similar,
maybe minimally improved. More importantly, changing of some header files
are not as cascading as earlier.
Finally, knowledge about where to attack the code in order to truly
gain something in compile times is growing, and we are going towards
the goal of reduce compile times.
When I'm satisfied with my "undangerous" operations, I'll begin to propose
a few code restructuring things that should really pay off. I believe we
have the potential to reduce compile times by maybe 10-25% from a clean
build, and, more importantly, make it less intrusive to change key files
in the future.
Greets,
Asger
