Some timing regarding ADL..

[David Nolden]

2010/11/26 Milian Wolff <mail at milianw.de>:
>
> While I agree, doing more profilling would maybe be good. I will try to do
> such testing in the future, at least on KDevplatform/KDevelop or similar sized
> projects.
>
> But As Andreas said, it stayed in the branch for weeks without *anyone* being
> interested in it. Only a few users told me that they really like this branch,
> that it works fine for them. So I trusted the, merged it and saw that it was
> working quite nice. Then a few days later I noticed the slowdown during code
> completion which was fixed ASAP by Ciprian, so nothing wrong there.
>
> And I'd rather have new contributors doing good work (even if it's not
> perfect) then no new things at all. And it's not like I merged this stuff
> blindly, I did skim through the changes and didn't see anything obvious. I
> mean hell, I don't even know what ADL is nor when it's applicable and when
> not, I wouldn't have seen how to improve this case. We simply need people like
> you in our project, who know Cpp really well *and* know their way around the
> code base. And really, if it's just 8h of work, what's the problem? I spent
> much more on some much less important features...
>
> Anyways, this doesn't help at all and I don't see why I should defend myself
> here. Who codes decides, and you sadly disappear for weeks sometimes and left
> me no choice but to try to improve KDevelop in the best way possible that I
> can.

This wasn't ment as an attack against you. I think you're doing a good
job of tracking the merge requests, and I thank you for that, as I
don't have that much time. I'm also not talking about "doing new
stuff", but about "changing existing stuff".

I don't expect anyone except me to directly see the performance
problems on such a patch, I have a feeling for the expenses of each
duchain operation, after all I've written most of them. However, I
think I have verbalized my concerns regarding this patch.

The thing is, when I sometimes _have_ time, then I want to do
something constructive (I have a lot on my list which is more
important for productivity than ADL, for example incremental parsing).
I don't want to be _forced_ to spend hours and hours just to get the
application back into a (to me) acceptable state. This is frustrating,
first because I don't get to the things I really want to do, and
second because it gives me the feeling that the stuff would fall apart
if I wouldn't monitor it closely.

So I simply wish for some more conservative changing behavior
regarding existing and well-working code. I don't care if there appear
new "unfinished" features in the master, but the existing stuff should
keep working right. If you don't fully understand the patch (btw. I
also don't fully understand some of the C++ support patches, the stuff
is simply too complex), then at least do some really thorough testing.

In other words: Please don't do "David will fix it" commits. ;-)

Greetings, David