[releng] QVTo build fails against OCL RC1 milestone

[eclipse-ocl-bot]

| --- | --- | | Bugzilla Link | 313857 | | Status | CLOSED INVALID | | Importance | P3 normal | | Reported | May 21, 2010 02:31 EDT | | Modified | May 27, 2011 02:54 EDT | | Version | 3.0.0 | | Depends on | 313124 | | Reporter | Sergey Boyko |

Description

We are unable to build QVTo bits against latest OCL 3.0.0RC1 milestone. \ The problem is that after normal source's compilation build procedure is failed to install target platform for testing. So tests are not launched and as a consequence build result is treated as failed.

Interesting that just the same sources and build configuration (Eclipse RC1 and Orbit S20100514103823) processed normally when OCL 3.0.0M7 milestone is used in place of OCL 3.0.0RC1.

While most probably the reason is somewhere in QVTo build configuration could you please take look at build console logs attached below. May be you can give some advise on these strange failures.

[eclipse-ocl-bot]

By Sergey Boyko on May 21, 2010 02:37

Created attachment 169464 QVTO build console log when using mdt-ocl-SDK-3.0.0RC1.zip

Note that mdt-ocl-SDK-3.0.0RC1.zip (comparing to mdt-ocl-SDK-3.0.0M7.zip) contains 'plugins' as addition root folder.

Problem that occurred during test platform installation:

Missing requirement: Operational QVT Core 3.0.0.v201005201551-7B1-AkF7B77RF7BRRB (org.eclipse.m2m.qvt.oml.feature.group 3.0.0.v201005201551-7B1-AkF7B77RF7BRRB) requires 'org.eclipse.ocl [3.0.0,4.0.0)' but it could not be found

:notepad_spiral: failed-with-ocl-sdk-rc1.txt

[eclipse-ocl-bot]

By Sergey Boyko on May 21, 2010 02:39

Created attachment 169465 QVTO build console log when using mdt-ocl-Update-3.0.0RC1.zip

Problem that occurred during test platform installation:

Missing requirement: Operational QVT Core 3.0.0.v201005201551-7B1-AkF7B77RF7BRRB (org.eclipse.m2m.qvt.oml.feature.group 3.0.0.v201005201551-7B1-AkF7B77RF7BRRB) requires 'org.eclipse.ocl [3.0.0,4.0.0)' but it could not be found

:notepad_spiral: failed-with-ocl-updatesitezip-rc1.txt

[eclipse-ocl-bot]

By Sergey Boyko on May 21, 2010 02:44

Created attachment 169468 QVTO build console log when using mdt-ocl-SDK-3.0.0M7.zip

Console output for successful build when OCL 3.0.0M7 milestone is used for the entirely same QVTO configuration as in previous attachments.

:notepad_spiral: ok-with-ocl-sdk-m7.txt

[eclipse-ocl-bot]

By Ed Willink on May 21, 2010 02:48

I see

Bundle org.eclipse.ocl.examples.xtext.completeocl.ui

as missing in your log. You are therefore using an MDT/OCL feature that\ includes the examples and so Xtext.

If this intentional, it should have failed on M7, and you should emulate the\ MDT/OCL build.properties.

Since you clearly did not have a problem with M7, I think this depends on Bug 313124.

[eclipse-ocl-bot]

By Sergey Boyko on May 21, 2010 07:03

(In reply to comment #4)\ Hi Ed,

Thank you for the comment. Then we'll wait for RC1a (RC2 ?) OCL milestone.

I see

Bundle org.eclipse.ocl.examples.xtext.completeocl.ui

as missing in your log. You are therefore using an MDT/OCL feature that includes the examples and so Xtext.

If this intentional, it should have failed on M7, and you should emulate the MDT/OCL build.properties.

Since you clearly did not have a problem with M7, I think this depends on Bug 313124.

[eclipse-ocl-bot]

By Alexander Igdalov on May 21, 2010 10:00

(In reply to comment #4)

I see

Bundle org.eclipse.ocl.examples.xtext.completeocl.ui

as missing in your log. You are therefore using an MDT/OCL feature that includes the examples and so Xtext.

If this intentional, it should have failed on M7, and you should emulate the MDT/OCL build.properties.

And it did fail on M7. See the attached log.

Since you clearly did not have a problem with M7, I think this depends on Bug 313124.

Hopefully, yes. Maybe an extra "plugin" folder if the root was the culprit...

[eclipse-ocl-bot]

By Ed Willink on May 21, 2010 11:09

Looking at your build.properties it seems you're making all the mistakes that we were. (I suspect we share a common ancestor.)

a) the "test" step seems to have been deprecated and replaced by "testLocal" which in conjunction with running XVNC around the entire build makes starting the tests much more reliable. (Bug 308759).

b) you are using "evil" dropin zips rather than repositories (Bug 309900).

Using the entire MDT/OCL ZIP I think ensures that you incur the all the dependencies so you must provide Xtext etc.

If you use just the repository, it may be that you don't get the dependencies of the examples you don't use.

I strongly recommend changing the dependencies part of your build.properties to emulate MDT/OCL's. (You won't need the residual Orbit ZIP since you're not fetching LPG for redistribution.) If necessary, put in Xtext, Xpand, Mwe, Mwe2 to keep life simple; you can delete them later.

[eclipse-ocl-bot]

By Alexander Igdalov on May 23, 2010 19:17

Builds with OCL RC1a fail with the same error message. When switching to repositoryURLs the build doesn't show the missing requirement error but it doesn't run tests for some reason, and thus, fails...

[eclipse-ocl-bot]

By Sergey Boyko on May 23, 2010 23:55

(In reply to comment #7)

I've switched to use 'repositoryURLs' as Ed suggested.

This helps to eliminate errors with test platform installation. Now it installs without p2 errors. However build procedure complains that there are no tests results. Which (I believe) indicates that for some reason tests are no actually run.

Also I tried to specify all dependencies from OCL build.properties (Xpand, Mwe, etc.). This did not help either.

Btw, I already faced such problem (all is run normally but build procedure complains about absence of test results) during M7 build before switching to use new Orbit bundle.

Also switching to OCL RC1a didn't help. I'll check M7 with our latest build configuration.

Looking at your build.properties it seems you're making all the mistakes that we were. (I suspect we share a common ancestor.)

a) the "test" step seems to have been deprecated and replaced by "testLocal" which in conjunction with running XVNC around the entire build makes starting the tests much more reliable. (Bug 308759).

b) you are using "evil" dropin zips rather than repositories (Bug 309900).

Using the entire MDT/OCL ZIP I think ensures that you incur the all the dependencies so you must provide Xtext etc.

If you use just the repository, it may be that you don't get the dependencies of the examples you don't use.

I strongly recommend changing the dependencies part of your build.properties to emulate MDT/OCL's. (You won't need the residual Orbit ZIP since you're not fetching LPG for redistribution.) If necessary, put in Xtext, Xpand, Mwe, Mwe2 to keep life simple; you can delete them later.

[eclipse-ocl-bot]

By Ed Willink on May 24, 2010 01:03

(In reply to comment #7)

a) the "test" step seems to have been deprecated and replaced by "testLocal" which in conjunction with running XVNC around the entire build makes starting the tests much more reliable. (Bug 308759).

You aren't running XNVC around the entire build.

Check the "Run Xvnc during build" in the Hudson configuration.

[eclipse-ocl-bot]

By Sergey Boyko on May 24, 2010 04:52

(In reply to comment #10)

Ed, you're sorcerer! Now build works!! Thank you very-very much :)

Tested for N-build and I hope nothing would prevent us from building our RC right now.

(In reply to comment #7)

a) the "test" step seems to have been deprecated and replaced by "testLocal" which in conjunction with running XVNC around the entire build makes starting the tests much more reliable. (Bug 308759).

You aren't running XNVC around the entire build.

Check the "Run Xvnc during build" in the Hudson configuration.

[eclipse-ocl-bot]

By Ed Willink on May 24, 2010 05:03

(In reply to comment #11)

Ed, you're sorcerer! Now build works!! Thank you very-very much :)

Not really; it's Nick that provided all the solutions.

I recognize the relief; this was a two day nightmare for me two weeks ago.

I'm disappointed that there seems to be no proactive mechanism for propagating the improved build practices, that arose from testLocal in February. I had a lot of pain. You had a lot of pain. We are making releng needlessly painful.

May be Buckminster and/or b3 will be wonderful.

[eclipse-ocl-bot]

By Ed Willink on May 24, 2010 05:05

Problem was out of date QVTo build procedures.

[eclipse-ocl-bot]

By Ed Willink on May 27, 2011 02:54

Closing