-
Notifications
You must be signed in to change notification settings - Fork 850
New issue
Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? # to your account
Introduced a plugin-API and show usage with rhino-xml #1699
base: master
Are you sure you want to change the base?
Conversation
@@ -33,7 +33,6 @@ public void initFromContext(Context cx) { | |||
allowMemberExprAsFunctionName = cx.hasFeature(Context.FEATURE_MEMBER_EXPR_AS_FUNCTION_NAME); | |||
strictMode = cx.hasFeature(Context.FEATURE_STRICT_MODE); | |||
warningAsError = cx.hasFeature(Context.FEATURE_WARNING_AS_ERROR); | |||
xmlAvailable = cx.hasFeature(Context.FEATURE_E4X); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Tried to remove all XML-Lib dependency to rhino-xml. There is still org.mozilla.javascript.xml.XMLLib
and org.mozilla.javascript.xml.XMLObject
in rhino base module. They are a bit difficult to remove
* | ||
* <p>The default implementation now prefers the DOM3 E4X implementation. | ||
*/ | ||
protected org.mozilla.javascript.xml.XMLLib.Factory getE4xImplementationFactory() { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I have removed both checks. If rhino-xml
is on classpath, we should also have org.w3c.dom.Node
.
Can re-add this check, if wanted.
Note: Kit.classOrNull
can be very slow, if class is not found, as every check a ClassNotFoundException happens
default void initSafeStandardObjects(Context cx, ScriptableObject scope, boolean sealed) {} | ||
|
||
/** Initializes the (unsafe) standard objects. */ | ||
default void initStandardObjects(Context cx, ScriptableObject scope, boolean sealed) {} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
CHECKME: Do we need the difference between safe and unsafe objects in the long term?
I think that this is the right direction, and we should pursue it. I know that with these changes, the tests work, because all the tests in the "tests" module will include all the modules. Can we also add a test in the "rhino" module that will ensure that whatever is supposed to happen happens if you try to use the E4X syntax when the rhino-xml module is not in the classpath? |
I have tested to remove the rhino-xml dependency from A few tests like I've added a small A/B test. In the long term we might think about to split the Roland |
Sorry to be dribbling questions out here one at a time but I want to understand the implications of this. I THINK that what happens here now is that to use E4X features, you need "rhino-xml" in your classpath, AND you need to enable the FEATURE_E4X flag in the context. (I admit, though, that I don't use this feature myself so I don't know.) I like using the plugin model here, but does the user experience change? If it won't, then I don't think we should deprecate the FEATURE_E4X flag -- I don't think that the special E4X syntax should be enabled unless you set this feature flag, since E4X requires the special syntax and all that. |
Not sure if we ought to deprecate/remove the E4X feature flag: even if the 'rhino-xml' is on the classpath, it doesn't mean every Context instance will have/need/is allowed to use E4X |
Which begs the more general question: just the fact that a plugin is available on the classpath, does that mean it needs to be automatically initialized into each TopLevel? Or does it do that only in for example the shell and/or by default when using the standard ContextFactory, with the ability for integrations to take control over it, for more fine-grained initialize behavior? |
All good, it's important to know the impact here.
The current implementation (=without this PR) behaves this way:
So, setting the The new implementation (=with this PR) behaves this way:
This change should only affect the edge case, when you set the FEATURE_E4X flag, but no dom3 is on classpath (current implementation can compile e4x code, the new one wont, if rhino-xml is also missing or probably run into an error elsewhere)
I would enable the e4x syntax and object only, if the plugin is in place. Now you need both, the plugin AND the flag. See here Do you think, there are use cases, where you want to compile e4x code, but not run it? When we decide to deprecate the E4X flag, this means, that the user must use a custom context factory and control e4x availability with the plugin presence or not. At this point I am also a little unsure whether we can burden this breaking change to the user
By default, the contextfactory loads all plugins from the serviceLoader here And you can set your own list of plugins with setPlugins Admittedly, this customization is a bit limited or cumbersome because unmodifiable lists are used to ensure that checkNotSealed etc. will work properly (I do not want to do something like I'm also unsure if we should get the plugins for each new factory from the ServiceLoader or only for the one created with the default constructor (or the GlobalFactory). I could imagine that if I call
Yes, that's what I would like to do, but the current implementation is probably not 100% suitable for that (see above) |
OK -- looking for more comments here from some of the others, but in general this sounds like a good approach. However, I started this conversation because I see that the "FEATURE_E4X" flag is marked deprecated in this PR. I think that if we want the above -- namely, to have E4X, you need BOTH to set the feature flag AND have rhino-xml in the classpath -- then we should not deprecate it. |
Yes, at the moment you still need both because I don't want to introduce incompatibility (= existing code does not compile). I would prefer if we can remove the flag completely and control the e4x support solely by whether the plugin is registered in the ContextFactory or not. But this means, the current API breaks. That's the reason, I've deprecated it. A plan could be:
I'll be honest here that I can't estimate exactly how big the outcry will be if the API breaks in a 1.7.x version. Maybe we should really wait until 1.8.x or 2.x to remove it |
I've rebased this PR and submitted a different approach how to deal with plugins and autoloader: By default, the standard-constructor of ContextFactory will create a factory with all plugins, found by serviceloader. Plugins can disabled with a system propery, so if you set Nevertheless, you can still construct your own ContextFactory, where you manually add the XML-plugin. In this use case, you have the standard factory, without e4x support and an explicit one with e4x support (of course, you still have to set the FEATURE_E4X, which I suggest to remove.) Also the other way is possible: |
*/ | ||
public static final int FEATURE_E4X = 6; | ||
@Deprecated public static final int FEATURE_E4X = 6; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think that this is a fine change and we should do it, but I still don't think that we should deprecate this flag until it really does something. IMO, deprecating this flag is like saying that we're going to deprecate the E4X feature and I don't think that we're doing that.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think there is still a misunderstanding here...
until it really does something
the flag currently does something, it controls, if E4X is enabled, when the plugin is present. So currently, you have to enable the flag (which is done by default for >= 1.6) and ensure, that the plugin is in classpath. I see these options:
add xml-plugin to rhino-all and keep the flag
If someone uses rhino-all as dependency, he will get the plugin automatically and the flag behaves like before. This solution provides maximum backward compatibility (and this is how it is implemented now, but it is annotated as deprecated)
remove the flag completely
As the flag is enabled by default (at least for languageVersion >= 1.6), we could move that check to the plugin. So if someone has decided to overrride FEATURE_E4X in his factory, this will be a breaking change, as code will no longer compile.
something between the 2 solutions
deprecate the flag - with a clear hint, that disabling E4X (that is the purpose that the flag currently has IMHO) with the flag will be removed in the future and should be controlled with the plugin mechanism.
This is my first try to introduce a plugin-api, so that logical separate code parts (rhino-xml, liveconnect) can register them into the rhino engine.
Before I continue here and invest a lot of time, I would like feedback on whether you think this is going in the right direction