Change initialization of typeRepository [Was: [Bug 201455] kdevelop crashes on startup on Windows]

[Andreas Pakulat]

On 07.08.09 08:51:42, David Nolden wrote:
> Am Freitag 07 August 2009 06:47:54 schrieb Andreas Pakulat:
> > Thanks Yaron for the analysis. The attached patch should fix the
> > initialization, however I'd like to have a comment from David wether
> > this could potentially break something. David?
> >
> > I couldn't test wether this helps on windows, but at least it doesn't
> > seem to break anything on linux.
> 
> Looks like it cannot break anything, but I hate introducing these new 
> additional functions everywhere. The class "RepositoryManager" is ment to fix 
> these problems, and it already contains one function-call with each access to 
> the repository. See "RepositoryManager::operator->()". So instances of 
> RepositoryManager should just be global variables, and not be double-hidden 
> behind a function.
> 
> In this case, it _really_ matters, as this is accessed _very_ frequently.
> 
> So the better solution would be solving the problem in RepositoryManager: 
> Instead of directly taking the "shareMutex" variable in constructor, which 
> leads to the problematic function call, it should take another template 
> parameter like:
> "
> struct InstantiationInformationManagerData {
> static AbstractRepositoryManager* shareMutex () {
> 	return typeRepositoryManager();
> }
> };
> "
> 
> And that would then be used in RepositoryManager::createRepository, which 
> means that the function would be called on first use of the repository, 
> instead of during static initialization, without paying an additional 
> performance penalty.

I don't have the code in front of me right now, but the description lost me
after the "very frequently" :) Do you have time to do this over the
weekend? If not I'll take a shot and try to post a patch.

Andreas
 
-- 
Learn to pause -- or nothing worthwhile can catch up to you.