Porting uses of 'accuracy' in KMimeType API

[Kevin Funk]

On Friday 12 September 2014 10:50:36 David Faure wrote:
> On Friday 12 September 2014 09:39:42 Kevin Funk wrote:
> > Heya,
> > 
> > Context: Forward-porting some patches in KDevelop involving KMimeType API.
> > 
> > I've just noticed that in Qt5, QMimeDataBase/QMimeType doesn't allow me to
> > retrieve the accuracy of a match anymore, while KMimeType did. For
> > example,
> > compare the possible arguments for QMimeDataBase::mimeTypeForUrl vs.
> > KMimeType::findByUrl.
> > 
> > What's the suggested way to deal with this? As far as I can see, there are
> > no porting notes about that particular matter.
> 
> You're telling me everything I know already, but not the important bit which
> is: what do you need the accuracy for?

Well, this is ancient, performance-critical code inside KDevelop. We basically 
cache a extension -> language mapping for performance reasons. Just recently 
we introduced code that avoids caching extensions for which the KMimeType 
lookup yielded a very low accuracy.

Also see: https://git.reviewboard.kde.org/r/120085/

Note: I never touched that code myself.

> The mime matching tries its best to find out the mimetype based on what you
> give it (filename and/or content). The output of that is the best idea we
> have about the mimetype. What difference does it make if that's a 20% or an
> 80% accuracy?
> 
> > (Interestingly, the QMimeDataBase implementation internally plays around
> > with accuracies a lot, it just isn't exposed in the public API)
> 
> Yes, as per the mimetype spec.
> 
> My guess is that it was used as a workaround for bad api or implementation,
> like "try with the content, and if that's not accurate, try with the
> filename", or vice-versa. But QMimeDatabase has the right all-in-one methods
> for this, following the spec, i.e. "if both filename and content are
> available, then trust the filename, unless there are multiple mimetypes
> claiming the same glob, then look at content, and select the one that
> matches (or is a subclass of) one of the candidates based on the filename".
> 
> The intended way to use this is to use isDefault() on the QMimeType,
> if that's true then it couldn't find any specific mimetype for the file
> (i.e. basically accuracy 0). And otherwise, that's the mimetype.

-- 
Kevin Funk | kfunk at kde.org | http://kfunk.org