On 12 Nov 1998, Ben Gertzfield wrote: > >>>>> "Jason" == Jason Gunthorpe <[EMAIL PROTECTED]> writes: > > Ben> I think that for 0.3, I should at least set up the stubs for > Ben> a libapt-pkg and libapt-pkg-dev package in debian/rules, and > Ben> only actually use them when the package gets moved out of > Ben> experimental. > > Jason> I fear that even in stable we will have version problems - > Jason> it is the same issue every libray faces.. If fixing a bug > Jason> means adding a new variable to a structure then you have to > Jason> bump the soname because they are no longer binary > Jason> compatible. > > Are you sure? Should a library really depend on structures and not > functions/methods -- ESPECIALLY a C++ library? Why are we using > a library if this is changing so often, then?
Well, already the library is used by 7 (working on 10) apt programs so yes it makes a lot of sense to have a shared library. As for using a design more resilient to binary incompatibility.. that is a -vastly- more complex subject that not many people tackle and is certianly not something you want to attempt with C++. Yes there are solutions (SOM, COM, padding, shear force of will, etc) but they all have enough issues to make them not worth while for this. Jason
