observe OSX application launch events in Go via a cgo->Objective-C bridge
You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
Jordan Orelli 04abc47ac0 finished readme 8 years ago
ProcWatcher.h reorganize 8 years ago
ProcWatcher.m reorganize 8 years ago
README.md finished readme 8 years ago
procmon.c reorganize 8 years ago
procmon.go reorganize 8 years ago
procmon.h reorganize 8 years ago

README.md

procmon

procmon is a small Go program for OSX that watches App launch and terminate events in AppKit. This project demonstrates the following useful techniques:

  • how to call C code from Go with cgo
  • how to link Apple frameworks into a cgo project
  • how to call Go code from C with cgo
  • how to integrate the callback-based concurrency model of AppKit into Go's CSP-style concurrency model

The Go program directly links against the AppKit framework and uses it to subscribe to the NSNotificationCenter notifications generated by the OS when the user launches or terminates an App. The observer itself is written in Objective-C. The Objective-C observer is accessed by the Go program through a simple C function. The Objective-C observer, upon seeing notifications, invokes a Go function directly, passing control back to our Go program.

installation

Via Go Get: go get github.com/jordanorelli/procmon

You can also clone this package and build it with go build. The Go toolchain will invoke cgo transparently on your behalf. There should be no reason to invoke the cgo toolchain manually; that should only be of interest for debugging and learning purposes.

construction

procmon.go is the single Go file of interest to the Go toolchain.

Accessing cgo requires importing the pseudo-package C. It's important to understand that there is no literal C package in the Go standard library. Every project that uses cgo generates its own C package transparently.

When invoking import "C", the comment that immediately precedes the import directive contains a set of instructions to feed to cgo, as follows:

/*
#cgo CFLAGS: -x objective-c
#cgo LDFLAGS: -framework AppKit
#include "procmon.h"
*/
import "C"

Any lines starting with #cgo indicate cgo directives. These are passed to the cgo tool and are used to invoke the necessary compiler and linker. We use these flags to indicate that we want to invoke the Objective-C compiler and link agains the AppKit framework.

The other lines in this comment, that is, the lines that do not begin with #cgo, are passed to the C compiler as if they were in a C header file. For our project, that is just one line: the line that includes procmon.h, the header file for the C code that we want to access.

Down in the Go program's main function, we spawn a goroutine to listen on a channel for changes:

    go reportChanges()

The reportChanges function simply reads values off of a channel and prints them:

func reportChanges() {
    for change := range appChanges {
        switch change.stateChange {
        case stateStarted:
            fmt.Printf("started: %s\n", change.appname)
        case stateEnded:
            fmt.Printf("terminated: %s\n", change.appname)
        }
    }
}

We then invoke the C function MonitorProcesses, which we declared in our C header file. In Go, the invocation looks like this:

    C.MonitorProcesses()

And in our header file, the declaration looks like this:

void MonitorProcesses();

The cgo toolchain automatically associated procmon.c with our header file procmon.h that we imported in our cgo import comment. The implementation of the MonitorProcesses function appears in procmon.c:

void MonitorProcesses() {
    [[ProcWatcher shared] startWatching];
    [[NSRunLoop currentRunLoop] run];
}

This function does two things: it starts by accessing a singleton of our Objective-C class ProcWatcher (that's [ProcWatcher shared], which is defined here) and invoking its startWatching method. This subscribes our ProcWatcher instance to OS notifications. We'll come back to how the ProcWatcher subscribes to events in a bit.

sidebar: the Run Loop

After signing up for the notifications, we access the current processes' Run Loop with [NSRunLoop currentRunLoop] and call its run method to run the Run Loop. There are two reasons why we need to start the Run Loop. The first has to do with the mechanics of AppKit. NSRunLoop represents the event loop underpinning our notification center. Without the Run Loop running, the notification center won't ever pick up any notifications. Apple has a wealth of documentation with respect to the mechanics of Run Loops. If you're extremely curious about this part of the project, this page has some great literature on how the Run Loop is operating inside of AppKit.

The other reason we invoke the Run Loop in this way is that calling our Run Loop's run method blocks until the Run Loop itself terminates. Since we're invoking the C function from within the Go program's main function, we're blocking Go's main function, thus preventing main from returning. If main returns in the Go program, the Go runtime ends the process, which is not what we want. So this call gives us two things: it sets up the notification system infrastructure, and it prevents our program from terminating.

end sidebar: back to observing NSNotifications

The startWatching method accesses the current OSX user's NSWorkspace. The NSWorkspace handle allows us to hook into NSNotificationCenter to subscribe to notifications in the user's workspace. We specifically subscribe to the NSWorkspaceDidLaunchApplicationNotification and NSWorkspaceDidTerminateApplicationNotification notifications. Here's the subscription to the NSWorkspaceDidLaunchApplicationNotification notification, which is signaled by the operating system to inform an observer that an application has been launched by the user:

    void (^handleAppLaunch) (NSNotification*) = ^(NSNotification* note) {
        NSDictionary* info = note.userInfo;
        NSRunningApplication* app = info[NSWorkspaceApplicationKey];
        NSString* bundleId = app.bundleIdentifier;

        AppStarted((GoString){bundleId.UTF8String, bundleId.length});
    };

    id observerLaunch = [notifications
        addObserverForName: NSWorkspaceDidLaunchApplicationNotification
                    object: workspace
                     queue: [NSOperationQueue mainQueue]
                usingBlock: handleAppLaunch];

The notable feature here is: we pass a callback to our notification center, and within that callback, we invoke a curious function: AppStarted. That function isn't defined anywhere in our C or Objective-C code: it's defined in our original Go file procmon.go:

//export AppStarted
func AppStarted(name string) {
    appChanges <- appStateChange{stateStarted, name}
}

The //export AppStarted line before the definition of the Go function informs cgo that we'd like the function to be exported for use by C with the name AppStarted. I gave it the same name in C and Go but the names don't have to be the same; you could //export SomethingElse or even //export something_else and invoke it from C as something_else.

Because we're exporting a function for use by C, cgo will generate some bridging code in C that can be imported by our own C code. This allows our own C code to call back into the Go program and invoke Go functions. cgo will silently generate this C header file behind the scenes. That C header file, which is given the totally obvious and well-documented name _cgo_export.h is generated by cgo when you run go build, used to help compile our C code, and then deleted. You won't notice it getting written and deleted because it goes by so quickly, but it's there, and it's on disk when our C code gets compiled. In order to access those definitions from our C code, our C code has to import this fleeting header file. In this project, that inclusion happens in ProcWatcher.m here, which looks like this:

#include "_cgo_export.h"

Any time you access a Go function from C, you almost certainly need to import the _cgo_export.h header file. Importing this header file makes the Go function accessible to the Objective-C code as a C function which will automatically cross-call into Go, having the following signature:

void AppStarted(GoString p0);

And that is the function that we're invoking in our NSNotification observer when we call this:

AppStarted((GoString){bundleId.UTF8String, bundleId.length});

The GoString type is used to convert a null-terminated C string into a Go string, which appears as a struct at the C level, having the following definition (and transitive definitions):

typedef struct { const char *p; GoInt n; } GoString;
typedef GoInt64 GoInt;
typedef long long GoInt64;

That intermediate type is generated automatically by cgo in the intermediate header file _cgo_export.h (before deleting it). If you want to look at this file as a reference to see what is visible to your C code, you can manually invoke cgo with go tool cgo ..., passing in the files to be processed by cgo. You shouldn't need to do this in the normal case, but it can be useful for debugging.

Anyway, calling that C function invokes the corresponding Go function AppStarted, which writes a value onto a channel. That value is read off of the channel by our reportChanges goroutine and used to print out the name of the App that had been launched or terminated by the user.

Feel free to run the project on OSX. After running it, you should get no output and your terminal should be under procmon's control. Open another OSX app and you should see a line like started: com.apple.Notes (if you start Notes.app, for example). If you don't ... pull requests welcome ;)