Complete the switch to ServiceObject

Switch Observatory to use only ServiceObject instances.

All fetched objects are upgraded to a ServiceObject.
ServiceObjectCache.
Uniform updating / refreshing.
Fix UTF8 string escaping in TextBuffer (should fix invalid character bug https://code.google.com/p/dart/issues/detail?id=16590)

R=turnidge@google.com

Committed: https://code.google.com/p/dart/source/detail?r=33541

[Cutch, 2014-03-10 15:01:29]

Not ready for review yet.

[Cutch, 2014-03-10 15:57:19]

PTAL

[turnidge, 2014-03-10 21:03:27]

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/app/location_manager.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/app/location_manager.dart:72: void
_setResponse(ServiceObject so) {
Consider using "obj" instead of "so" - I always read "so" as the word "so" and
not an abbreviation.  Your call.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/elements/class_view.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/class_view.dart:17:
cls.refresh().whenComplete(done);
Simpler now.  Nice.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/elements/class_view.html (left):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/class_view.html:29: <library-ref
app="{{ app }}" ref="{{ cls['library'] }}"></library-ref>
It's good to see the "app" arg go away.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/elements/code_view.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/code_view.dart:22: code.deref();
I'm wondering whether it makes sense to rename "deref" to "load" and "refresh"
to "reload".  It might make the relationship between the two methods more
obvious.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/elements/instance_ref.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/instance_ref.dart:43: if (expand)
{
Would it work to replace this with ref.refresh(done)?

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/elements/nav_bar.html (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/nav_bar.html:162: height:19px;
I have some questions about this css -- don't see a span tag anywhere in your
html.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/nav_bar.html:170:
</polymer-element>
Where do you use this custom element?  It looks like you aren't using it in
<script-view>.  Is this dead code?

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/elements/script_ref.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/script_ref.dart:25: String get
name {
I'm beginning to wonder if it makes sense to have this shared "name" function. 
Once each model defines a decent set of fields, it probably will make sense to
grab what we want directly from the model instead of building naming smarts into
the <script-ref> element.  I think this was more useful before we broke out the
various subtypes of service ref.

Thinking out loud.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/elements/script_view.html (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/script_view.html:15: <label>Show
Coverage Data</label>
Did you mean to use your nav-menu-item-checkbox thingy here?

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/elements/service_ref.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/service_ref.dart:25: return 'NULL
REF';
Do these nulls happen a lot?  Do they happen at all?  If not, maybe we could
just use the model object instead.

Not necessary for this cl.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/elements/service_view.html (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/service_view.html:72: <template
if="{{ object.serviceType == 'Smi' }}">
We should eventually merge all of the instance-view cases together into one
template if.  Not required for this cl.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/service/cache.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/cache.dart:17: bool cachesId(String
id);
What do you think of blank lines between these decls?  I would find it less
dense.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/cache.dart:19: bool
cachesType(String type);
I'm not convinced that we want to have a data-driven approach to dispatching to
the appropriate cache.  I don't see the benefit yet.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/cache.dart:20: /// Upgrades [obj]
into a [T].
More verbose comment about what "upgrading" means.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/cache.dart:21: T
upgrade(ObservableMap map);
Could upgrade be private (_upgrade) or is it part of the public api of a cache?

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/cache.dart:36: operator[]=(String
id, T so) {
"so" -> obj?

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/cache.dart:54: T
cachedOrUpgradeAndAdd(ObservableMap obj) {
I would like a different name.  Maybe something like "putIfAbsent".  When does
this get used -- cause I would like to look at the use of the function and
puzzle it out a bit.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/cache.dart:76: bool
cachesType(String type) => ServiceObject.unreffedType(type) == 'Script';
unreffedType... the double f is weird.  Maybe "stripRef"?

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/service/object.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/object.dart:13: ScriptCache
_scripts;
consider blank lines between decls.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/object.dart:78:
Future<ServiceObject> fetch(String serviceId, [Map options]) {
I find the distinction between the names "get" and "fetch" to be hard to make
out.  To me they mean the same thing as words.  What do you think of taking the
common one (get?) and keeping its name short and taking the uncommon one
(fetch?) and giving it a more awkward name.  Something like getSkipCache or
getNoCache?

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/object.dart:82: 
Remove one blank line here?

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/object.dart:383: // Preemptyively
mark that this is a reference.
typo in comment.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/service/service.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/service.dart:10: Isolate _isolate;
consider blank lines between decls.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/service.dart:36: assert(serviceType
== 'Error');
We want to consider whether we want to assert on bad data from the vm.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/service.dart:67:
Future<ServiceObject> deref() {
I suggested this elsewhere but will write it here again...

If we rename:

deref -> load
refresh -> reload

it may make their relationship more clear.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/service.dart:73: return new
Future(refresh);
Can this be just

return refresh();

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/service.dart:105: static bool
isRefType(String type) {
Could change to "isRef" instead of "isRefType".  Not sure.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/service.dart:110: static String
unreffedType(String type) {
Consider renaming to "stripRef"

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/service.dart:132: if
(isolate.scripts.cachesType(type)) {
The data-driven thing here doesn't buy much.  I would prefer a switch until
being data-driven buys us something.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/service/vm.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/vm.dart:16: Future<ObservableMap>
fetchMap(String id) {
What do you think of renaming this to "getAsMap"?

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/vm.dart:56: Future<String>
_fetchString(String id);
again, "get" and "fetch" mean almost the same thing to me, but I don't care here
because it is a private function.

https://codereview.chromium.org/192443004/diff/90001/runtime/platform/json.cc
File runtime/platform/json.cc (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/platform/json.cc...
runtime/platform/json.cc:679: static const uint32_t kMaxCodePoint = 0x10FFFF;
I don't like duplicating the existing unicode decode logic that is in
vm/unicode.{h/cc}.  Can you find a way to move that code up to platform so it
can be shared?  Perhaps the unicode change should become part of a separate cl?

https://codereview.chromium.org/192443004/diff/90001/runtime/vm/debugger.cc
File runtime/vm/debugger.cc (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/vm/debugger.cc#n...
runtime/vm/debugger.cc:717: jsobj->AddProperty("type", "StackFrame");
Why are you adding an id/type to the frame?

https://codereview.chromium.org/192443004/diff/90001/runtime/vm/object.cc
File runtime/vm/object.cc (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/vm/object.cc#new...
runtime/vm/object.cc:6879: return s.raw();
Why this delta?

https://codereview.chromium.org/192443004/diff/90001/runtime/vm/object.cc#new...
runtime/vm/object.cc:15415: ASSERT(Utf8::IsValid(result, len));
Is this expensive for our debug builds?

https://codereview.chromium.org/192443004/diff/90001/runtime/vm/service.cc
File runtime/vm/service.cc (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/vm/service.cc#ne...
runtime/vm/service.cc:525: jsobj.AddProperty("id", "");
Do we require an empty id now for Errors?  Why?

[Cutch, 2014-03-11 03:17:48]

Patch Set 8 is ready.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/app/location_manager.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/app/location_manager.dart:72: void
_setResponse(ServiceObject so) {
On 2014/03/10 21:03:28, turnidge wrote:
> Consider using "obj" instead of "so" - I always read "so" as the word "so" and
> not an abbreviation.  Your call.

Done.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/elements/code_view.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/code_view.dart:22: code.deref();
On 2014/03/10 21:03:28, turnidge wrote:
> I'm wondering whether it makes sense to rename "deref" to "load" and "refresh"
> to "reload".  It might make the relationship between the two methods more
> obvious.

I like it. Done.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/elements/instance_ref.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/instance_ref.dart:43: if (expand)
{
On 2014/03/10 21:03:28, turnidge wrote:
> Would it work to replace this with ref.refresh(done)?

Yep.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/elements/nav_bar.html (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/nav_bar.html:162: height:19px;
On 2014/03/10 21:03:28, turnidge wrote:
> I have some questions about this css -- don't see a span tag anywhere in your
> html.

Done.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/nav_bar.html:170:
</polymer-element>
On 2014/03/10 21:03:28, turnidge wrote:
> Where do you use this custom element?  It looks like you aren't using it in
> <script-view>.  Is this dead code?

Dead code. Removed.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/elements/script_ref.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/script_ref.dart:25: String get
name {
On 2014/03/10 21:03:28, turnidge wrote:
> I'm beginning to wonder if it makes sense to have this shared "name" function.

> Once each model defines a decent set of fields, it probably will make sense to
> grab what we want directly from the model instead of building naming smarts
into
> the <script-ref> element.  I think this was more useful before we broke out
the
> various subtypes of service ref.
> 
> Thinking out loud.

Agreed. Let's discuss how we can make generating display, hover text, and url
uniform.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/elements/service_ref.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/elements/service_ref.dart:25: return 'NULL
REF';
On 2014/03/10 21:03:28, turnidge wrote:
> Do these nulls happen a lot?  Do they happen at all?  If not, maybe we could
> just use the model object instead.
> 
> Not necessary for this cl.

Let's discuss this tomorrow.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/service/cache.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/cache.dart:21: T
upgrade(ObservableMap map);
On 2014/03/10 21:03:28, turnidge wrote:
> Could upgrade be private (_upgrade) or is it part of the public api of a
cache?

Done.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/cache.dart:36: operator[]=(String
id, T so) {
On 2014/03/10 21:03:28, turnidge wrote:
> "so" -> obj?

Done.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/cache.dart:54: T
cachedOrUpgradeAndAdd(ObservableMap obj) {
On 2014/03/10 21:03:28, turnidge wrote:
> I would like a different name.  Maybe something like "putIfAbsent".  When does
> this get used -- cause I would like to look at the use of the function and
> puzzle it out a bit.

I've renamed it to putIfAbsent. This gets called by the ServiceObject factory:
_upgradeToServiceObject

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/cache.dart:76: bool
cachesType(String type) => ServiceObject.unreffedType(type) == 'Script';
On 2014/03/10 21:03:28, turnidge wrote:
> unreffedType... the double f is weird.  Maybe "stripRef"?

Done.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/service/service.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/service.dart:10: Isolate _isolate;
On 2014/03/10 21:03:28, turnidge wrote:
> consider blank lines between decls.

Done.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/service.dart:36: assert(serviceType
== 'Error');
On 2014/03/10 21:03:28, turnidge wrote:
> We want to consider whether we want to assert on bad data from the vm.

I'm convinced that it's a bad idea. But, we need a uniform way of handling this
internally and bubbling the error back up to the application. Let's discuss
tomorrow morning.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/service.dart:67:
Future<ServiceObject> deref() {
On 2014/03/10 21:03:28, turnidge wrote:
> I suggested this elsewhere but will write it here again...
> 
> If we rename:
> 
> deref -> load
> refresh -> reload
> 
> it may make their relationship more clear.

Done.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/service.dart:73: return new
Future(refresh);
On 2014/03/10 21:03:28, turnidge wrote:
> Can this be just
> 
> return refresh();

Done.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/service.dart:105: static bool
isRefType(String type) {
On 2014/03/10 21:03:28, turnidge wrote:
> Could change to "isRef" instead of "isRefType".  Not sure.

Disagree. "isRef" is an instance method on the ServiceObject. We could probably
make these static functions private.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/service.dart:110: static String
unreffedType(String type) {
On 2014/03/10 21:03:28, turnidge wrote:
> Consider renaming to "stripRef"

Done.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/service.dart:132: if
(isolate.scripts.cachesType(type)) {
On 2014/03/10 21:03:28, turnidge wrote:
> The data-driven thing here doesn't buy much.  I would prefer a switch until
> being data-driven buys us something.

I've reworked this.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
File runtime/bin/vmservice/client/lib/src/service/vm.dart (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/vm.dart:16: Future<ObservableMap>
fetchMap(String id) {
On 2014/03/10 21:03:28, turnidge wrote:
> What do you think of renaming this to "getAsMap"?

Done.

https://codereview.chromium.org/192443004/diff/90001/runtime/bin/vmservice/cl...
runtime/bin/vmservice/client/lib/src/service/vm.dart:56: Future<String>
_fetchString(String id);
On 2014/03/10 21:03:28, turnidge wrote:
> again, "get" and "fetch" mean almost the same thing to me, but I don't care
here
> because it is a private function.

Renamed to _getString

https://codereview.chromium.org/192443004/diff/90001/runtime/vm/object.cc
File runtime/vm/object.cc (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/vm/object.cc#new...
runtime/vm/object.cc:6879: return s.raw();
On 2014/03/10 21:03:28, turnidge wrote:
> Why this delta?

Removed.

https://codereview.chromium.org/192443004/diff/90001/runtime/vm/object.cc#new...
runtime/vm/object.cc:15415: ASSERT(Utf8::IsValid(result, len));
On 2014/03/10 21:03:28, turnidge wrote:
> Is this expensive for our debug builds?

I haven't noticed but I've removed it.

https://codereview.chromium.org/192443004/diff/90001/runtime/vm/service.cc
File runtime/vm/service.cc (right):

https://codereview.chromium.org/192443004/diff/90001/runtime/vm/service.cc#ne...
runtime/vm/service.cc:525: jsobj.AddProperty("id", "");
On 2014/03/10 21:03:28, turnidge wrote:
> Do we require an empty id now for Errors?  Why?

We detect that a JSON map is a ServiceObject by checking for a type and an id. 

Longer term, the client side should canonicalize errors and never "reload" one.

[turnidge, 2014-03-11 16:40:06]

lgtm

https://codereview.chromium.org/192443004/diff/130001/runtime/bin/vmservice/c...
File runtime/bin/vmservice/client/lib/src/elements/script_view.html (right):

https://codereview.chromium.org/192443004/diff/130001/runtime/bin/vmservice/c...
runtime/bin/vmservice/client/lib/src/elements/script_view.html:30: <td style="{{
hitsStyle(line) }}">  </td>
Unrelated to this review: you could use classes/styles here to move the css
stuff out of the dart file and into the html file for hitstyle.

[Cutch, 2014-03-11 17:48:41]

Committed patchset #10 manually as r33541 (presubmit successful).

[Ivan Posva, 2014-03-12 21:17:15]

DBC -ip

https://codereview.chromium.org/192443004/diff/170001/runtime/vm/json_stream.cc
File runtime/vm/json_stream.cc (right):

https://codereview.chromium.org/192443004/diff/170001/runtime/vm/json_stream....
runtime/vm/json_stream.cc:405: 
?