Jakarta EE TCK, 7 Jun 2023

Previous Meeting

Next Meeting

⏯

youtube image

►

From YouTube: TCK call#32

Description

June 7, 2023 Jakarta EE Platform TCK call #32.

Minutes can be viewed via https://docs.google.com/document/d/1V1dDLJkd14EDRMPeuI0VzPtU4Lbli8FFBd1pLDLlOrY/edit#bookmark=kix.boj22qlndjhi

A

B

C

A

B

Great okay um and um Scott um welcome Ed was just saying yeah as as we entered this in a minute. You know that um he contacted um uh you, you I'm gonna butcher. You you you're.

A

B

You're fine, thank you. You find about um what was conveyed on on the tck mailing list in terms of the repos and the output and we'll yeah. We'll expect to hear back- um and um um you know shortly in the next week or two about a follow-up- is.

A

There a change that needs to be applied still because I gave myself a date, but I don't see it here did I forget to save it or something.

A

So I guess that change got lost I'll make it again.

A

uh No later, then,.

D

D

Sounds good um on the next one, so.

E

What I have I'm a preliminary version of uh a couple of open, rewritable rules that are going to be the? This is again applying the basically the changes that Folly had done on the on the certain branch and it's working for a simple, simple test case for them when I applied for the whole Branch, it's it's missing a month, so I have to increase the test cases and see where it's it's falling down, um but it does look like it could be. You know possible, certainly not I mean a large chunk of it.

E

The big problem as I was mentioning was the fact that.

E

The the libraries, the libraries that we include in the test, artifacts, are kind of difficult to block out of um it's just the source code itself and even with the combination of trying to identify with the Legacy artifact bundle had included.

E

um You know there are some issues with.

E

How do we uh pull that out so that that's like the one of the things that I'll have to do more work on, but for now we're basically dropping in a stud that says utility class, so they need to add the libraries for this test, and so you know even more cases if you had to fill those in with many tests that has regard it back.

E

um That's there would be a lot a lot of boilerplate but a lot of hand. Coatings removed.

E

um I'll put the links to the the open relent rules and Families.

A

Hey Scott Stark, your audio is kind of muffled um microphones.

E

Okay, is that any better.

B

Yeah much better okay, I thought it was my hearing.

E

B

That isn't the best and.

E

Yeah mine, mine too, but yeah no I, I guess I was had the other microphone hooked up all right. So that should be better.

E

um Do you want me to repeat anything, or it looks like you got it yeah.

F

E

Case worst case, I think is better off, because at a minimum we would have like The The J units.

A

E

Point added annotations added in a stub for The Arcadian. It's just. We have some. You know hand coding work to do around the art of adding the any embedded Library artifacts, that a test might use.

B

And and and I I think, there's like in some other test um uh vehicles, I mean test test buckets um we see more like um like, like you know whether it's a Parent, Directory reference or you know three, you know four, you know Parent Directory references up. You know you know a lot of code sharing via you know other directory references. So so it's like Library references but there's like I think so. I think like code, it's just like shared with other test groups as.

E

Well, yeah, but those those are kind of straightforward because they CA they still come out of the um they still come out of the Legacy uh test artifact, and so we just when we generate the deployment method, we're just referencing the fully qualified name of the classes that are in there. So it kind of doesn't matter where they're located. Oh.

B

Oh great, okay, yeah.

E

Yeah, the only problem is when there's been a library that was actually in the Legacy, build created from ads. That's the problematic piece, because there's no code representation and even the contents of that we can get.

E

um Actually, there's the yeah there's a couple things I need to try um going deeper into the the Legacy archive and and kind of doing the same thing. Recreating the the library archive is one thing to try. We had tried uh one approach where we're basically just trying to use the archive, as is, but then it required having the Legacy tck available and that's not going to be practical. You know for them uh going forward. It's not practically a way to run a new tck to have the old tck bundle available.

B

What what what what like? What about the case when there's nowhere Killian involved like if we're just you know when we want to generate like just just a standalone tck, um which is a Subs? You know a subset, um um you know. Should we automate that as well and um I.

E

Mean well, the question is what how does it show up in the Legacy archive, because the only thing what's driving the generation of the deployment method is, if there's a an artifact that needs to be deployed um as part of the test.

E

You know so when, when I'm, using the your jar to shrink wrap utility, which it goes and looks for a given test package and sees that there's an uh an archive that needs to be deployed yeah to The Container, then then that's when it will assume that we need our Killian. If there's not such a thing, then basically it should have just converted it to a Java or a junit test and not include our Killian.

B

Yep that makes sense.

E

um Yeah so I mean we'll have like I said once once I get it running a little better, we're gonna have to go through and double check. You know uh some cases to make sure it's actually doing that correctly, but um so because there are, there are two rules: there's a rewrite rule for converting the the javadoc test, name to junit, test annotation and then there's a separate rule for adding the arcillian deployment method.

E

Based on whether or not there is a test artifact, so you know either they run um I think they run independently. Nice.

C

B

Yeah, that's that's yeah! That that'll be um great um because I think you know those are the the two main cases really.

E

Yeah I mean I, don't know if we're, if there's like still some uh configuration, that we need to further incorporate um that that's sitting out in property files or something like that that we'll run into. But you know those the book of the work is certainly those two things uh so hopefully, by the end of the week, I get I've got that in sufficient shape that people can take a look at it and start evaluating the Transformations nice.

B

um And I I was looking at I started. Looking at you know and um like try, you know trying um you know the other. You know the other um cases like you know. Just like you know, you know. Regular jars should be fine, but also ears and.

D

B

Want to add I want to add um you know the you know the ear support, definitely yeah.

E

Yeah, that's the other, that's actually, then, after this first phase is picking up the you know. If.

C

E

The non-wars basically we're just focusing on Wars at this point, yep.

B

um Which is which is I think sufficient for the you know for the stage, but you know um it doesn't hurt to add. You know to do the ears too, and.

E

B

Yeah, just you know, get it all going.

B

Nice and and any any any work that we do manually that we've been doing over the last year, or so you know still yeah, which all was the question at all yeah all he had. uh You know you know this. You know I think the output is, is gonna, be yeah, a a a whole request that you run and if, if there's already work, that's been done, you know you just don't train yeah, we just don't do that for the ones.

E

B

E

No, like I, said responded to all I mean the first thing is his work will certainly be a mechanism by which or validating the automated transformation. uh I I, don't think it's you know, like I, pointed out a case where you know, maybe that could be. There could be a semantic difference, because a class loading is obviously we see. Class loading is a a problematic aspect of uh you know EE and the tck, because he had moved classes from an embedded jar into the the war classes uh directory.

E

You know so you know, maybe one implementation would have a different behavior from that, but in general um you know those are there's only a few of those.

B

And one and one other thing, I wanted to point out that came in the conversation. Email conversation with um with Ollie is um or some other pull. You know you know, um issue um is the you know. The servletes tck is only is only for the standalones serve the tck. We still need to do the platform tck servlet test, um which.

A

B

Know he you know haven't been just haven't been done yet, so that's just the pendant work um just to be. You know just to mention that in case others weren't aware of that um I think the same will be true for.

C

B

For you know, as we do other stamp, you know Standalone tcks, like yeah, you know the first Standalone persistence tck. You know that will um um I I'm hoping that'll be something that the platform tck persistence tests can depend on.

B

um But um you know you know we'll yeah, we'll yeah we'll see how that goes, but um I think I think um we'll create these tcks as separate module, Maven modules, sub modules and um um where.

D

I I think where.

B

You know you know: I I, don't know if there's any, you know if there should be like one Maven module, yeah sub module for the platform. Tck are several for each different technology. You know under the platform tck but um um yeah I think that's yeah.

E

um Well, I mean that's, that's how we have to structure, uh uh that's how we're structuring the build anyway right, yep that the current the current tck refactor branch is one module per technology right, yeah.

B

E

Which is good, because that also that separates the dependencies better anyway. So.

E

um You know if you actually had separate artifacts um well, if you're releasing a bundle of separate Maven artifacts as the platform cck, um you can update. You know just the ones that have changes if uh they're Upstream they're Upstream dependencies change, which is a lot better than having to rebuild the whole thing.

B

Yep yep good point, and uh but yes, so I I think certain like, like the server technology on you know, under the platform tck would, you know, would would have both and pause. You know possibly the cert, the Standalone servlet tck Source would just move to the servlet. You know, spec, you know being an active spec um for you know not as active specs if they're not as active specs, the the spec teams, probably won't. It won't receive the source for their Standalone tck and that would probably stay in the platform dck.

B

But you never know yeah yeah, which ones will stay and which ones will go I. Think based yeah based on feedback. So far,.

B

And what I mean is yeah? We want we we do. We would like the tcks to be in the respective um spec repos, um so I would expect servlet to move. You know the Standalone server with spec tck, to move that you know, hopefully at some yeah at some point and that's on something yeah yeah. We have to discuss that, though yeah you know you know further, but.

C

B

Does anyone else have other? Yes, like the item uh for yeah, anyone else have updates on other tck refactoring. You know, maybe, on you know the the manual efforts that we've been doing in other areas.

B

They want to share.

D

uh uh Albany so I've been working on uh JSP uh basic case um like it's manually, but I was able to convert at least a few of the tests and do a quillion engine, um so probably by maybe within a week, I should be able to get a PR that converts all the JSP. uh This get tests into a quilling.

D

um Also we have the eltck refactored pr and already reviewed. uh Maybe I'll wait for some more time uh for any more comments before we merge it. Oh.

C

D

um Yeah, both of them are like manual effort. um Yeah uh eltc did not have to use any equivalent, it's only the J unit, uh but for JSP uh Pages tck. We are using aquiline and J unit uh also to note that uh I'm not using any uh vehicle implementation here. So basically, both of them include the Standalone vehicle implementation. Only uh until uh the vehicle uh related code is figured out so.

B

D

Both of them are like manual effort without any information.

B

um So so so JSP all yeah, also no arcillion. Is that what you said so.

D

No DSP has both the earthquake.

B

D

Only the unit- um okay, okay,.

F

F

Websocket uh refactoring PR uh is open. That includes both the platform and standard modificated that uses our cool name and there is no vehicle dependency for web socket and, apart from that, I'm working also on JTA 88 tck, that is working progress. The refactoring code is complete.

F

The testing is in progress with failures, which I am debugging post refactoring.

B

And which, which one did you just say,.

F

uh Websocket PR is open for review.

F

There are no open comments. All the review comments has been addressed. If any more review comments, I get I will address it as soon as I get uh JTA APK.

F

Like that, the code is complete, refactoring code is complete, testing is in program. There are failures.

F

During testing and I am debugging and fixing them.

C

B

um And just to mention it, you know you know again um here because I we mentioned it um before we're expecting to use ee10 implementations for validation are ee-10s. You know relevant specs for validation of tcks, for, for you know, for the development aspect of the of the tck work. Once we start add an ee 11 dependencies, then we're going to lose the ability to be able to validate our work and, and then it's yeah.

B

We've we've done that and yeah we've done that a few times now we kind of we kind of go into the into the dark side of um or um where we don't know if test failures are caused by bugs in the test or implementation.

B

Problems are um environmental problems of some sort and um yeah, so yeah yeah it'll it'll be good to um have that to make uh it more effect, make us more effective when we're validating you know the rewriting the rewriting and, at some point will be ready to add ee 11 tests and yeah that'll be great.

B

When that's going to be I, don't know I.

C

B

I think that's gonna yeah. That would be when yeah, that would be when we have the tcks Rewritten and pass and against ee-10 or the relevant specs.

B

So that'll be it'll, be great. Yeah that'll be a really great milestone for us to. You know to reach and uh it'll we'll have a big celebration for that.

C

B

um I, let's see if there's nothing else, I added um um a tck challenge, which we discussed in the platform quality yesterday and I just wanted to mention it here in case you know people in this call you.

C

Know you know weren't.

B

Aware of it or um have feedback or wanted to discuss it all and um I I think I mean my current thought on this particular challenge, which is basically we we have a ee8 implementation that is was previously compatible. They obviously made some change in their code base and there are now challenging a test that they've passed and lots of other people. Implementations have passed before and you know going back for so you know for early uh before ee8.

B

Maybe this ee6 or or so I I, I I, don't know I, don't remember, but um um the gist of it I think is the platform specification, maybe not be exact about calling out the requirements, um but there's you know yeah there there's um um basically I an ear with um that contains um a ejb and a war, and both the ejb jaw and the war contain the egb interface class and the current wording as I read it of the of the platform spec, which I I pasted in here is portable applications must not.

B

That must not depend on having or not have and um the egb classes in the you know, I I think if I re, if I understand, if I understand it correctly, having the egb classes in both the war and the ejb you know versus in the you know the share in the ear you know, air lib would be the another place where it could move to, but portable applications, you know, will I think would be able to work with the interface being in both places.

B

Is one interpretation of that and I think that's the interpretation that was previously used, but the request for the challenge is to make a call that follows the exact wording and I don't know it. Yeah we're gonna have to come up with a with a call on this, and um so just wanted to give you all a chance to chime in. If you wanted to and either on the issue on the platform mailing list, we can raise it on.

B

You know, I think ion's going to raise it on the servlet mailing list and I raised it on the egb mailing list already and um so yeah any feedback.

E

Well, it seems like it seems. Like you know, the norm for a war is war, first class loading. So as soon as you introduce ejbs that are external to the the web application, it seems like you're. Gonna have to have a configuration of the application server that allows for parent first delegation is, it has to. You, have to be able to take the interface. That's from the egb jar, as the type I mean.

E

What they're running into is that they have the type loaded in both the egb class, loader, egb, container class loader and the war and so they're having an injection failure.

E

um You know I, how are how did I I would seems like to me. We would have had to configure the class loading for this taste, so um there were the referencing that glass fish is is doing parent first delegation, but is that because they have uh a glass fish descriptor that they're, incorporating or I would find it unlikely that they would delegate to that model because that's not the norm for the web. The web container.

B

Yeah at all things I, don't think they I don't think they have.

E

I mean this is the deal so I mean this I guess this is the question I had asked them is I mean, do they have a way to configure parent first delegation on a on a warning, or you know they have a descriptor that allows for that, because I expect that that's what's the requirement of this is.

C

B

Up a recent copy of it a little test, but I, don't I, don't think there's any I mean this is probably probably is a probably is a let's.

C

B

uh Deployment descriptor in there, but I had to look to see what.

B

Here are some application.

E

Well, we have all the standard ones, but.

E

I mean in the past, we probably included glassware specific descriptors in here, because it was the reference implementation do. Do we not have those as Legacy anymore.

B

Well, we we we, we we have, we, we have not yet removed the glass fish. um um There were some implementation.

E

Deployment descriptors, yeah, Sun application.

E

But that's for security.

C

Just that's the.

E

E

That's the standard.

E

Now that's still security.

C

B

I mean I think last fish is, you know they probably didn't need to do anything special and I. Don't think I, don't yeah I, don't yeah I, don't think.

E

Well, I would certainly expect that there's some tests that require a uh the cluster that loads from the war. First.

B

um Let's see so I I guess I guess the key is um so the server the servlet is invoking the ejb and I.

C

Guess the problem, the.

B

Problem was on the call they're passing in uh when yeah yeah, the I think they said the wrong yeah, the the wrong. You know, the the other.

B

uh You know the other egb into interface um was was um um what was being referenced and um yeah. So so they got got a failure, instead of them both using the same consistent. You know right both the ejb and and the server using.

E

The same from from the exception, what it looked like was happening is that they were during the injection of the ejb interface on the servlet.

E

uh That was failing, uh presumably because two types they were trying to inject one type and the other types of referenced.

B

Yeah yeah, it's it's yes! Yes, certainly it seemed like that, um but yeah, so that so, but.

E

I mean uh so like I said that I would first ask them: do they have the ability to configure the war to use parent first delegation and then just add it I mean it's. That's that's an allowable change because class voting is, you know, non-trivial yeah.

B

I I didn't get it. I didn't get into that yeah that and into that they I mean they passed before they are. You know they are ee8 certified already and I. Don't know why they've made a change um but something I I. Don't you know I didn't get into the you know what the why they pat? Why did you pass before and and what what did you change? I mean they're, closed, Source, I, think I, don't think they're open source and yeah so yeah we could. You know we could certainly ask um um yeah.

E

I mean so, for example, wildflies certainly um do we have to add a a wildfire, specific descriptor to pass this test.

B

I didn't look, I, don't think so, though I think we.

E

What's our isn't our default Warfare still uh class loading, or do we do parents.

B

B

I don't know because we we changed, we Let's see, we um we made it I know we made a change um in the in the like three or four years ago, um but I. Think um like like.

B

Let's see if we just look at the is it is this a let's just want to see, is it a um remote? Is it a remote.

C

E

The check test interface- you know it's just uh a pojo, egb, okay, yeah I mean it could be I mean. What makes sense is that you know you could have war first class loading when you're, it's just a war, because that's all there is, but as soon as you're embedded in the air, you know then switch to parent first delegation or you know, but.

B

um In the so, in this case the um well the I mean you know, the ejb should be in its own. You know neither it's not the ejp, isn't in the top level deployment. Isn't it all yeah it's going to be a.

E

Yeah, it's an AGB jar yeah, but.

B

E

The only reason to include ejb's in a war is if that war itself could be deployed independent. You know independent that year I mean it seems like to me. This test is they've created. You know three container artifacts two words in an ear, presumably because those could be run separately and then there's also a test to bundle these in an ear to see how it behaves but I'm, not sure I.

B

So I I think I mean my my read of the uh oops. Let's see my read of you find the text, my read of what I I think the you know, the part about um I I, think I. Think the implementations have to be consistently you like in in an ear they need to consistently um use. You know you know, load from um the ejb class for sorry the ejb, the ejb, you know class loader, you know first um and um or you know, I I guess you know they can dip it.

B

You know they they can load. You know, you know they can depend on the war, but that's going to be. That would have to be both I would I would expect, and you know the ejb copy of the class to be. You know to be respected, um but I think it's I think I. Think the test has the interface in both because it's.

C

B

Loader specific thing of how an implementation wants to make.

E

Things work, that's the question, I mean what is what what um what's back and what? um What assertions is that the test making I.

B

Mean yeah I looked at those and um let me see um see the um the you know that the test is not trying to. um Let's see, I mean we're just going to see the um the assertion.

C

The assertion IDs here.

B

um I looked I looked at this at the assertion at I, looked up the assertions and, and neither of them.

E

Yes, security um I mean what they're, what they're trying to test is. Security has nothing to do with class loading.

B

Yes, exactly yeah, so I think that part so I think. Basically the test was written to be portable. You know independent of how the implementation um is doing its class loading and covering multiple just covering multiple, multiple possibilities to avoid class enough, not found errors if the implementation needed them in both like if you yeah.

E

B

E

You had, it certainly brings up the possibility that there's going to be a conflict too, if you have, if they're loaded in both that they're loaded in more than one class loader well,.

B

I think if it's loaded in both you'd have to be treating it as a remote call.

E

Yeah true or that, but it's definitely not.

B

um It's yeah, it seems like they're, you know, they're not, and it's it's typical to.

E

Know is there any way that this is a legacy that this used to be a remote call, and then it got changed, and so really it's not valid for it to be in the war I.

B

Don't have the Old Source, you know the pre-chakata source anymore, but um but I don't think so. I, don't think I. Don't think it would be unusual for that to for this to change.

B

That you know from from the I mean there's with millions of lines of test code. Who knows, but you know just going by memory. I I. Don't think that that you know that that tests like these were ever you.

E

B

Were you know they think they were ridden and then they stayed the way they were and.

E

Look at the build at XML. It's definitely meaning to include the egb interfaces in both and all three yeah.

B

E

um The war yeah okay.

A

um So Scott Marlo and Stark I want to ask um I appreciate the detailed thought experiment. Discussion you've been having and I recognize that someone on the servlet team, who would need to reason about this problem and potentially make a decision or, as we talked about be informed, might benefit from having some kind of brief description of the conclusion that you two have come to after this discussion.

A

uh Do you think that's possible worthwhile to capture in some text, so we can make sure.

E

That yeah I think that should be added. The issue is yeah well, I'll, I'll follow up, I mean basically I want to ask: can they configure their their server to behave? Obviously, the way this expects, and two uh you know, certainly give the you know what what the situation is, so that because it's going to be up I, think more than likely. This still is going to fall back on the platform, because it's going to depend on the class loading behavior of an ear where we have duplicate classes.

E

So um you know I. We can have the serverless group ping us on whether or not they believe that's the case or whether or not there's specific servlet uh text that covers this, but um in other words, I think we're gonna have to yeah go back to the platform group, but yeah I'll I'll. Take that as action item so.

A

That goes into the issue and I agree. That's the best place to put it I just want to make sure this useful and.

B

I I will upload the um um the you know the recording um Ed you kind of froze for a second while.

A

You're yeah I started.

C

B

Because I, because you seem to be you know you paused, but right.

A

I just want to make sure that's useful and very difficult to elicit understanding super deep context. Understanding is encoded somewhere in a reusable form.

B

A

B

And before I put the mic down on it, I'm just I'm just going to mention quickly just for this recording um that I I think that the the question that you mentioned Scott back to the Implement there um I I, think that you know I see that as a um um you know, as also um considering application, portability will applications that work on other ee. Eight implementations also work on their implementations and when people write specifically to their implementation, will those be portable to others and that's yeah I I that yeah.

B

That would be the you know to me. That would be the you know. A reason to not. You know, exclude this particular test or considered it. You know it excluded. Yeah,.

A

E

A

Exclude unless we must.

E

Yeah I, don't again I, don't really see that this is trying to test um well, because, by definition, this is going to require application specific descriptors anyway, because it's testing security, so you're gonna have to do a mapping of you know, because security is out always has required that there's, no certainly an application server can take and like have a default Behavior where it takes um the the security roles, and this have like kind of a a pattern for transforming that into an actual mapping, but in general, security always requires a separate application.

E

Server-Specific mapping file, which we see here for the sun. You know the Legacy sun, which I guess it's the origin of glassfish right. Yes,.

B

It's yeah yeah, definitely and it still works with glass, fish and.

E

B

It's easy to remap to other implementations and I. Think that's typical from conversations with others and you know yeah, we've done you know, we've done. You know that as well, so yeah.

E

um Yeah I mean I I. Don't expect that this test is of the form that will this run without any application server specific descriptors, oh.

B

Yeah, it definitely needs some descriptors for sure yep. Definitely yep yeah.

E

And certainly configuring class loading behavior is is allowable, but um yeah I'm, not sure why?

E

For a local interface I'm, not sure why you would include the egbs in the war in the first place, I'm, not sure what the history of that was.

E

Because it seems like to me, we went through this exercise of getting rid of remote at one point, and so I still would expect that historically, these were because, frankly, there was a point where there wasn't a local interface right.

E

Remote interfaces at one point, yeah.

B

E

And they're in they're in the wars, for you know that historical artifact and it's you know it's probably not correct here, but I.

B

I do remember challenging some other tests to remove the duplicate duplicate interface.

E

Yeah, it seems like to me I've I've had I recollect. We had issues with where we were going to do remote, removal and um yeah. There was the dude, you know you don't want duplicate classes in a year for sure in general, you know, unless that's unless there is a serialization of the invocations yeah.

B

Yeah definitely- and it's not it's it I looked at all the assertion IDs and none of them uh from the spec. um um You know sections that we, you know that talk about you know, ear um like egb classes in in ears and in in use in servlets and yeah yeah, which I.

C

A

B

Some of which I pasted here, okay, um so the yeah, it's not yeah, that's not assertion related. So all right, so um yeah, great um I'll ping, you later Scott um and I'll, take a look to see what we you know if we have anything in wildfly specific for that test. Just you know just.

E

So you know yeah that'd be good, because I mean uh you know and because, in the old days when I was still heavily involved in this piece that uh you we would have needed that because we used to default to the class loading added classes out of the Warfare. So you know, if you say, we've changed that about that. Would obviate the need for them.

B

Yep great point: okay,.

E

All right, we'll follow up on the issue, then.

C

B

Right, that's anything else from others.

E

C

B

Let me stop uh sharing and stop.