- We use the ASC compiler, which works quite differently from compc/mxmlc. It doesn't attempt any mapping between class hierarchy and directory structure. You feed it source, that's it. This compiler is under continuous development, and the one shipped in Flex SDK 3 is buggy, so we supply our own in lib/asc.jar for the time being.
- Since much of Narya relies on classes that don't exist on Tamarin (e.g. DisplayObject), we have to isolate a specific subset of source. Right now that's done ridiculously explicitly as a list of class files, in etc/asc-files.txt. In the long run we will want to resolve this more elegantly, but there are no trivial solutions at hand.
- We have not implemented the full flash.utils.describeType() for Tamarin, but rather a subset of its functionality: one method to see if one type derives from another, and another method to extract the public variable names in an object. The environment-specific code has been isolated to com.threerings.util.env.Environment, and build.xml simply copies either Environment.as-tm or Environment.as-fp into place depending on what it's targeting.
- The ASC compilation links against lib/builtin.abc, which supplies information about core classes such as Object, Array, etc, and lib/thane.abc which is our homegrown emulation library containing e.g. flash.net.Socket and other things we need that's not part of the open-source code Adobe donated to Mozilla, but rather still in their proprietary Flash Player code.
I'm committing thane.abc and builtin.abc directly now, but they will shortly be built by the Thane project and published as maven whatnots. We could auto-publish lib/asc.jar too, but I think we may want more explicit control over how we version that.
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5045 542714f4-19e9-0310-aa3c-eee0fc999fb1
(I just noticed marshallers are importing themselves but the flash compiler
doesn't seem to mind, so committing now)
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5043 542714f4-19e9-0310-aa3c-eee0fc999fb1
clean action script files with no unnecessary imports
* New class ImportSet to facilitate the operations being performed manually
throughout the code. Also removes the need for rawimports
* Threaded ImportSet through all places where imports and rawimports were used
and removed magic parameters to control import generation
* Added an ImportSet instance to ServiceListener, this was key to the action
script fixes
* Moved all the import tweaking and munging logic into the generate* methods
and made specific to code being generated (removed overreaching stuff
from ServiceMethod, this was part of the problem)
* Removed blanket imports from tmpl files
* Removed unused methods
* Fixed tabs from last commit
* Quick and dirty script to check for unused imports
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5041 542714f4-19e9-0310-aa3c-eee0fc999fb1
* ServiceMethod import options:
- Made argument type imports ignore first N args (0 for old behavior)
- Made marshaller listener imports optional
* Added InvocationTask._verbose and several dumps of current imports to help
track down future superfluous imports
* Made the import of ServiceListener method arguments' types optional
* Wrapped up most of processService code into protected inner class
ServiceDescription and made configurable:
- Optional import of the types of the first arguments of the service methods
- Optinoal import of the types of listener method arguments
* Changed generate methods to take the service class and construct their own
local ServiceDescription configured for their target
* Removed global import of service class and do it explicitly where required
(via method addServiceImport)
* Made some imports conditional
- InvocationResponseEvent for mashallers only if service has some listeners
- InvocationException for providers only if service has some methods with
listener arguments
* Removed unused variables and imports
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5036 542714f4-19e9-0310-aa3c-eee0fc999fb1
* Keep an Invoker member
* Added internal Launcher class to run ProcessBuilder.start on the invoker
thread because it is blocking
* Removed synchronized from methods, all ooo-servers are single threaded,
hurray!
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5033 542714f4-19e9-0310-aa3c-eee0fc999fb1
* Added RegistryTester smoke test which randomly creates and destroys agents
in batches
* Fixed bug where correct agent id was never making it to client
* Implemented Bureau.toString (+refactored summarize method)
* Fixed nasty bug due to my assumption that SafeSubscriber =~ Subscriber, they
are in fact completely different
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5032 542714f4-19e9-0310-aa3c-eee0fc999fb1
* Fleshed out client-side Agent class, made abstract
* Delegate the creation of the bureau director to subclasses of BureauClient
* Moved BureauDirector's BureauReceiver and Subscriber implementations to
anonymous classes that delegate to protected methods.
* Use SafeSubscriber
* Added new service method for notiyfing the server when the creation of an
agent fails so that the registry is not perpetually waiting
* Made BureauCredentials more conformant
* Made BureauRegistry respond properly to client logging off
* Got rid of superfluous uses of safeToString
* Changed Launcher to CommandGenerator and now use ProcessBuilder and
copyMergedOutput
* Changed bureau process parameters to server name and port instead of URL
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5030 542714f4-19e9-0310-aa3c-eee0fc999fb1
* Fixed signature of Receiver methods
* Implemented create/destroy mechanism and bootstrap request in BureauDirector
* Added BureauContext class to expose bureau information on top of usual
presents stuff
* Added BureauContext implementation in BureauClient
* Added code to kick off the director when the client logs on
* Added credentials for a bureau logon, which is just a token passed to
command line launcher
* Added concept of bureau types to distinguish between executable and command
line arguments; includes interface BureauRegistry.Launcher
* Implemented backend for the registry
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5027 542714f4-19e9-0310-aa3c-eee0fc999fb1
the deployment build must be using stricter options (need to find them so I can replicate these issues locally in the future)
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5026 542714f4-19e9-0310-aa3c-eee0fc999fb1
Made the generatorFn optional, in case the source values are already ready-to-go,
although 99% of the time it's going to be better to use the generator function
to get things ready to go because MultiLoader will take care of figuring out
the sources structure.
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5022 542714f4-19e9-0310-aa3c-eee0fc999fb1
logic. If we're inner and we're not static, we have a problem.
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5019 542714f4-19e9-0310-aa3c-eee0fc999fb1
non-static member. Pre-1.5 VMs can't do all of these things and Retroweaver
freaks out at runtime when put to the task.
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5018 542714f4-19e9-0310-aa3c-eee0fc999fb1
This is a potentially more dangerous change, but nothing
seems awry. Eyes peeled!
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5017 542714f4-19e9-0310-aa3c-eee0fc999fb1
and instead my checkin message got mangled.
*clearing throat*
..So our old way of doing things was pointlessly waiting a frame to handle
network events.
We also ran object subscribe/unsubscribe requests through the same queue,
and now those are processed immediately. Everything seems ok, but let's
keep an eye out.
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5016 542714f4-19e9-0310-aa3c-eee0fc999fb1
This was residual from the initial conversion from Java, but I don't think
it's very necessary on the flash side, and Zell is working to decouple
the presents layer from flash-player specific classes and events like
ENTER_FRAME.
So now we just process everything immediately.
I did some tests, it seems that the flash player dispatches ENTER_FRAME,
then it dispatches any network events, then the frame is rendered.
So our old way of doing things was pointlessly
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5015 542714f4-19e9-0310-aa3c-eee0fc999fb1
EnumSet, because Java's implementation of that is private). The
performance could conceivably be improved; Java uses two
implementations, one using a single long to store the element
flags (for enumerated types with 64 or fewer constants) and one
using an array of longs (for types with more than 64). This uses
an array of bytes.
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5011 542714f4-19e9-0310-aa3c-eee0fc999fb1
freakoutery as we try to stream the instance's implicit reference to its
containing class.
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@5001 542714f4-19e9-0310-aa3c-eee0fc999fb1
calling the callback. Loaders are not supposed to do this, but that may
only be when loading over the network and not from a ByteArray.
We know that retaining a reference to the MultiLoader fixes it, but
that increases the burden on the users of this class. Let's try this
instead, I think it will work: just retain references to active MultiLoaders
in the class.
Jamie's going to test this, as well.
(Also: in foreach mode, free references to results after reporting them.)
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@4999 542714f4-19e9-0310-aa3c-eee0fc999fb1
ChatCodes.PLACE_CHAT_TYPE that gets used for any feedback generated from the tell process. This
will allow chat systems that display tells in a custom way (tabs) to distinguish tell-related
feedback from other feedback.
git-svn-id: svn+ssh://src.earth.threerings.net/narya/trunk@4996 542714f4-19e9-0310-aa3c-eee0fc999fb1
Par Winzell