Add GitHub Actions builds for oci8 and pdo_oci drivers

[morozov]

Q	A
Type	improvement
BC Break	no

Fixes #4075.

[morozov]

Not sure why code coverage didn't increase but it's not the immediate goal. Once the PR is approved, I'll remove the project from ContinuousPHP.

[greg0ire]

Not sure why code coverage didn't increase but it's not the immediate goal

There is some insight about this in the output, apparently the file can't be downloaded.:

##[warning]Codecov warning: connect ETIMEDOUT 35.199.43.247:443
bash codecov.sh -n  -F 
codecov.sh: line 1: undefined: command not found
##[warning]Codecov warning: The process 'bash' failed with exit code 127

Consider removing coverage-related parts from this PR and reporting the bug to the action's author.

EDIT: the issue only happens with the PDO_OCI job…

Also, a cool thing to do (possibly in a subsequent PR) would be to leverage job dependencies: https://docs.github.com/en/actions/reference/workflow-syntax-for-github-actions#jobsjob_idneeds

[morozov]

Consider removing coverage-related parts from this PR and reporting the bug to the action's author.

Why remove it if it doesn't fail the build? Given the error message and the fact that the oci8 build successfully uploaded its report, I'd just consider it a one-off runtime issue.

UPD: after closing and reopening the PR, both coverage reports were successfully submitted.

[greg0ire]

Yeah it does look transient. I see a lot of S in the output, they don't correspond to code that is supposed to run in these jobs do they? Because that would explain the coverage not increasing, and if it does, it would mean we no longer run OCI-related tests.
I'm referring to that kind of thing:

dbal/tests/Doctrine/Tests/DBAL/Functional/Driver/PDOOracle/DriverTest.php

Line 10 in 61a0113

* @requires extension pdo_oci

If you have a doubt about that, please commit a failure in such a test, and let's see if the jobs fail?

[greg0ire]

And if you don't have a doubt, just merge

[morozov]

I see a lot of S in the output, they don't correspond to code that is supposed to run in these jobs do they?

This is probably because the PHP runtime doesn't have any DB extensions except for the one tested against. That's why a lot of S.

[…] it would mean we no longer run OCI-related tests.

It would be only possible if the OCI-based extensions weren't installed as claimed by the action but it's not possible because the first commit fixes the failure of the test that only runs if oci8 is installed:

dbal/tests/Doctrine/Tests/DBAL/Driver/OCI8/OCI8StatementTest.php

Lines 11 to 13 in 8abd320

	/**
	* @requires extension oci8
	*/

[greg0ire]

Ok, maybe CodeCov is lost because it does not have a report for 2.10.x AND oci…

[morozov]

And if you don't have a doubt, just merge

I'm not 100% sure what's going on but I'd rather merge this as is and work on the coverage later. I'd like to get rid of ContinuousPHP and complete all the up-merges first.

[morozov]

There's definitely something weird going on beyond our code coverage settings. I just ran the tests locally with the configuration from the repository, and it worked:

$ phpunit -c ci/github/phpunit.oci8.xml --coverage-html $HOME/report

[greg0ire]

This url shows that 2 builds were received from GH. When I download them, they contain references to OCI8Exception, such as

<file name="/home/runner/work/dbal/dbal/lib/Doctrine/DBAL/Driver/OCI8/OCI8Exception.php">
        <class name="Doctrine\DBAL\Driver\OCI8\OCI8Exception" namespace="Doctrine\DBAL\Driver\OCI8">
          <metrics complexity="2" methods="1" coveredmethods="0" conditionals="0" coveredconditionals="0" statements="6" coveredstatements="0" elements="7" coveredelements="0"/>
        </class>
        <line num="17" type="method" name="fromErrorInfo" visibility="public" complexity="2" crap="6.00" count="0"/>
        <line num="18" type="stmt" count="0"/>
        <line num="19" type="stmt" count="0"/>
        <line num="20" type="stmt" count="0"/>
        <line num="21" type="stmt" count="0"/>
        <line num="23" type="stmt" count="0"/>
        <line num="24" type="stmt" count="0"/>
        <metrics loc="25" ncloc="17" classes="1" methods="1" coveredmethods="0" conditionals="0" coveredconditionals="0" statements="6" coveredstatements="0" elements="7" coveredelements="0"/>
      </file>

So the issue would be that they are not "merged" with the other ones?

Also, this screen shows only reports from GH, nothing from Travis… but the docs say having multiple sources should not be an issue.

[greg0ire]

Ah wait, this graph shows a covered OCIException: https://codecov.io/gh/doctrine/dbal/commit/8cce1649b0deea971d8e3730685f36672bbd2145/graphs/sunburst