[poky] About the operator "??="

[Tian, Kevin]

>From: Xu, Dongxiao
>Sent: Thursday, December 09, 2010 11:33 AM
>>
>> Thanks, that is quite a significant difference.
>>
>> Could you try working out how often getVar returns a "None" value?
>> I'm wondering if we should add something to getVar that if it is
>> about to return None, it checks for the default value flag and
>> returns that if set. Originally ??= wasn't implemented that way do to
>> performance concerns but I think we need to recheck that!
>
>Hi Richard,
>
>During the file parsing process, the getVar is called "1435156/1148654" times (actually I am
>still confused what "/" means...), and among them, 322121 times of getVar call return
>"None", thus about 1/3 or 1/4 of the total calls.
>
>I tried to use the following patch to have a test. Also I paste a profile log.
>From the result we can see it has over 15% performance gain. (from 39.3 secs to 33.7
>secs).
>
>In usermanual, the "??=" definition is:
>
>Setting a default value (??=)
>	A ??= "somevalue"
>	A ??= "someothervalue"
>If A is set before the above, it will retain that value.  If A is unset prior to the above, A will
>be set to someothervalue.  This is a lazy version of ?=, in that the assignment does not
>occur until the end of the parsing process, so that the last, rather than the first, ??=
>assignment to a given variable will be used.
>
>Our patch logic isn't strickly following the "??=" definition, since the assignment doesn't
>occurs in end of parsing process...
>Except this one, I think other behaviors for ??= do not change.

I think it is, as long as a getVar on a variable happens after all the evaluations of that
variable which I think should be true or else even current design has problem. In that way
when evaluating variable assignments, you keep 'defaultvalue' updated until the last "??="
assignment. Then later a getVar() on that variable happens which then you just return the
right defaultvalue. :-)

>
>diff --git a/bitbake/lib/bb/data_smart.py b/bitbake/lib/bb/data_smart.py
>index b9d9476..14ac305 100644
>--- a/bitbake/lib/bb/data_smart.py
>+++ b/bitbake/lib/bb/data_smart.py
>@@ -246,6 +246,9 @@ class DataSmart:
>     def getVar(self, var, exp):
>         value = self.getVarFlag(var, "content")
>
>+        if value == None:
>+            value = self.getVarFlag(var, "defaultval")
>+
>         if exp and value:
>             return self.expand(value, var)
>         return value
>diff --git a/bitbake/lib/bb/parse/ast.py b/bitbake/lib/bb/parse/ast.py
>index 870ae65..1ccda82 100644
>--- a/bitbake/lib/bb/parse/ast.py
>+++ b/bitbake/lib/bb/parse/ast.py
>@@ -109,10 +109,8 @@ class DataNode(AstNode):
>         if 'flag' in groupd and groupd['flag'] != None:
>             bb.data.setVarFlag(key, groupd['flag'], val, data)
>         elif groupd["lazyques"]:
>-            assigned = bb.data.getVar("__lazy_assigned", data) or []
>-            assigned.append(key)
>-            bb.data.setVar("__lazy_assigned", assigned, data)
>             bb.data.setVarFlag(key, "defaultval", val, data)
>+            bb.data.setVar(key, None, data)

This I think is incorrect. You actually wiped out previous assignment before ??=,
and thus end up to always have default value favored unless there're other direct
assignments after ??= evaluation. Just touch the flag should be enough.

>         else:
>             bb.data.setVar(key, val, data)
>
>@@ -301,10 +299,6 @@ def handleInherit(statements, m):
>     statements.append(InheritNode(m.group(1)))
>
> def finalize(fn, d, variant = None):
>-    for lazykey in bb.data.getVar("__lazy_assigned", d) or ():
>-        if bb.data.getVar(lazykey, d) is None:
>-            val = bb.data.getVarFlag(lazykey, "defaultval", d)
>-            bb.data.setVar(lazykey, val, d)
>
>     bb.data.expandKeys(d)
>     bb.data.update_data(d)

Thanks
Kevin