I recently compiled R14B04 version of Erlang OTP suite and tried to make it work with Distel (current trunk version). Pretty good how-to is described on Clemensont’s blog but still had two major problems with Erlang/Distel/Emacs24 combo:
* Connect to node (C-c C-d n) and node-ping (C-c C-d g)
would always result in message “nodedown”
* (after first one is fixed): Sometimes when adding breakpoint
I would get a message “Module is not interpreted, can’t set breakpoints”
Solutions are as follows:
“Nodedown” problem
This message indicated that Emacs is unable to connect to inferior Erlang process. This is a well known problem of Distel/Erlang for versions of Erlang starting from R14. I downloaded the patch from Ralfs post and applied it to Distel epmd.el file. Now I can correctly connect to Erlang node.
“Module is not interpreted, can’t set breakpoints” problem
This is also Distel issue (so it seems). Fortunately, there is a simple workaround – when this message appears do the following:
* C-c C-d m (opens the monitor buffer)
* k (kills the monitor buffer)
* C-c C-d i (start interpreting) in the .erl buffer will now work.
If you never heard of mach system calls and specifically task_for_pid() call on Mac OS X, you can consider yourself lucky. If you want to stay that way – stop reading now! Still here? In that case let’s start with disclaimer – author of this text is not and can not be in any way responsible for damage produced or influenced by this article.
Prior to the Mac OS X 10.4.X (Tiger), it was completely legal for one process to control another for the purpose of influencing its execution (single stepping, resuming, stopping etc) and inspecting or modifying its memory and registers. In one of the patches for Tiger, this policy was changed so that only a process owned by root or with a “primary effective group of procmod or procview” has this privilege. In Leopard (Mac OS X 10.5), this policy was changed again (that much about consistent security policy – nice work Apple) such that an inspector process now depends on the security framework to authorize use of the task_for_pid system service which gives a process the capability to control another process.
To build a utility that will use task_for_pid(), you need to do the following:
1. Create Info.plist file which will be embedded to your executable
file and will enable code signing
2. Create self-signed code signing certificate using Keychain access
3. Write your program that uses security framework to obtain rights
to execute task\_for_pid()
4. Compile your program and code-sign it.
So let’s get started.
Step 1 – Create Info.plist
I used one of the standard Info.plist files I could find in Xcode and changed some particular parts as can be seen in following example:
When created – this certificate will be untrusted by default – change “When using this certificate” to “Always Trust” and you should be OK and ready to go for the next step.
Step 3 – Write your program
I wrote a very simple program that takes PID of a process you want to investigate (ran by your UID), connects to it and writes current register values for it. Code is pretty self-explaining so I won’t go into nifty details: