The new scripts support both ideas now: You can use any repo from your 
filesystem and only if you specify the dest parameter or use a revision other 
than "WORK" the code will be copied to ~/lab/revision-cache.

As the blacklist for *.obj files also has been implemented I think I can close 
this issue, but feel free to reopen if something is missing.

I like your ideas from the last paragraph. I'll handle it accordingly
for now.

I think having a cached checkouts directory might still be useful to save
multiple compilations when doing multiple experiments for the same base
revision, and also to facilitate things like compiling on a different machine
from the one where the experiment is generated (which was occasionally necessary
for me for various reasons).

One possibility is to consider "checkouts" a "cache" of checked out (and
compiled) planner versions that can optionally be used, but doesn't have to be
used. I wouldn't mind making the name clearer in this respect (e.g. calling it
"cached-revisions", "revision-cache" or whatever; I guess "checkout" is not
technically the term one would use for hg in this case anyway?).


But even if this would be a nice to have feature, it's not of critical
importance, I would say. With the scripts living outside the main codebase, I
guess it would make sense to augment the current "use the working copy code"
feature so that we can tell it where the working copy to use lives in the
filesystem. If we do that, we could also support multiple different revisions
living in different directories, which would be enough to quite easily emulate
something like "cached-revisions" manually.

For the new-scripts 2.0, do you think we should make the hg clones directly into 
the experiment dir or use a "checkouts" dir as before?

add option to blacklist files (e.g. *.obj).

This would remove the maintenance burden of keeping the binaries up-to-date.

should be optional at least for local experiments

after copying call ./build_all in the experiment directory