Skip to end of metadata
Go to start of metadata
Jan 19 18:59:06 jmcconnell	this is a pretty big change i think, opens up a lot of potential for great things...and abuse, rather like profiles in that respect
Jan 19 18:59:31 jmcconnell	people have been asking for a global context for plugin communication for a long time
Jan 19 19:00:20 jcasey	bporter: I'm not sure I know what you mean about the project as a way to centralize the info between plugins, other than using properties and static config that's manually edited
Jan 19 19:01:47 jmcconnell	jcasey: basically you just have a mechanism for one plugin to say X=FOO and another plugin to get that variable state, right?
Jan 19 19:02:02 jcasey	right
Jan 19 19:02:17 jmcconnell	ya, thats hugely powerful
Jan 19 19:03:01 jmcconnell	the crux is that up til now the only way to do that was through external methods like files
Jan 19 19:03:29 jmcconnell	and it makes plugins able to be dependent on other plugins
Jan 19 19:03:42 jmcconnell	_wholey_ dependent
Jan 19 19:04:14 jmcconnell	personally I love it, its just dangerous :)
Jan 19 19:04:35 jcasey	right, dependency is possible...but it's possible using files too
Jan 19 19:04:53 jcasey	it's just info passing, and you could even do it using sockets...
Jan 19 19:05:07 jcasey	you can't keep people from screwing themselves up if they're determined to do so
Jan 19 19:05:16 jmcconnell	right, like profiles :)
Jan 19 19:05:26 jcasey	well...yeah
Jan 19 19:08:17 jmcconnell	the only thing I worry about is how it could poliferate into plugins in the wild
Jan 19 19:08:34 jmcconnell	where plugin X needs plugin Y to determine some state
Jan 19 19:09:10 jmcconnell	imo there ought to be some kinda registration of that dependency is this is the case, so error reporting can be useful
Jan 19 19:09:38 jmcconnell	know what I mean?
Jan 19 19:10:40 jcasey	yeah, I think the registration would be in the form of a <dependency/>...
Jan 19 19:11:06 jcasey	the way I've been using it is as a DTO of sorts that can serialize/deserialize primitive info to/from the container context
Jan 19 19:11:36 jcasey	you have to have the DTO in your classpath, so if you use it as an extension, or have one plugin extend the other where the DTO exists, it works
Jan 19 19:12:19 jmcconnell	DTO? sorry, acronym challenged atm
Jan 19 19:12:46 jcasey	data-transfer object
Jan 19 19:13:03 jcasey	this has always been possible, using ${session} and the core container context, BTW
Jan 19 19:13:12 jcasey	I just made it a little more formal
Jan 19 19:13:22 jmcconnell	ah
Jan 19 19:13:26 jcasey	I've been doing this for quite some time now
Jan 19 19:14:08 jcasey	and I've just been careful to make plugins resilient to nulls for this stuff
Jan 19 19:14:09 jmcconnell	do you think <dependency> is enough
Jan 19 19:14:41 jcasey	for the object, I think so...but if you don't formalize it into a DTO, then you have to be resilient to possible nulls coming out of the context
Jan 19 19:15:03 jcasey	you don't have to have inter-plugin deps with this, it's not a foregone conclusion
Jan 19 19:15:17 jmcconnell	I would almost be in favor of something like ${globalContext.maven-foo-plugin.value} so it could complain about the plugin not having the said value
Jan 19 19:16:01 jmcconnell	default-value=${that thing}
Jan 19 19:16:14 jcasey	we could do that too, but remember that the value in context doesn't have to be tied to a plugin
Jan 19 19:16:25 jcasey	in fact, many plugins could be used to inject the same value, or read it
Jan 19 19:16:56 jcasey	and, the context throws an exception in case the value isn't there, so it forces you to think about it
Jan 19 19:17:11 jmcconnell	sure they could, but to make a relationship clear and concise you could argue against allowing that
Jan 19 19:17:40 jcasey	sure, but why force the plugin-plugin relationship if it's not a 1:1 mapping?
Jan 19 19:17:55 jcasey	if two plugins could put it on the wire, how would that expression look?
Jan 19 19:18:05 jcasey	I suppose ${globalContext.value}
Jan 19 19:18:06 *	jmcconnell is just thinking of ways to limit your new found power and enforcing accountability to a value going missing where one is required
Jan 19 19:18:40 jcasey	IMO, it's a good practice to inject it via plugin parameter
Jan 19 19:18:45 jcasey	then, you can use @required
Jan 19 19:18:49 jmcconnell	accountability in such a way that maven itself can complain and let the user know what plugin failed on the chain that getting built up now
Jan 19 19:18:54 jcasey	right
Jan 19 19:19:05 jcasey	well, sort of...
Jan 19 19:19:16 jcasey	that's part of what makes this hard
Jan 19 19:19:20 jmcconnell	cause thats really what this is...you are allowing a chain of plugins to execute now
Jan 19 19:19:31 jmcconnell	thta carry a context with them
Jan 19 19:19:31 jcasey	sure, kind of
Jan 19 19:19:54 jcasey	I can also read System properties out of the context without using System.getproperties (which is not embeddable,really), and without passing them around explicitly
Jan 19 19:20:08 jcasey	that's more powerful than just inter-plugin comm, IMO
Jan 19 19:20:13 jmcconnell	and if god forbid you have a chain of 10 plugins in a state map...it needs to be able to pin the one that failed and how
Jan 19 19:20:32 jcasey	hopefully, if it really does fail, it should kill the build...that's what the exceptions are for
Jan 19 19:20:35 jmcconnell	sure, but no one allows that kinda usage of system
Jan 19 19:20:52 jcasey	not sure what you mean
Jan 19 19:20:55 jmcconnell	you would be beaten upside the head if you tried it...or I would if I did at least :)
Jan 19 19:21:02 jmcconnell	using System for your context
Jan 19 19:21:02 jcasey	:)
Jan 19 19:21:07 jcasey	no, not what I mean
Jan 19 19:21:10 jcasey	the opposite
Jan 19 19:21:24 jcasey	pulling the sysprops, and storing them in a build-level context for reference
Jan 19 19:21:38 jcasey	without reading from System.getProperties() like we are in core in some places now
Jan 19 19:21:43 jcasey	for example, in profile activation
Jan 19 19:22:03 jcasey	Milos has worked around that, but it's passing the properties from hand to hand that don't need them...
Jan 19 19:22:17 jcasey	the project builder, and the profile manager both have to know about them, needlessly
Jan 19 19:22:57 jmcconnell	ok, implementation can be countless ways...but the part that concerns me is the declaration of the dependency
Jan 19 19:23:49 jmcconnell	its not so much a <dependency> of the second plugin on the first, which would imply class file requirements, but its really a <stateDependency>
Jan 19 19:24:13 jmcconnell	which is all I am concerned about having clear
Jan 19 19:24:15 jcasey	I'm starting to think this debate would be better on dev@...but I agree that the globalContext thing works for accounting, though I'm not sure it would help tell you where the value should come from...
Jan 19 19:24:31 jcasey	what if it's not a plugin at all, but an extension?
Jan 19 19:24:36 jmcconnell	so when it really fails it can be reported in a way that you have a hope of figuring it out
Jan 19 19:24:59 jcasey	yeah, a state dep of sorts makes some sense
Jan 19 19:25:28 jmcconnell	and ya, I think this is probably what brett wanted talked about on dev@
Jan 19 19:25:29 jcasey	like "something was supposed to set the global state: systemproperties" ?
Jan 19 19:25:36 jmcconnell	yep
Jan 19 19:25:49 jcasey	I'm not sure how that would help, unless the plugins doc what they set (which they should do, I suppose)
Jan 19 19:26:03 jmcconnell	so if I fail I can blame maven-sparky-plugin for putting me in this unrecoverable state
Jan 19 19:26:12 jcasey	would be cool if we could provide a mechanism for setting state that would track it somehow
Jan 19 19:26:17 jcasey	hehe
Jan 19 19:26:33 jmcconnell	${globalContext.maven-sparky-plugin.value} would at least tell you the plugin and value
Jan 19 19:26:49 jcasey	right, this is the discussion we should be having. I just wanted to make sure it was feasible before firing off an abstract email
Jan 19 19:26:50 jmcconnell	hm..
Jan 19 19:27:02 jmcconnell	sure thing, I love the idea frankly
Jan 19 19:27:21 jmcconnell	its been the topic of many mails I have read in the last year+
Jan 19 19:28:00 jcasey	yup
Jan 19 19:28:06 jmcconnell	a plugin listener could do it maybe
Jan 19 19:28:09 jcasey	I'll draft something and send it off tonight
Jan 19 19:28:14 jmcconnell	cool
Jan 19 19:28:27 jmcconnell	c&p from here if you want, you have my permission :P
Jan 19 19:28:52 jcasey	I was thinking about a new method, like export(Object key, Object value) on AbstractMojo, but that would still leave open the possibility of rogue values like I used to do
  • No labels