[PyQt] Looking for clarification and ideas on %ConvertToSubClassCode for PyKF5
Phil Thompson
phil at riverbankcomputing.com
Thu Jan 5 10:01:33 GMT 2017
On 5 Jan 2017, at 9:27 am, Shaheed Haque <srhaque at theiet.org> wrote:
>
> Hi all,
>
> The effort to generate PyKF5 semi-automatically (to replace PyKDE4) is making some headway but I am uncertain as to how best to tackle the problem of %ConvertToSubClassCode. The first clarification question is to check I have understood correctly that the basic process for designing %ConvertToSubClassCode is:
>
> For each .sip file in a SIP module
> For each type in the .sip
> For all known subclass trees
> Generate a fragment of code (like the sample below)
>
> Is that right?
>
> Assuming that is correct, the second question is about how to implement the above. In the PyKDE4 days several thousand lines of code like this were lovingly hand-crafted:
>
> %ConvertToSubClassCode
> // CTSCC for subclasses of 'KConfigSkeletonItem'
> sipType = NULL;
>
> if (dynamic_cast<KCoreConfigSkeleton::ItemBool*>(sipCpp))
> sipType = sipType_KCoreConfigSkeleton_ItemBool;
> else if (dynamic_cast<KCoreConfigSkeleton::ItemDateTime*>(sipCpp))
> sipType = sipType_KCoreConfigSkeleton_ItemDateTime;
> else if (dynamic_cast<KCoreConfigSkeleton::ItemDouble*>(sipCpp))
> sipType = sipType_KCoreConfigSkeleton_ItemDouble;
> else if (dynamic_cast<KCoreConfigSkeleton::ItemInt*>(sipCpp))
> {
> sipType = sipType_KCoreConfigSkeleton_ItemInt;
> if (dynamic_cast<KCoreConfigSkeleton::ItemEnum*>(sipCpp))
> sipType = sipType_KCoreConfigSkeleton_ItemEnum;
> }
> ...
>
> (I have de-duplicated this by hand, and I still have over 1000 lines of code). Now I have over 2000 .sip files across over 100 components that make up PyKF5 and a minor nightmare in prospect :-). It seems to me that I have several manual options:
> • Ignore the problem at this level, and only manually create %ConvertToSubClassCode on an as-requested basis, release-by-release. (On occasion, I may be able to use the PyKDE4 code as a starting point, but the shape and details of the type hierarchy in PyKF5 is often rather different).
> • For each of the 100+ components, don't release the component until the work is done.
> Neither of these is attractive from a user perspective, and perpetuates the problem from a maintainer perspective. The other way forward might be some kind of automation in the SIP generation automation layer:
> • Let's say that the presence of RTTI is a given.
> • We know we are dealing with a Qt-like system including all its magic macros and the like.
> • We can run some kind of offline analyser (the SIP generator uses libclang, and can either operate on a single .h file at a time, or we can even run over a whole directory hierarchy if needed).
> Has anybody attempted this before? Any other suggestions? Can we do something at runtime?
>
> All input welcome...
For the record I've sent you a copy of the Python scripts I use for PyQt5 via private email. If anybody else wants a copy then please let me know.
Phil
More information about the PyQt
mailing list