Bug 7668

I had a look at doing this, it turns out to be quite hard/ugly.

The issue is that the event handler code is precompiled and takes a single argument, e. In order to get the same behaviour as anonymous python, we'd need to pass d into the locals() of the function but to do that, as far as I can tell we need exec/eval. exec/eval don't work with functions which take parameters.

There are tricks you can use to exec/eval functions with parameters but they work by wrapping the function, then we lose the ability to influence locals again.

I also tried injecting d into globals() however globals() is module specific and we therefore can't access globals past the first function exec where a different namespace takes over.

This left __builtins__ and something like:

diff --git a/bitbake/lib/bb/event.py b/bitbake/lib/bb/event.py
index f755eb2..80e3796 100644
--- a/bitbake/lib/bb/event.py
+++ b/bitbake/lib/bb/event.py
@@ -72,6 +72,7 @@ _eventfilter = None
 
 def execute_handler(name, handler, event, d):
     event.data = d
+    __builtins__['d'] = d
     try:
         ret = handler(event)
     except bb.parse.SkipRecipe:
@@ -87,6 +88,7 @@ def execute_handler(name, handler, event, d):
             logger.error("Execution of event handler '%s' failed (exit code %s)" % (name, exc.code))
     finally:
         del event.data
+        del __builtins__['d']
 
 def fire_class_handlers(event, d):
     if isinstance(event, logging.LogRecord):


does work. Do we accept that ugliness though?

Is there a reason we don't just add a second argument? I realize it's probably not conceptually ideal, since setting d is just a convenience, but it would work. As far as I know, the only code which compiles the handler functions is bb.event.register(), so it should be trivial to change the signature of the handler functions.

Alternatively, we could inject d = e.data as the first line of the functions, but then we'd need a way to adjust the line numbers for tracebacks, so probably not worthwhile.

diff --git i/bitbake/lib/bb/event.py w/bitbake/lib/bb/event.py
index ba0de11..1399ba5 100644
--- i/bitbake/lib/bb/event.py
+++ w/bitbake/lib/bb/event.py
@@ -73,7 +73,7 @@ _eventfilter = None
 def execute_handler(name, handler, event, d):
     event.data = d
     try:
-        ret = handler(event)
+        ret = handler(event, d)
     except (bb.parse.SkipRecipe, bb.BBHandledException):
         raise
     except Exception:
@@ -180,9 +180,9 @@ def register(name, handler, mask=[]):
     if handler is not None:
         # handle string containing python code
         if isinstance(handler, basestring):
-            tmp = "def %s(e):\n%s" % (name, handler)
+            tmp = "def %s(e, d):\n%s" % (name, handler)
             try:
-                code = compile(tmp, "%s(e)" % name, "exec")
+                code = compile(tmp, "%s(e, d)" % name, "exec")
             except SyntaxError:
                 logger.error("Unable to register event handler '%s':\n%s", name,
                              ''.join(traceback.format_exc(limit=0)))

http://git.yoctoproject.org/cgit.cgi/poky/commit/?id=df123ddab43a221cc26028e0aff211ac34c1eafc(In reply to comment #2)
> Is there a reason we don't just add a second argument? I realize it's
> probably not conceptually ideal, since setting d is just a convenience, but
> it would work. As far as I know, the only code which compiles the handler
> functions is bb.event.register(), so it should be trivial to change the
> signature of the handler functions.
> 
> Alternatively, we could inject d = e.data as the first line of the
> functions, but then we'd need a way to adjust the line numbers for
> tracebacks, so probably not worthwhile.
> 
> diff --git i/bitbake/lib/bb/event.py w/bitbake/lib/bb/event.py
> index ba0de11..1399ba5 100644
> --- i/bitbake/lib/bb/event.py
> +++ w/bitbake/lib/bb/event.py
> @@ -73,7 +73,7 @@ _eventfilter = None
>  def execute_handler(name, handler, event, d):
>      event.data = d
>      try:
> -        ret = handler(event)
> +        ret = handler(event, d)
>      except (bb.parse.SkipRecipe, bb.BBHandledException):
>          raise
>      except Exception:
> @@ -180,9 +180,9 @@ def register(name, handler, mask=[]):
>      if handler is not None:
>          # handle string containing python code
>          if isinstance(handler, basestring):
> -            tmp = "def %s(e):\n%s" % (name, handler)
> +            tmp = "def %s(e, d):\n%s" % (name, handler)
>              try:
> -                code = compile(tmp, "%s(e)" % name, "exec")
> +                code = compile(tmp, "%s(e, d)" % name, "exec")
>              except SyntaxError:
>                  logger.error("Unable to register event handler '%s':\n%s",
> name,
>                               ''.join(traceback.format_exc(limit=0)))

The reason I didn't do this is that that under that if block, the register function would appear to take functions as handlers as well as strings. Those functions would then break. At the time I assumed something used that code path, perhaps it doesn't though?

http://git.yoctoproject.org/cgit.cgi/poky/commit/?id=df123ddab43a221cc26028e0aff211ac34c1eafc
http://git.yoctoproject.org/cgit.cgi/poky/commit/?id=8c3dd2d9eb2c593d5b7cf28d7627e6725ab7dbfe
Still considering the change Chris proposed though...