Profiling Perl 6 code on IRC

On the #perl6 IRC channel, we have a bot called camelia that executes small snippets of Perl 6 code, and prints the output that it produces. This is a pretty central part of our culture, and we use it to explain or demonstrate features or even bugs in the compiler.

Here is an example:

10:35 < Kristien> Can a class contain classes?
10:35 < Kristien> m: class A { class B { } }; say A.new.B.new
10:35 <+camelia> rakudo-moar 114659: OUTPUT«No such method 'B' for invocant of 
                 type 'A'␤  in block <unit> at /tmp/g81K8fr9eY:1␤␤»
10:35 < Kristien> :(
10:36 < raydiak> m: class A { class B { } }; say A::B.new
10:36 <+camelia> rakudo-moar 114659: OUTPUT«B.new()␤»

Yesterday and today I spent some time teaching this IRC bot to not only run the code, but optionally also run it through a profiler, to make it possible to determine where the virtual machine spends its time running the code. an example:

12:21 < moritz> prof-m: Date.today for ^100; say "done"
12:21 <+camelia> prof-m 9fc66c: OUTPUT«done␤»
12:21 <+camelia> .. Prof: http://p.p6c.org/453bbe

The Rakudo Perl 6 Compiler on the MoarVM backend has a profile, which produces a fancy HTML + Javascript page, and this is what is done. It is automatically uploaded to a webserver, producing this profile.

Under the hood, it started with a patch that makes it possible to specify the output filename for a profile run, and another one to clear up the fallout from the previous patch.

Then came the bigger part: setting up the Apache virtual host that serves the web files, including a restricted user that only allows up- and downloads via scp. Since the IRC bot can execute arbitrary code, it is very likely that an attacker can steal the private SSH keys used for authentication against the webserver. So it is essential that if those keys are stolen, the attacker can't do much more than uploading more files.

I used rssh for this. It is the login shell for the upload user, and configured to only allow scp. Since I didn't want the attacker to be able to modify the authorized_keys file, I configured rssh to use a chroot below the home directory (which sadly in turn requires a setuid-root wrapper around chroot, because ordinary users can't execute it. Well, nothing is perfect).

Some more patching and debugging later, the bot was ready.

The whole thing feels a bit bolted on; if usage warrants it, I'll see if I can make the code a bit prettier.