it seams like to be and problem at travis itself? As the file that throws the error is /home/travis/.travis/job_stages: line 104: ?

It's either an issue with the Travis platform or with the PHP 7.3 RC itself. Laravel too has the same problem.

Well apparently the solution is to ignore it... e5157f2

Solution is not the right term but as this is something out of the control of our code and seams to be a problem on other repos too we should give travis time to take a look but not blocking all PRs to be merged when Travis fails.

The test suite runs perfectly, and completes, on my mac with PHP 7.3beta3.

Yes it seams that travis now uses RC5 and as said before the error message looks like a travis platform issue not a PHP issue itself lets check what happens at the time RC6 comes out.

With PHP 7.3RC5 in a docker container I get:

and that comes from this test

Starting test 'JComponentRouterRulesNomenuTest::testConstruct'.

undefined index SCRIPT_NAME ? Can you identify the failing test? Can you confirm that docker container runs the tests successful on an other PHP Version? (just to make sure anything required is running there)

failing test was

Starting test 'JComponentRouterRulesNomenuTest::testConstruct'.

I removed that, next failure with the same error is

Starting test 'JHelpTest::testCreateUrl'.
EUndefined index: SCRIPT_NAME

removed that test and then

Starting test 'JHtmlIconsTest::testButtons'.
EUndefined index: SCRIPT_NAME

removed that test and then

Starting test 'JLayoutFileTest::testEscape'.
EUndefined index: SCRIPT_NAME

Congrats, Phil. You've just run into the same issues what tests we have left have always had. There's global state bleedover between tests and they aren't portable across environments.

====

Yes it seams that travis now uses RC5 and as said before the error message looks like a travis platform issue not a PHP issue itself lets check what happens at the time RC6 comes out.

There's no assurance that it is not a PHP issue. Just because the file path has what is presumably the file that the YAML config is parsed into does not automatically mean it is a Travis-CI issue. FWIW I did get a project to run its test suite on Travis-CI with PHP 7.3 RC 5 so that does give some credit to it not being a platform (whatever the system the Travis-CI run is being made on) problem and some credit toward a potential PHP bug (and considering both Joomla's and Laravel's test suites are having the same issue, that does seem to strengthen that argument).

Congrats, Phil. You've just run into the same issues what tests we have left have always had. There's global state bleedover between tests and they aren't portable across environments.

This project never ceases to amaze me. The one whole point of tests is that they are runable and portable... /facepalm

After faking SCRIPT_NAME in TestCase.php setUp function (and directly in a load of tests) I was able to run the complete test suite on PHP 7.3RC5. in a docker container (Alpine based, PHP Earth packages - same build that ALL the LIVE myJoomla.com service containers are running in production)

As a core file is generated, is there any interest in looking at something like http://jsteemann.github.io/blog/2014/10/30/getting-core-dumps-of-failed-travisci-builds/ ?

It's a Travis-CI issue.

Framework Repos:

• https://travis-ci.org/joomla-framework/application/builds/454758565 passed on 7.3
• https://travis-ci.org/joomla-framework/archive/builds/454758629 passed on 7.3
• https://travis-ci.org/joomla-framework/authentication/builds/454758677 failed on 7.3
• https://travis-ci.org/joomla-framework/controller/builds/454760241 passed on 7.3
• https://travis-ci.org/joomla-framework/crypt/builds/454760386 passed on 7.3
• https://travis-ci.org/joomla-framework/data/builds/454760437 passed on 7.3
• https://travis-ci.org/joomla-framework/database/builds/454762736 is still going but all 7.3 builds have passed as of this writing

CMS Repo:

• https://travis-ci.org/joomla/joomla-cms/builds/454764451 passed building this commit
• https://travis-ci.org/joomla/joomla-cms/builds/454764377 passed building this commit (this is the one that tried to get the core dump)

very very strange :)