Fix broken tests #199
Fix broken tests #199
Conversation
Tests were relying on _rest action calls which have now been removed in master (and 2.8) and therefore tests were failing. Makes the tests work, although there is probably some effort required in determining why metadata_modified is not being returned from package_show calls. Also caches the pip output, and tests against newer versions of CKAN
| - CKANVERSION=master POSTGISVERSION=2 | ||
| - CKANVERSION=2.2 POSTGISVERSION=2 | ||
| - CKANVERSION=2.3 POSTGISVERSION=2 | ||
| - CKANVERSION=2.4 POSTGISVERSION=2 |
There was a problem hiding this comment.
These versions can be definitely be dropped but we need to ensure it works with 2.5 to 2.8. Can you enable these versions and see if tests pass or what we need to fix?
There was a problem hiding this comment.
I can. When I tried switching to 2.6, 2.7, 2.8 pip installing the pg lib complained that it couldn't find a version from the string 10.1, which I think seems to be a problem with earlier versions of the postgres library. Not sure why it was trying at all with 10.x though ....
Will try getting them to work this evening ...
There was a problem hiding this comment.
Yes, we've had to fix this in many other extensions as well, this should do the trick: frictionlessdata/ckanext-validation@c652cee
Fixes the problem testing multiple CKAN versions by making sure it uses psycopg2 2.7 rather than the broken psycopg2 2.6
|
Mostly seems to fix the problem, but now downloading and building the patched libxml occassionally fails. Maybe we can vendor that file that is downloaded? Am asking because it's 5Mb but if you think it might be safer for the tests? |
|
@rossjones I'd try to avoid adding the libxml2 file to the repo. Maybe that's not needed any more or we can update the image used by Travis. Or if the failures are persistent we can just host the file on S3, to make it more reliable |
|
🎉 |
|
Seems to be happy now testing 2.5 -> 2.8 + master |
|
Thanks @rossjones! |
Tests were relying on _rest action calls which have now been removed in
master (and 2.8) and therefore tests were failing.
Makes the tests work, although there is probably some effort required in
determining why metadata_modified is not being returned from
package_show calls.
Also caches the pip output and only tests again latest version of CKAN. For some reason, apart from the first test the following tests were attempting to run with postgres 10.1