Exactly. I want to be conservative and not make an optimization I can't prove,
however obscure the possibility.
As to user lzo's, they output lzs (i.e., our class-ified version of
javascript), so the class analysis tools can still be run over lzo's and do the
right thing. In fact, the test case that I compiled has a zillion independent
lzo-compiled libraries, so I am confident that this works as expected.
Eventually, I would like warnWithThis to default to on. And also have a
warnUnknownGlobal that defaults to on.
On 2011-03-09, at 14:16, Donald Anderson wrote:
> To clarify, you're concerned about:
>
> var myglobal = 7;
> class devilish {
> var myglobal = 666;
> }
> class myclass extends devilish {
> function void bar() {
> return myglobal;
> }
> }
>
> where for whatever reason, we don't know about devilish. I agree with your
> analysis. As a practical matter, the classes
> we tend to not know about are in the LFC or components (since we have the
> complete script for the user program (or can they
> break it up into LZOs?)) Even though we hope the LFC would never alias its
> own globals, the user could still create a global that
> matches a name in the devilish class.
>
> On Mar 9, 2011, at 1:42 PM, P T Withington wrote:
>
>> I missed a glaring logic error in JavascriptGenerator in my previous review:
>>
>> JavascriptGenerator#1419:
>>
>> The global table is a red herring. It started out as an attempted solution
>> to this problem, but it was an incorrect one. I'm sorry that it is still
>> there, because I think it led you down the garden path...
>>
>> We _can't_ remove globals from possibleInstance if the classdesc is not
>> complete. If it's not complete, we don't know that there might not be
>> instance vars that alias globals. If the classdesc _is_ complete, we don't
>> need possibleInstance, because we know the actual instance from the
>> classdesc.
>>
>> In fact, the compiler's knowledge of globals is really useless for the
>> purposes of computing if we need with(this) or not. The only use for
>> globals would be to emit a warning when there are free references that the
>> compiler cannot identify as a global.
>>
>> The real logic of this section should be:
>>
>> if classdesc complete OR if incomplete but all free references are in
>> classdesc:
>> rewrite all free references that are in classdesc
>> otherwise:
>> emit a warning and use with(this)
>>
>> As a bonus:
>>
>> if classdesc complete and there are free references _not_ in classdesc:
>> compare those free references to known globals and emit a warning for any
>> that cannot be identified
>>
>> Sorry that I missed this.
>
>
> --
>
> Don Anderson
> Java/C/C++, Berkeley DB, systems consultant
>
> voice: 617-306-2057
> email: [email protected]
> www: http://www.ddanderson.com
> blog: http://libdb.wordpress.com
>
>
>
>
>
