I can affirm that the ConfigParser package is super easy to use, and I've used it to aggregate both ALPHA and X86 checkpoints together arbitrarily to make new checkpoints. An updater shouldn't be too hard, I've done a fair amount of fixup along the way in that regard.
I also want to point out I added a UNSERIALIZE_OPT_SCALAR macro in changeset 6820 that tries to unserialize a field and returns a bool if it's not there. This helped with old checkpoints because the unserialization code could set its own default for a field that was not serialized in old checkpoints, based on the return value of the macro. Lisa On Sun, Dec 12, 2010 at 5:40 PM, nathan binkert <n...@binkert.org> wrote: > > I think this is a good idea. I'm pretty sure you can parse the file with > the python ini package, correct? What would be great is if we could > automatically have a version added to the root object. This seems pretty > impossible though, so perhaps we should create a checkpoint_version > parameter on the root starting with 1. As versions change, it probably > shouldn't be too bad to increment the version number and call a method to go > from checkpoint version to max version by calling every function needed for > the translation. > > > I really like this idea. The ConfigParser stuff should do just fine > with our files and if there is something wrong in our grammar, we > should just fix it. Steve and I have even had several discussions > about using something like ConfigParser to build the checkpoint in > memory and then spit it out at the end instead of our total hack of > generating the checkpoint on the fly. This would make it far simpler > to serialize sub-objects. Anyway, maintaining a conversion utility > doesn't sound like a huge pain. We need to make some sort of > provision for marking changes in the checkpoint functions that are not > backward compatible. > > Nate > _______________________________________________ > m5-dev mailing list > m5-dev@m5sim.org > http://m5sim.org/mailman/listinfo/m5-dev > >
_______________________________________________ m5-dev mailing list m5-dev@m5sim.org http://m5sim.org/mailman/listinfo/m5-dev
