Fix r19564: Don't free decoder_ in ~Simulator().

Fix r19564: Don't free decoder_ in ~Simulator().

The simulator never allocates decoder_, and it is always passed from the
calling scope (if set at all). The caller should be responsible for
deleting it.

This fixes the failures in test-assembler-a64.cc.

BUG=

[jbramley, 2014-02-27 17:54:48]

Jochen, is this the right fix?

[jochen (gone - plz use gerrit), 2014-02-27 22:43:49]

I think we should instead allocate the decoder in the test as well.

For Simulator::current(), we have to rely on Simulator's dtor to cleanup
everything.

[jbramley, 2014-02-28 09:33:16]

On 2014/02/27 22:43:49, jochen (OOO from March 1) wrote:
> I think we should instead allocate the decoder in the test as well.
> 
> For Simulator::current(), we have to rely on Simulator's dtor to cleanup
> everything.

Ah, I didn't spot that. I don't like imbalanced allocation APIs* though; is
there no way that we can free the decoder when we free the simulator that
Simulator::current allocates? After having a quick look, I can't actually see
where the simulator is deleted. In ARM and MIPS, there is a TODO(146) which
suggests that it is not deleted.

* To clarify, if the caller allocates an object, then passes a reference or
pointer into another class, I don't expect that class to take ownership of it.
This is especially true for the visitor model decoder, where it might also be
attached to a disassembler, for example. If the simulator's destructor runs
before the disassembler's destructor, but both assume that the decoder is still
valid, then bad things will happen.

[jbramley, 2014-02-28 09:38:12]

On 2014/02/28 09:33:16, jbramley wrote:
> After having a quick look, I can't actually see
> where the simulator is deleted. In ARM and MIPS, there is a TODO(146) which
> suggests that it is not deleted.

Ah, I see that you're addressing that here:
https://codereview.chromium.org/180243010/

Still, can we free the decoder explicitly, or is there no nice way to do that?

[jochen (gone - plz use gerrit), 2014-02-28 09:45:04]

On 2014/02/28 09:38:12, jbramley wrote:
> On 2014/02/28 09:33:16, jbramley wrote:
> > After having a quick look, I can't actually see
> > where the simulator is deleted. In ARM and MIPS, there is a TODO(146) which
> > suggests that it is not deleted.
> 
> Ah, I see that you're addressing that here:
> https://codereview.chromium.org/180243010/
> 
> Still, can we free the decoder explicitly, or is there no nice way to do that?

src/isolate doesn't know about the decoder :-/

note that the disassembler has its own decoder as far as i can see.

[jochen (gone - plz use gerrit), 2014-02-28 10:45:00]

(I committed a fix for this to green the tree up)

[jbramley, 2014-02-28 12:39:57]

On 2014/02/28 10:45:00, jochen (OOO from March 1) wrote:
> (I committed a fix for this to green the tree up)

Thanks!

I think we could probably simplify the simulator constructors a bit, so I might
have a look at that to tidy things up.