On Sep 20, 2010, at 10:14 AM, Eric MSP Veith wrote: > > Ok, thought I'd better ask. I came to it because you mentioned automatic > generation. >
Well I mentioned automated doco generation rather than SWIG. But I should separate my opinion from generally using SWIG; perhaps someone can figure a way to use SWIG intelligently. I tried, but my POV wrto RPM is rather differently uniquely specific to me. Let me try to illustrate the SWIG issues with some examples: libselinux uses SWIG (iirc) for python bindings. What is gained is access to libselinux routines for doing development in python. Python is certainly easier development than C, nothing wrong there at all. The problem is that its common in C to return 0 on success, -1 on error. OTOH, in python, with exceptions, the programming paradigm for returns is often to return the same object (or nil) so that nested functions can be used for a series of common operations. While SWIG can certainly arrange for most any usage case, one ends up with a complex recipe for automagic generation for lots of rather simple getters/setters/creators/destructors. The code for all of getters/setters/creators/destructors -- conformant with the programmatic expectations -- usually isn't that hard to generate directly, using cpp macros if needed. For an alternative approach to SWIG, augeas python bindings use PYCpbject to attach to the Augeas symbols in -laugeas. Augeas (unlike SELinux) has an extremely well designed API, with ~10 methods of interest. For Augeas, SWIG is overkill, and PYCobject wrappings are enough. With languages that have -lffi, one can often call/return library routines without the need for SWIG wrapper code generation. GPSEE (and mozjs and other JS libraries) use -lffi rather than SWIG. Subversion makes extensive use of SWIG, and it seems to Just Work. But if you look at the SWIG wrappers, it appears (to me) that there's some serious efforts that are/were involved in using SWIG, the SWIG templates are quite complex/subtle/sophisticated/obscure. For RPM, there's probably a usage case for SWIG for "bindings-lite". I.e. most bindings want RDONLY access tags in headers from an rpmdb. The methods involved are sufficiently simple that SWIG automation might be a win because of the number of possible language bindings that might be targeted. 73 de Jeff ______________________________________________________________________ RPM Package Manager http://rpm5.org Developer Communication List rpm-devel@rpm5.org
