|
Authored by: Anonymous on Wednesday, May 09 2012 @ 12:58 AM EDT |
Obviously that should have been "Waste disposal operative" == Garbage
collector [== Binman in UK].[ Reply to This | Parent | # ]
|
|
Authored by: Anonymous on Wednesday, May 09 2012 @ 06:59 AM EDT |
Where's that post on "the courtroom API"? [ Reply to This | Parent | # ]
|
|
Authored by: greed on Wednesday, May 09 2012 @ 02:28 PM EDT |
I would tend to agree that "simulated execution" sounds a bit like
"parsing".
But what if they actually do "simulate" execution: clone the stack and
whatever machine registers the virtual CPU has. Clone the heap. Set up traps
so that only certain instructions and address ranges are acceptable. Run
execution against the clone, until X instructions are executed, Y state change
is made, or one of the traps fires.
(For "clone" above, I'm assuming some copy-on-write technique is
available. Or even just trap-on-write.)
Now you can see the changes that would have been made; you control the VMMU, so
you know which addresses were accessed. You can see what changes were made.
And you did it without having to "understand" the virtual instructions
in two different places.
Not being willing to pull the JVM source at this point, and certainly not
willing to read the patent--we're already being sued for software patent
infringement--I'm not sure how it is actually implemented.
But there is certainly at least one way to "simulate" execution in a
way that almost sort of makes sense. (Anything that eliminates dual maintenance
should at least be considered useful; dual maintenance is a fault injection type
you don't need.)
[ Reply to This | Parent | # ]
|
|
|
|
|