merge gdcc dataverse jsf tests into upstream · dev

Stream: dev

Topic: merge gdcc dataverse jsf tests into upstream

Ash Manda (Oct 05 2026 at 19:43):

@Don Sizemore, @Akio Sone, @William Hahn, @John D. Martin III, and the rest of the UNC team has given the go-ahead on merging the full https://github.com/gdcc/dataverse-jsf-tests repository into upstream.

UNC will be moving into a lower-overhead model depending heavily on upstream validation here on out.

The proposed model is a phased deployment. gdcc/dataverse-jsf-tests and uncch-rdmc/dataverse-jsf-tests will be immediately put into read-only mode. The tests will be merged into upstream, and the repositories will remain as such for a few months or so, while we at UNC figure out how we can pull upstream tests for our own use cases. Thank you!

@Philip Durbin 🚀

Philip Durbin 🚀 (Oct 05 2026 at 19:45):

Sounds great! You'll be making a PR?

Ash Manda (Oct 05 2026 at 19:46):

There's some technical debt that I want to address first, but yes I'll be making the PR!

Ash Manda (Oct 05 2026 at 19:46):

do you have a timeline on when upstream wants it merged?

Philip Durbin 🚀 (Oct 05 2026 at 19:47):

yesterday :smile:

Ash Manda (Oct 05 2026 at 20:02):

Which directory within the upstream repo?

Ash Manda (Oct 05 2026 at 20:03):

tests/dataverse-jsf-tests ? or tests/playwright ?

Philip Durbin 🚀 (Oct 05 2026 at 20:04):

Hmm, tests/jsf?

Shorter, sweeter. :smile:

Ash Manda (Oct 05 2026 at 20:04):

SGTM

Philip Durbin 🚀 (Oct 05 2026 at 20:05):

@Oliver Bertuch unless you want them under src/test somewhere?

Ash Manda (Oct 05 2026 at 20:10):

https://github.com/IQSS/dataverse/issues/12797

Ash Manda (Oct 05 2026 at 20:10):

https://github.com/IQSS/dataverse/pull/12798

Ash Manda (Oct 05 2026 at 20:11):

This ports the repository as is

Philip Durbin 🚀 (Oct 05 2026 at 20:15):

Hmm, can we remove references to UNC?

Ash Manda (Oct 05 2026 at 20:16):

I was planning to do that post-merge, should I do it now? It'll be a while if I get to cleaning it

Ash Manda (Oct 05 2026 at 20:17):

these tests are currently being used by CI/CD, so its a bit risky to try and edit or clean up everything

Philip Durbin 🚀 (Oct 05 2026 at 20:17):

Hmm, can you at least add a note at the top of tests/jsf/README.md? TODO: Clean up these docs, blah blah or something?

Ash Manda (Oct 05 2026 at 20:17):

for sure

Philip Durbin 🚀 (Oct 05 2026 at 20:18):

awesome, thanks

Philip Durbin 🚀 (Oct 05 2026 at 20:18):

and thanks for the PR!

Ash Manda (Oct 05 2026 at 20:18):

You have @Don Sizemore to thank. He was one of the strongest advocates from the very beginning

Philip Durbin 🚀 (Oct 05 2026 at 20:19):

it's like a shiny new toy

Philip Durbin 🚀 (Oct 05 2026 at 20:19):

Wait, what about the GitHub Action? It's still pointing at the old repo, now archived, you said.

Ash Manda (Oct 05 2026 at 20:20):

It will continue to point at the old repo for a while

Ash Manda (Oct 05 2026 at 20:20):

Until I address the technical debt and remove the UNC references

Philip Durbin 🚀 (Oct 05 2026 at 20:21):

so this is more of a dump, a snapshot

Ash Manda (Oct 05 2026 at 20:21):

Exactly!

Philip Durbin 🚀 (Oct 05 2026 at 20:21):

not a living thing

Philip Durbin 🚀 (Oct 05 2026 at 20:21):

dead code

Philip Durbin 🚀 (Oct 05 2026 at 20:21):

doesn't execute

Ash Manda (Oct 05 2026 at 20:21):

Yep

Ash Manda (Oct 05 2026 at 20:21):

For now

Philip Durbin 🚀 (Oct 05 2026 at 20:21):

seems a bit suboptimal :sweat_smile:

Ash Manda (Oct 05 2026 at 20:21):

It is

Ash Manda (Oct 05 2026 at 20:21):

But if it is used in its current state and y'all start adding tests

Ash Manda (Oct 05 2026 at 20:22):

it will be unusable in a few weeks

Ash Manda (Oct 05 2026 at 20:22):

There's a lot of hardcoding done because I was developing it so quickly

Ash Manda (Oct 05 2026 at 20:23):

Give me a few weeks (two or three) and I will make it fully extensible and usable for everyone

Philip Durbin 🚀 (Oct 05 2026 at 20:24):

Sure, please take your time. Should I assign the PR to you? And put it in "in progress" on the board? Is that the right state?

Ash Manda (Oct 05 2026 at 20:25):

Yep. I can also make it a draft PR if that helps

Ash Manda (Oct 05 2026 at 20:30):

Philip Durbin 🚀 said:

Hmm, can you at least add a note at the top of tests/jsf/README.md? TODO: Clean up these docs, blah blah or something?

Added these to the top

Philip Durbin 🚀 (Oct 05 2026 at 20:37):

Thanks and thanks for making it draft. I added it to "in progress" and assigned it to you.

For me, part of the definition of done is that the code not be dead. It should be executed by CI.

Philip Durbin 🚀 (Oct 05 2026 at 20:37):

Mostly I'm just glad we're getting the tests upstream! :tada:

Ash Manda (Oct 05 2026 at 20:39):

This PR has to be merged before the workflows can repoint to it anyways

Ash Manda (Oct 05 2026 at 20:39):

we could take them live immediately after the merge

Philip Durbin 🚀 (Oct 05 2026 at 20:39):

Huh. I imagined it would be one PR.

Ash Manda (Oct 05 2026 at 20:40):

That would cause conflicts in this: https://github.com/IQSS/dataverse/pull/12783

Ash Manda (Oct 05 2026 at 20:41):

I'm trying to go as slow as possible so that nobody's workflow is disrupted and tests remain green throughout

Ash Manda (Oct 05 2026 at 20:44):

By phased deployment my general idea was:
Phase 1: Get a dead dump/snapshot into upstream
Phase 2: Wait for 12783 to pass review
Phase 3: Clean up UNC references, technical debt within upstream
Phase 4: Activate the tests/jsf folder
Phase 5: Re-jig this pipeline for UNC's use

Philip Durbin 🚀 (Oct 05 2026 at 20:45):

I think I missed the part about phased development.

Ash Manda (Oct 05 2026 at 20:47):

I would advocate for this as it keeps disruption to a very minimum

Philip Durbin 🚀 (Oct 05 2026 at 20:47):

#12783 has merge conflicts. I put it in "ready for triage" earlier today. Is that the right status?

Ash Manda (Oct 05 2026 at 20:48):

It's in the making releases.rst file, I might git rm it tomorrow?

Ash Manda (Oct 05 2026 at 20:48):

thank you so much for working with me on this!

Philip Durbin 🚀 (Oct 05 2026 at 20:56):

sure!

Philip Durbin 🚀 (Oct 05 2026 at 20:56):

I was thinking you'd merge the latest from develop but whatever works.

Ash Manda (Oct 05 2026 at 20:56):

Yep, I did exactly that

Philip Durbin 🚀 (Oct 05 2026 at 20:57):

great

Ash Manda (Oct 07 2026 at 12:59):

I went ahead and closed the draft PR. I liked your idea more Phil, it's supposed to be one full PR where the code is both added and activated. In fact, this might actually be a really good excuse for me to try out: https://docs.github.com/en/pull-requests/get-started/about-stacked-prs

My ETA is about 3 weeks

Philip Durbin 🚀 (Oct 07 2026 at 13:16):

sounds fine, thanks!

Oliver Bertuch (Oct 15 2026 at 08:18):

Promise not to hate me. Here's a crazy thought. What if we'd create a Maven Submodule (as mentioned for API tests, too) and use a JUnit test to trigger running the Playwright tests? (Not asking to transform the tests from Node to Java here, just make them more accessible :wink: ) This might be a pipe dream around using https://github.com/eirslett/frontend-maven-plugin

Oliver Bertuch (Oct 15 2026 at 10:26):

I just left a lengthy review on #12783. Good job @Ash Manda! :heart: I do still have some more things for you... :grimacing: :see_no_evil:

Ash Manda (Oct 15 2026 at 16:13):

It will take a long time to address some of those review comments

Ash Manda (Oct 15 2026 at 16:14):

We cannot run JSF and API tests in parallel. Runners limits for free orgs is 20, we'd increase the number of parallel runners to 4. Dataverse frontend would suffer

Ash Manda (Oct 15 2026 at 16:16):

Coveralls also doesn't support piecemeal XML reports. So we need to fold in your unit tests action into our workflow and generate a combined report.

Ash Manda (Oct 15 2026 at 16:18):

There is a reason why I made the integration tests run three times, which is flaky tests. A lot of the time seeing an integration tests job fail, simply re-running it would get it to pass. If they run three times in parallel, say one passes and two fails, we know on the spot which test is flaky. That's a whole 40min re-run saved for developers. If all three fail, we know for sure that its not a flaky test

Ash Manda (Oct 15 2026 at 16:22):

With GHCR uploads/downloads we are adding another point of failure into our workflow. GitHub artifact upload/download for our purposes here is "free". It costs 10 seconds up and down and it gives devs an easy download if they want to replicate the exact container setup.

Ash Manda (Oct 15 2026 at 16:22):

Thank you for the detailed review! I'll share my thoughts there too

Oliver Bertuch (Oct 15 2026 at 17:06):

Anyone up for a campfire and a fireside chat in a cozy Zoom room to talk about this?

Ash Manda (Oct 15 2026 at 19:25):

Raw notes from today's meeting. Very raw.

**Push to GHCR if secrets are available AFTER all the tests pass. Multi-arch image.
container_base_push.yml => we want to look Into ARM runners
We need to look into CodeQL parallelized runners running them sequentially
Phil, Oliver, Ash agree on unnecessary runners. Cut them down. Especially CodeQL
We want to decouple the API and JSF tests, before that CodeQL needs to go away. It is possibly replaced by SonarQube
**R1: Integration Tests

R2: Chrome -> R2: Firefox

-> R3: Webkit

All end at the same time
pom.xml <JaCoCo>
Take a JaCoCo report from each and merge them together:
-> Unit Tests
-> Integration Tests
-> API Tests
-> JSF Tests Chrome
-> JSF Tests Firefox

-> JSF Tests Webkit
maven_unit_test.yml does codecoverage
Code Coverage is out of scope for the current PR
Instead of mangling the JaCoCo agent, add a script to the base container to attach the JaCoCo agent.

Container base / modules /docker script / startinForeground.sh
compose.yml
Add more agents: Java Agents setting
Tell maven to fold the jacoco cli and agent images into the build file (maybe you don’t need the cli)
This is part of the scope for this PR
Maven Cache Management: Figure out NPM Caching
Playwright, Chromium, Firefox, and Webkit => Once they’re merged into upstream.
NPM Cache
Future Optimization: Figure out how to use the maven cache more to reduce per-branch maven. Then look into npm caches
mvn verify by default runs mvn package BUT BUT BUT there are skip flags present to skip the packaging steps.

Don Sizemore (Oct 15 2026 at 20:01):

@Ash Manda you're supposed to be on vacation!

Philip Durbin 🚀 (Oct 15 2026 at 20:06):

That's what I said!! :smile:

Ash Manda (Oct 15 2026 at 22:33):

I am always on vacation! This is fun!

Ash Manda (Oct 15 2026 at 22:35):

I addressed almost all of the concerns we had in today's meeting except for caching. Which we can't do anyways because the JSF tests aren't in the main repo

Ash Manda (Oct 15 2026 at 22:36):

The way I'm planning to handle it is that I am also rewriting the JSF tests repository from TypeScript into Java. By the time we're ready to bring it in, it should be able to fold directly into upstream

Philip Durbin 🚀 (Oct 15 2026 at 23:40):

Ah, I was wondering about that. Backend devs would probably prefer Java anyway. :smile:

Ash Manda (Oct 16 2026 at 01:09):

Tests pass successfully in Java: https://github.com/uncch-rdmc/dataverse/actions/runs/38826031860/job/116733594566?pr=108

Ash Manda (Oct 16 2026 at 16:00):

Okay so i have a problem

Ash Manda (Oct 16 2026 at 16:00):

I translated the entire suite into Java yes. But it still uses the microsoft playwright library

Oliver Bertuch (Oct 16 2026 at 16:01):

WOWzers! That must have been a lot of work!

Ash Manda (Oct 16 2026 at 16:01):

This library is known for "strongly" preferring Ubuntu. I don't know if it'll break anything if its added to the project. Its known for not playing very well on Fedora/RHEL

Ash Manda (Oct 16 2026 at 16:02):

The project is live at: https://github.com/gdcc/dataverse-jsf-tests/tree/java

Oliver Bertuch (Oct 16 2026 at 16:02):

If we put the tests into a submodule, most people probably will never get in touch with them. Maybe that's fine then?

Ash Manda (Oct 16 2026 at 16:02):

I was wondering if someone can help me understand the blast radius of folding this into the upstream repo

Oliver Bertuch (Oct 16 2026 at 16:07):

Sure. What would you need?

Ash Manda (Oct 16 2026 at 16:08):

going through the pom.xml and seeing if any libraries would cause problems for developers?

Ash Manda (Oct 16 2026 at 16:08):

playwright is definitely one

Ash Manda (Oct 16 2026 at 16:08):

The whole point of this exercise is to let developers run tests freely right

Ash Manda (Oct 16 2026 at 16:09):

If that library is the only one, I might try switching to Playwright in Containers

Oliver Bertuch (Oct 16 2026 at 16:09):

I can take a look. But as I said, by making it a submodule, this is a part of upstream, but not mixed into the Dataverse source tree.

Oliver Bertuch (Oct 16 2026 at 16:09):

That makes it alot less risky

Ash Manda (Oct 16 2026 at 16:09):

I thought we wanted to make it part of the source tree?

Ash Manda (Oct 16 2026 at 16:10):

In any case, thank you so much!

Oliver Bertuch (Oct 16 2026 at 16:11):

I would advise against that. Please note that the two of us never talked about this here before, so my opinion might be different to what @Philip Durbin 🚀 said / wants / needs. :see_no_evil:

Philip Durbin 🚀 (Oct 16 2026 at 16:11):

I just want the JSF tests in the main "dataverse" git repo so that tests are versioned alongside the code.

Ash Manda (Oct 16 2026 at 16:11):

submodule ain't gonna do that right?

Oliver Bertuch (Oct 16 2026 at 16:12):

My advice would be to put the code under modules/dataverse-jsf and make it a submodule to the parent as well as using the parent in the module.

Oliver Bertuch (Oct 16 2026 at 16:12):

Ah, I think I know where the confusion comes from!

Oliver Bertuch (Oct 16 2026 at 16:13):

I'm talking about a _Maven_ submodule, not a _Git_ submodule!

Philip Durbin 🚀 (Oct 16 2026 at 16:13):

Maybe just modules/jsf?

Or modules/jsf-tests?

Philip Durbin 🚀 (Oct 16 2026 at 16:13):

Or under tests/jsf. I thought that was the original plan. :smile:

Oliver Bertuch (Oct 16 2026 at 16:13):

Again, I'd advice against that. Looking at a lot of other Java projects, if you have more tests for another module, they mostly use the module name and add a suffix

Ash Manda (Oct 16 2026 at 16:14):

I still have to give developers a way to run these tests locally on RHEL

Philip Durbin 🚀 (Oct 16 2026 at 16:14):

I'm not sure what we're talking about. Getting lost.

Ash Manda (Oct 16 2026 at 16:14):

my bad.

Little bit of context

The playwright tests are now in Java. Which is good

Ash Manda (Oct 16 2026 at 16:15):

But playwright doesn't work well on RHEL. Only on Ubuntu. The maven package

Philip Durbin 🚀 (Oct 16 2026 at 16:15):

right, moving from JS/TS to Java was a big change

Ash Manda (Oct 16 2026 at 16:15):

Most of our developers use RHEL afaik

Philip Durbin 🚀 (Oct 16 2026 at 16:15):

who uses RHEL?

Oliver Bertuch (Oct 16 2026 at 16:15):

Not for development. And I'm not most :sweat_smile:

Philip Durbin 🚀 (Oct 16 2026 at 16:15):

The containers in our dev env are Ubuntu.

Ash Manda (Oct 16 2026 at 16:15):

Don't they? I thought RHEL was the recommended dev platform?

Oliver Bertuch (Oct 16 2026 at 16:16):

No, deployment platform

Philip Durbin 🚀 (Oct 16 2026 at 16:16):

yes, RHEL for prod

Oliver Bertuch (Oct 16 2026 at 16:16):

IQSS AFAIK uses Macs exclusively, I'm on Fedora.

Philip Durbin 🚀 (Oct 16 2026 at 16:16):

right

Oliver Bertuch (Oct 16 2026 at 16:16):

Not sure about the other core team members Vera and Eryk (not tagging them for now)

Oliver Bertuch (Oct 16 2026 at 16:17):

I _think_ Jim uses Mac as well?

Ash Manda (Oct 16 2026 at 16:17):

I'd still feel better if someone more knowledgable used AI to see if any of these modules will cause an issue: https://github.com/gdcc/dataverse-jsf-tests/blob/java/pom.xml

Philip Durbin 🚀 (Oct 16 2026 at 16:17):

Eryk is on Linux. Jim is on Windows. Not sure about Vera.

Ash Manda (Oct 16 2026 at 16:17):

I don't see anything other than Playwright

Oliver Bertuch (Oct 16 2026 at 16:17):

/me checks out the code now

Ash Manda (Oct 16 2026 at 16:18):

we probably want to run these tests on a deployment platform too

Philip Durbin 🚀 (Oct 16 2026 at 16:18):

As long as the tests are not tied into the core code in any way (via Maven), they won't hurt anything in the core code.

Ash Manda (Oct 16 2026 at 16:18):

Say a QA or Test server with an RC

Philip Durbin 🚀 (Oct 16 2026 at 16:19):

We already have a couple little Maven projects in the core code that aren't tied in.

Philip Durbin 🚀 (Oct 16 2026 at 16:19):

(I think @Oliver Bertuch has argued that they should be tied in.) :sweat_smile:

Ash Manda (Oct 16 2026 at 16:19):

once again thank you so much for taking time to help me through this!

Philip Durbin 🚀 (Oct 16 2026 at 16:19):

oh sure

Philip Durbin 🚀 (Oct 16 2026 at 16:20):

Instead of <groupId>edu.harvard.iq</groupId> should we use org.dataverse? :thinking:

Oliver Bertuch (Oct 16 2026 at 16:21):

I agree we should use profiles per browser. The profile should keep the execution config, making the "skip" flags obsolete.

Philip Durbin 🚀 (Oct 16 2026 at 16:23):

@Oliver Bertuch is your vision for this that devs run something like mvn test:jsf from the root of the repo?

Or will they cd tests/jsf first and run whatever mvn command?

Oliver Bertuch (Oct 16 2026 at 16:24):

Either cd modules/dataverse-jsf; mvn verify ... or mvn -f modules/dataverse-jsf verify ...

Oliver Bertuch (Oct 16 2026 at 16:24):

Normal Maven stuff

Oliver Bertuch (Oct 16 2026 at 16:24):

Just like any other submodule :sweat_smile:

Ash Manda (Oct 16 2026 at 16:26):

If you have any reviews for that repo, they're always welcome! It's best to make changes right now

Philip Durbin 🚀 (Oct 16 2026 at 16:27):

Ok, so a maven submodule. Tied in. Not a separate Maven project on the side that happens to be in the same git repo.

Philip Durbin 🚀 (Oct 16 2026 at 16:28):

Why call it "dataverse-jsf" and not just "jsf-tests"? Why is "dataverse" in there? :thinking:

Ash Manda (Oct 16 2026 at 16:28):

I wanted to separate it from the other JSF tests that I work on specifically for UNC.

Oliver Bertuch (Oct 16 2026 at 16:28):

Philip Durbin 🚀 said:

Why call it "dataverse-jsf" and not just "jsf-tests"? Why is "dataverse" in there? :thinking:

:point_down:

Oliver Bertuch said:

Again, I'd advice against that. Looking at a lot of other Java projects, if you have more tests for another module, they mostly use the module name and add a suffix

Philip Durbin 🚀 (Oct 16 2026 at 16:31):

Is this where we define the module name as "dataverse"? https://github.com/IQSS/dataverse/blob/v6.12.1/pom.xml#L17

Oliver Bertuch (Oct 16 2026 at 16:32):

Aye!

Philip Durbin 🚀 (Oct 16 2026 at 16:32):

ok, thanks

Oliver Bertuch (Oct 16 2026 at 16:33):

It would be so nice to have a proper parent/child layout in our repo... :see_no_evil: Maybe in DV 10?

Philip Durbin 🚀 (Oct 16 2026 at 16:33):

ha, maybe

Philip Durbin 🚀 (Oct 16 2026 at 16:34):

Can we change it from edu.harvard.iq to org.dataverse then too?!?

Ash Manda (Oct 16 2026 at 16:34):

sure, I have no problem

Oliver Bertuch (Oct 16 2026 at 16:34):

No one would stop us. We don't have to change the package folders when relocating the groupId, but we should to be aligned with good practices of Maven.

Philip Durbin 🚀 (Oct 16 2026 at 16:34):

in Dataverse 10, I mean :smile:

Oliver Bertuch (Oct 16 2026 at 16:35):

These days, something like this should be fairly transparent anyway. Moving the file in Git without touching the package name makes sure the past history is still available in Git blame.