Broken on Trusty (because of ASDF 3) about roslisp

[airballking]

Thanks, @gaya- I'll look into it, tonight.

[airballking]

I had a look at the code, but am not 100% sure (from reading) whether split-string is actually necessary functionality for us. Right now, I don't have roslisp_repl installed on my 14.04 system. I will experiment with this on my 12.04 system with printouts on Monday morning.

[airballking]

I'm further digging into this. First observations:

• load-manifest is indeed calling (asdf:initialize-source-registry ...)
• asdf3 is split into two packages which are released together: asdf/defsystem (asdf itself) and uiop (some utilities)
• split-string is part of uiop

[airballking]

I attempted a first fix. You can find it here: airballking/roslisp@c980434

With this in place, load-manifest still works for me under Ubuntu 12.04 with ROS Hydro. @gaya- could you please check whether it also works under Ubuntu 14.04 with ROS Indigo? Thank you.

[gaya-]

The fix looks good. I like your check for ASDF version. In some slime code I have seen a compile-time version of this checks (e.g. https://github.com/slime/slime/blob/master/contrib/swank-asdf.lisp#L56), which could be even more efficient.

On a more global scale, it looks like load-manifest is one big mess. The most clean (and most time consuming...) solution would be, IMO, to rewrite load-manifest (at least some functions of it) and the rest of roslisp that works with paths and code compilation / building in such a way that would have clean implementations for all the different versions of ASDF (maybe not ASDF 1, it is really old after all). Something like:

#+asdf1
(defun some-function ...)

#+asdf2
(defun some-function ...)

#+asdf3
(defun some-function ...)

The thing that bothers me the most is the way roslisp works with the ASDF source registry. It uses the interface from ASDF 1 (asdf:*central-registry*) which was reimplemented much nicer in the newer versions. I haven't looked into this that closely though, so I might be wrong...

In addition to that, if there really is much code that actually isn't used at all on modern platforms with modern ROS versions, it would be nice to either completely get rid of it, or deprecate it, or label it with #+asdf1 or something similar.

@moesenle, you seem to be the only person who has a comprehensive knowledge about what's happening in load-manifest. Could you, please, give us some input on this?

[bbrieber]

I tested your fix and seems to work.
I was able to load cram packages in indigo.

I have not tested more complex stuff because I don't have my projects ported to indigo yet but it looks very promising...

[airballking]

@bbrieber thanks for taking a look at this. If this fix makes roslisp usable with the new ASDF, I'd vote for sticking with the fix for the time being.

@gaya- I suggest we postpone cleaning up load-manifest until after the ICRA submission deadline in autumn. What do you think?

[gaya-]

Agreed! Things turned out much simpler than I expected.
If I could just have another 24 hours to test it with more complex packages that include C++ code and a foreign function interface, that'd be great.

[airballking]

@gaya- Sure! I think a couple of days are no problem at all.

[gaya-]

Ok, I tested it on more complex projects, works well. I do get linkage errors from CFFI on really complex rosbuild packages but I'm pretty certain it has nothing to do with roslisp. Let's merge the fix.

[airballking]

Done. @Gaya thanks for testing!