bad url join in celery task when site_url has a path components

[dlax]

Consider ckan.site_url = http://somehost/ckan in CKAN configuration file. With this, when building api_url here, one gets the wrong URL http://somehost/api/action because the "ckan" path component gets dropped by urljoin. One solution would be to have a trailing / in site_url configuration option but this is apparently not recommended. So I guess some url manipulation would be needed on extension side.

Note that other extensions (such as archiver and datastorer) have the same problem.

[davidread]

You're quite right.

How about changing it to:

api_url = urlparse.urljoin(context['site_url'] + '/', 'api/action')

Perhaps you could create some pull requests with this or similar change?

[dlax]

Yes I could. However, I'm not sure what the proper way to fix this. Sure your suggestion would work, but since this appears to be broken in other extensions as well, I was looking for a standard way to handle this... And looking into ckan source code, I could find a few different "solutions":

• controllers/feed.py has '/'.join([site_url, resource_path]) and also base_url + h.url_for(controller='package', ...)

• controllers/package.py has
  c.datastore_api = '%s/api/action' % \ config.get('ckan.site_url', '').rstrip('/')
  similar things in lib/dictization/model_dictize.py or model/package.py

• lib/cli.py has

  fetch_url = config['ckan.site_url']
  url = h.url_for(controller='package', action='read', id=dd['name'])
  url = urlparse.urljoin(fetch_url, url[1:]) + '.rdf'
  

  which, I guess, would also fail in case site_url has a path component.

Did I miss something or should we really go and fix all usages of urljoin the way you suggests?

[wardi]

site_url shouldn't have path components, that's what root_path is for. The url_for helper can be used to generate full paths correctly, I think.