random-state.net

Nikodemus Siivola

<< next | top | previous >>

August 25th 2004 #
random, August 25th 2004

Common-lisp.net was broken in again — or so we thought and shot it down. Turns out however, that we got spooked by a misconfiguration that caused stuff that wasn't even supposed to be installed (specifically Samba) to run. Embarrassing but true. In retrospect I do still feel that the decision to take CLnet down was a reasonable one, given the amount of information he had at hand at the moment. We're now in the process of moving the hosting to better premises, which was planned anyways — this just prompted us to move the schedule ahead. Mario Mommer had been working on the new setup for some time now, so most things are in place already. So maybe it's not all dark and gloomy. CLnet is hopefully up and running soon again — maybe today, a few days on the outside if the road turns out to have unexpected bumps.

More upbeat stuff:

Debugging SBCL with gdb can be painful — even more so then gdb in general. (Of course, in the normal course of lispy stuff you don't do this. When hacking on SBCL itself it is sometimes unavoidable.)

It might be interesting to teach SBCL itself to debug another SBCL via ptrace (or other, equivalent interfaces.) SBCL already knows how to deal with its own data and stack-frames, so large chunks of functionality are already somewhat there — teaching SBCL more about foreign frames would probably be inevitable, and replacing instructions with trapping ones would have to be done... except that the latter is more-or-less already there, in the form of non-encapsulating trace and (broken) breakpoint based single-stepper. With some rearrangement groveling the debuggee's type tags from its heap should also be possible — and if all else fails they could be written out as sexps during genesis, right alongside header generation.

Oh, I like this idea! Not that it'd be either easy or straightforward by any means, but definitely in the realm of doable, and definitely shiny. Any takers?