Comment #3 on issue 2148 by [email protected]: Null(Isolate* isolate)
crashes if we pass 0 for isolate
http://code.google.com/p/v8/issues/detail?id=2148
- I agree that V8 wants to avoid the NULL check.
- On the other hand, V8 bindings cannot avoid using an optional Isolate*
argument, since the method implementation is exactly the same except for
Null(Isolate*) or Null(), like this:
Handle<Value> foo(..., Isolate* isolate = 0) {
...; // method implementation
return isolate ? Null(isolate) : Null();
}
- Since it is too dirty to write a ternary operator here and there, V8
bindings should use a macro, like this:
#define V8_NULL(isolate) (isolate) ? (Null(isolate)) : (Null())
Handle<Value> foo(..., Isolate* isolate = 0) {
...;
return V8_NULL(isolate);
}
- Ideally we should use the macro only for the optional Isolate*, and
should use v8::Null(isolate) for the non-optional Isolate*. But It is
really buggy to mix them. If someone misused v8::Null(isolate) for a NULL
isolate in the future (Note: custom bindings are written by
non-V8-professional folks), it causes a crash bug. Thus I think V8 bindings
need to use V8_NULL(isolate) _almost_ always. "_almost_" means that V8
bindings can use v8::Null(isolate) directly for a few cases in which the
performance is really critical.
- Consequently, if V8 does not implement the NULL check, V8 bindings would
instead implement the macro that always does the NULL check.
So my feeling is that...
- If V8 could implement the NULL check without decreasing the performance,
that would be best. (BTW, what happens if you use cmov for the NULL check?
Does it still decrease performance?)
- Otherwise, V8 bindings would implement the macro for the NULL check and
use the macro in almost all places in V8 bindings, which would definitely
decrease performance.
Any good idea?
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
