Since Dion and I were really the only ones writing code for the memory profiling runtime, we haven’t actually documented the API yet. It’s pretty simple, though; every object in the Firefox runtime is referred to by a unique integer ID, and various functions can be used to get metadata about an object. For instance, getObjectInfo(id) returns a JSON-able data structure containing information about the object with the given ID, such as its prototype, its parent (i.e., its global scope), what other objects it points to, and so forth. Dion then used this API to write a memory profiler server.
One nice thing about the memory profiling runtime, though, is that it doesn’t have to embed a web server: it could just get some information about the heap and return it as a JSON object to the Firefox runtime, which could then do something useful with it. There’s lots of interesting things we’d like to eventually see—for instance, a visualization of the JS heap over time would be pretty cool.
Of course, we also ran into our own share of problems, many of which we’re still trying to resolve. They’re pretty technical in nature, but you’re welcome to read the write-up on our Memory Profiling Notes from July 2009 on the wiki.
Creating this tool wouldn’t have been possible without the excellent SpiderMonkey documentation on the Mozilla Developer Center or the friendly folks on #jsapi at irc.mozilla.org—particularly David Baron and Blake Kaplan. I’m definitely looking forward to tinkering more with the JSAPI in the future, and working with Dion and Ben to make developing for the Open Web a lot more fun.
Update: I’ve since documented the API for the memory profiling binary component and runtime.