Abstract

Package Relationships have not taken off in the 18 months we've had them in the API. There are some issues with them and we need to spend more time improving them or consider getting rid of them.

The Problem

Original use cases are expressed here: #253 Here are comments about how we could handle these specific examples better:

1. groups of packages - maybe better with a custom tag?
2. fragment resources - soon to be covered by 'kind' resource field #957

3&5. derived resource - better to have some sort of resource relationship perhaps?

4. linked resource - again better to have some sort of resource relationship perhaps?

Outstanding issues needing serious effort to fix:

• #256 Editing them in Web UI (not done yet)
• #1288 Package edit/creation can't include 'relationships' field

Specification

Remove relationships from model, API, tests, Web UI. Data migration to remove from db.

Why do it this way

Getting frustrated having problems with the code, when it's not used much. Often asked about what it's for, but rarely used. Seems an overly complicated design.

Backwards Compatibility

n/a

Implementation plan

Deliverables

See Specification

Risks and mitigations

Risk: a customer suddenly wants this, and the new ways to relate resources are not in place yet.

Mitigation: discuss this decision thoroughly to make sure we are confident the use cases are not important. Discuss with team, ckan-discuss and specifically the LOD people who have some related packages on thedatahub.org.

Participants

David Read

Progress

Not yet.

In password reset, it gets confused if you have two similar users. This is because with the string the user provides, it searches several fields, not just name but also fullname and email address, allowing you to search for these. But only name is unique.

Specific problem: Ira searches for "Irina" then it finds both: <User name=irina fullname=Irina [email protected] ] and <User name=shevski fullname=Ira email=> (I think)

Maybe need to choose which field it searches?

Ticket #787 described join auth between CKAN and Drupal. The authentication part is live and implemented. This ticket is a placeholder for work that will be needed in the new auth system to link authorization functions to Drupal. It is dependent on the groups refactor.

A possible bug in CKAN when I tried deleting users using "paster --plugin=ckan user delete" command.

To reproduce the bug do the following:

1. Create a user with an ID (which in my case was a user's full name)

that contains non-unicode caracters like Norwegian "æ", "ø", or "å".

2. Make sure that you can see something like the example below:

(pyenv) rm@mycomputer:$ paster --plugin=ckan user Users: name=Rustæm

3. Then try deleting the user with following command:

(pyenv) rm@mycomputer:$ paster --plugin=ckan user delete "Rustæm"

You should now get a python encoding error. I know that this is quite rare case, but in our case it caused some trouble. Could you guys have a look at this bug?

CKAN ver. 1.3.3.

We get this i18n error for the URL ​http://no.ckan.net/authorizationgroup (which runs 1.4.2 with Norwegian translation). I'm not sure why it occurs. Fixing it is low importance though, because this site doesn't use Authorization Group feature

Usually caused by bots. The user does not appear to be logged in.

WebApp Error: <type 'exceptions.IndexError'>: list index out of range
...
Module genshi.filters.i18n:177 in _generate
<<                  msgbuf.append(*previous)
                       previous = None
                   for event in msgbuf.translate(gettext(msgbuf.format())):
                       yield event
                   if previous:
>>  for event in msgbuf.translate(gettext(msgbuf.format())):
Module genshi.filters.i18n:1029 in translate
<<                      )
       
               parts = parse_msg(string)
               parts_counter = {}
               for order, string in parts:
>>  parts = parse_msg(string)
Module genshi.filters.i18n:1143 in parse_msg
<<      if string:
               parts.append((stack[-1], string))
       
           return parts
>>  parts.append((stack[-1], string))
IndexError: list index out of range

I have no idea whether this was a deliberate decision or not, but there is a total absence of any UI with which to delete resources from the currently deployed version of thedatahub.org.

SOLR search doesn't support searching for part of an extra field, but it does for other fields.

i.e. title="One Two Three" matches q=one AND q=title:one and geographic_coverage="England Scotland" matches q=England BUT NOT q=geographic_coverage:England

This problem emerged when we went to SOLR in #1275 (CKAN 1.5a). Tests were skipped.

This is could be a problem for DGU and maybe elsewhere.

Needs to update on:

1. language - It's broken on thedatahub.org if you change to German and then click on the CKAN icon top-left to go to thedatahub.org again.

AND

2. whether signed in or not - #1373

BUT not latest revisions (which is what it was)

Or get rid of etag caching on this page and others?

RSS 'subscribe' buttons appeared in many places on the site but were not very helpful. They took (confused) users pointed to the raw feed code, and Google Reader could not understand the feed. Safari, however, could interpret it correctly.

Their presentation needs to be clear and consistent; the RSS feed really needs testing in a variety of readers; and we need to decide exactly which items should get a feed. (Package updates? Groups?)

I couldn't log into theDataHub with OpenID today. I tried both Google ID and MyOpenID. Both times the login on the remote auth server went fine, but when it returns you to theDataHub you get error "Login failed. Bad username or password."

ckan-discuss discussion suggests changes to the 'openness' indicator ( ​http://lists.okfn.org/pipermail/ckan-discuss/2011-October/001786.html )

Dataset view page:

• If there is an explicit but non-OKD compliant license, such as CC-BY-NC, then this should be stated explicitly, perhaps: “This dataset is Not Open. License: Creative Commons Attribution Noncommerical. This is not an open license as it does not meet the Open Knowledge Definition.”
• If the license is marked as “Other::License Not Specified”, then this should be stated explicitly, perhaps: “This dataset is Not Open. It is published without an explicit license, the publisher reserves all rights to the dataset.”
• 3. If the license field was left empty by the contributor of the Data Hub record, then again this should be stated explicitly, perhaps: “This dataset is Not Open. The license of this dataset is unknown or unspecified. Start an enquiry on IsItOpenData? »
• There is a bug so that non-open licenses doesn't have an openness notice.
• If downloadable resources are not available, this should not affect 'openness' - check this has been removed.

Super ticket: #1190

A data processing system which utilizes the Webstore. One could get a long way with simple javascript running in the browser for development with this javascript then run offline using something like nodejs. Alternatively one could allow one to specify a url to e.g. a python file which would then be run in a sandbox (with access to some specified set of python modules)

Many actions in the Action API require parameters. What params are needed should be listed and checked. Because currently, if you get them wrong you simply get a useless 500 error.

Currently they are listed in the docs, extracted from the code manually.

So you could GET /action/api/package_list to receive not only the help text, but a list of arguments.

And if you send an extra or missing argument then an intelligent error message can be returned.

implementation

How about some sort of decorator on the action function:

@logic_params(id, offset, limit)
def get_package_list(context, data_dict):
    ...

This would do the param checking, and is there a way to extract these params from the function? Or do a registration of the logic function?

I'd certainly like to keep the list of the list of params for the function with the function, for ease of reading the code.

Another good thing would be to pass in the params named as themselves, rather than having them contained in the data_dict.

"when logging into ​http://register.data.overheid.nl/ with OpenID, the /user/me page gives a 404. CKAN version 1.3.4."

n the manual it says an API key kan be created via ​http://test.ckan.net/user/me /. However, when I try the corresponding ​http://register.data.overheid.nl/user/me, I get a 404 error (not found).

e.g. package_register_get.

The current harvesting implementation can only have one instance per server.

We could either:

• Accept this
• Change the RabbitMQ code to support multiple instances
• Switch the entire harvesting to use Celery
• Wait for James's redis based live feedback based system

ordering on homepage by name instead of group count

e.g. ​http://thedatahub.org/dataset/osm%402011-07-12T12%3A16%3A47.590358 gives:

This is the current revision of this dataset, as edited at 2011-07-12 12:16.

but it should say the date 2011-07-12 (which is the data being displayed).

The problem is looking at PackageRevision?, when it might be a tag or extra.

Because of #1514 we should just disable name changing, until #1514 is done.

The use of username is problematic because username's can change.

• Change all revision creation code to use user id (simplest is to change c.author field in lib/base.py (?))
  • (?) Add a field ipaddr for ip address of anonymous users? (or just keep putting this in author field on Revision and then acception that those won't match when we do a look up against user table)
• Change user view page to look up against user id rather than name
• Perform migration on existing Revision objects
  • Match should probably be against both openid and username when searching Revisions' author field (especially true on CKAN where some people have already changed their username from being their openid)

A sysadmin should be able to easily examine a suspect group or package, determine if it was created by a spammer (as opposed to being a legitimate object that has been graffitied by a spammer) and purge it.

The existing two-stage revision delete is currently unreliable and perhaps too laborious.

Olav and Richard have needs along this line.

CKAN is pulling a number of resources from external locations. This causes problems when connectivity is limited and you have to work locally. Maybe some of them cold be moved to CKAN source to avoid external requests.

Quick search:

./ckan/templates/layout_base.html:            <img src="http://assets.okfn.org/images/logo/okf_logo_white_and_green_tiny.png" id="footer-okf-logo" />
./ckan/templates/layout_base.html:            <a href="http://opendefinition.org/"><img alt="This Content and Data is Open" src="http://assets.okfn.org/images/ok_buttons/od_80x15_blue.png" style="border: none ; margin-bottom: -4px;"/></a>
./ckan/templates/package/resource_read.html:                <img src="http://assets.okfn.org/images/ok_buttons/od_80x15_blue.png" alt="[Open Data]" />
./ckan/templates/package/read.html:          <img src="http://assets.okfn.org/images/ok_buttons/od_80x15_blue.png" alt="[Open Data]" /></a>
./ckan/templates/_util.html:                    <img src="http://assets.okfn.org/images/ok_buttons/od_80x15_blue.png" alt="[Open Data]" />
./ckan/templates/_util.html:                  <img src="http://assets.okfn.org/images/ok_buttons/od_80x15_blue.png" alt="[Open Data]" />
./ckan/public/scripts/vendor/ckanjs/1.0.0/ckanjs.js:      this.$dialog.html('<h2>Loading results...</h2><img src="http://assets.okfn.org/images/icons/ajaxload-circle.gif" />');
./ckan/public/scripts/vendor/ckanjs/1.0.0/ckanjs.js:          self.setMessage('Uploading file ... <img src="http://assets.okfn.org/images/icons/ajaxload-circle.gif" class="spinner" />');
./ckan/public/scripts/vendor/ckanjs/1.0.0/ckanjs.js:      self.setMessage('Checking upload permissions ... <img src="http://assets.okfn.org/images/icons/ajaxload-circle.gif" class="spinner" />');
Binary file ./ckan/lib/app_globals.pyc matches
./ckan/lib/app_globals.py:                                  'http://assets.okfn.org/p/ckan/img/ckan.ico')
./ckan/config/deployment.ini_tmpl:ckan.favicon = http://assets.okfn.org/p/ckan/img/ckan.ico

It should be possible to pass userid/username and api key and obtain a response from CKAN for external services that use CKAN auth. Those services shouldn't be talking to the DB directly.

(details embargoed until 31/1/2012)

I want to browse tags, but filtered for a particular group. Currently our tag API doesn't allow for filtering by group.

This is important for improving groups as communities within a site #1521. It would be easy to do this by adding an option to filter by a group. BUT are there any other use cases that would warrant a more complete faceted tag search?

--

BTW I can currently draw a tag cloud for a group - I can get the top tags used in a group like this:

curl ​http://thedatahub.org/api/action/package_search -d '{"q":"groups:country-ca", "facet.field":"tags", "rows":"0"}'

but it only contains the top 20 tags.

ckanext-moderatededits requires an old and possibly development version of CKAN. It would be good to update it for later CKAN versions.

According to the README, you need CKAN from branch feature-1141-moderated-edits-ajax but the changelog suggests this branch went into version 1.4.2. So it possibly works with 1.4.2 and 1.4.3(.1). But CKAN 1.5 has revamped templates, so the genshi stream filters definitely don't work.

BTW history_ajax/read_ajax calls have been deprecated in CKAN since 1.5.2a and will need fixing up to use the Action API too as part of this.

This would allow deployment via Nginx or Apache using proxy to Paster, uWSGI. At the moment CKAN isn't aware of the proxy's IP address so when you perform an action which does a redirect (such as adding a package), CKAN redirects you to the *internal IP* not the external *proxy IP*.

We would like this work to facilitate testing within VMs as part of our new build infrastructure.

It would also be nice if CKAN worked when mounted at a path other than /. That could be dealt with in another ticket because it isn't a problem at the moment.

I had a package descriptions with URLs that contain "group:foo". This produces garbled output as the system tries to generate two sets of links: the outer link and an inner link.

Need to fix the parser.

Text:

Webdienst basierende Bereitstellung von Geobasisdaten der Freien und Hansestadt Hamburg. Folgende Geobasisdaten werden als WebMapTileService? (WMT-S) für die Dauer des Wettbewerbs netzbasiert unter der Creative Commons Lizenz zur Verfügung gestellt: Digitale Orthophotos 40 cm Auflösung (Layer: apps4d_DOP40), Digitale Stadtkarte (Layer: apps4d_DISK), Digitale Regionalkarte (Layer: apps4d_DIRK), Digitale Karte 1:5000 (Layer: apps4d_DK5).

Metadateneinträge zu den Daten im PortalU:

• [apps4d_DOP40](​http://www.portalu.de/trefferanzeige?docuuid=D6082285-6CAC-49C5-B7ED-B5A72BC17DEC&plugid=/kug-group:kug-iplug-udk-db_hh)
• [apps4d_DISK](​http://www.portalu.de/trefferanzeige?docuuid=D2DA3359-3DC9-4CA4-AE9B-7C63D53B220F&plugid=/kug-group:kug-iplug-udk-db_hh)
• [apps4d_DIRK](​http://www.portalu.de/trefferanzeige?docuuid=B7DDD599-F677-4C80-BED5-F11D82D5D1DD&plugid=/kug-group:kug-iplug-udk-db_hh)
• [apps4d_DK5](​http://www.portalu.de/trefferanzeige?docuuid=2AE6D23E-48A5-4D85-BC0A-160737E0C8D2&plugid=/kug-group:kug-iplug-udk-db_hh)

One fix is quoting the URLs

Instead of alphabetically as we do currently, alternatively by most viewed datasets

for ​http://thedatahub.org/dataset

Super ticket: #1506

Need to decide which term to use and then have the same for editing as well as viewing a dataset.

In creating/editing a dataset, want more explanation about adding extra fields (probably as a tooltip or similar).. i.e. that this let's you add extra custom metadata such as 'location: uk' which is then searchable etc

Jaakko Louhio reported that the wrong Routes version got installed during CKAN's package install.

He is using Ubuntu 10.04 and he believes it install Routes 1.12.3 instead of what we use which is 1.11.

There's a discrepency in repoze.who versions between the source and package installs:

• repoze.who - package 1.0.18 vs source 1.0.19
• repoze.who-friendlyform - package 1.0b3 vs source 1.0.8

We get a test failure [1] with the 1.0b3 version (from the ubuntu 10.04 python-repoze.who-plugins package). But we've not noticed any problems on s057 instances (br, no, ie etc) which have the package versions of repoze.who.

The reason the package install uses the earlier packaged versions rather than the ones we'd like is that repoze uses all sorts of horrendous import hacks, making it too difficult to put into our 'ckan-conflict' source package.

James suggests we 'do something horrible like dynamically patch repoze on CKAN import'.

[1] ​http://buildbot.okfn.org/builders/builder-ckan/builds/1371/steps/shell/logs/stdio ERROR: ckan.tests.functional.test_user.TestUserController?.test_user_create_unicode

The 'editor', 'anon_editor' and 'reader' roles are intended to have immutable actions. This was designed to prevent their names being subverted - e.g. an editor should always be able to edit! It also meant that when we add Actions (e.g. DELETE-PACKAGE) then it can be added sensibly to these roles in an upgrade just by changing the defaults table (ckan/model/authz.py).

The problem is that this immutability is only enforced on 'db upgrade'. So you can happily change the editor role using the paster command and it works, right up until you do an upgrade and realise permissions are different.

We should stop the paster commands being able to edit these roles. Or get rid of the immutability completely. Views?

Add image attribute for Dataset and Group which would be shown in prominent place.

For dataset this could be a relevant image and for Group this would be more of a logo.

I think this has very high value from a UX point of view.

Issues

• Do we need to support uploading and storing the image (or do users store elsewhere)
  • Could use something like transloadit service
  • Use our own storage facility with a reserved directory
• Do we need to do resizing etc?
  • Maybe. Perhaps can prefer this in reasonably small size and then just size down in css for thumbnails etc.

1. Make it possible to search by tag (e.g. by typing tag:csv into the search bar and clicking enter, it should add the 'csv' tag facet to the search)

2. Rename and standardise the list of format tags, on search page this should also be called 'Format' instead of 'res_format' (in the right hand side bar on search page).

3. Make it possible to view full list of tags, formats and groups by clicking on the name. From here you should be table to click on a classification and go back to a search page faceted by that classification. E.g. from search page, click on 'tags', on tag page click on 'london' or whatever, and be navigated back to search page with search within 'london' tag only. Or y'know, a better way of doing it.

4. More standard classifications, such as 'Location' and 'Theme' - like on publicdata.eu

5. Blue search button should be displayed in line with the search bar, not underneath

6. Datasets should be displayed in order of most viewed or downloaded instead of alphabetically. For alphabetic search we could consider adding a way to facet by first letter of dataset name

Changes to make search both more exploratory and more satisfying to use

• Search query build - #1603
  • Ability to add new facet fields "live"
  • That is add fields which then contain faceted options (a bit like data.hri.fi)
• (??) Autocomplete / drop down on search (i.e. search while you type)
  • Dubious about value / cost ratio here

Probably would involve to pure JS and HTML implementation.

Implementation

Probably require

• API changes to expose solr style API directly #1737

Currently the activity streams database tables just get longer and longer over time. Do we want to eventually delete the oldest activities, to keep the length of the table within limits?

Move ckan/lib/activity.py moved to into the model - say ckan/model/activity_extension.py, because it's so tightly knit with the model code, whereas most of the lib code is used in the controllers.

So users can rename a tag and/or move it between vocabularies using the API.

Currently we have create_tag() and delete_tag(), but if you were to 'update' a tag by deleting it and then recreating it all the datasets that had that tag will have lost it and you'll have to re-add it to them all.

What should happen to datasets that have the tag, if the tag gets moved between vocabularies? All the datasets just keep the tag with the new vocabulary? This will become a problem if/when we support 'radio button'-style vocabularies (where each dataset can only have one tag from the vocabulary).

You get an exception if you use the API to search packages and specify a non-ascii character in a field other than q.

For example:

http://catalogue.data.gov.uk/api/search/package?title=%E2%80%93

This "N-dash" (Unicode character 2013) causes this exception:

Module ckan.controllers.api:460 in search
<<                      if ver in u'12':
                               # Otherwise, put all unrecognised ones into the q parameter
                               params = convert_legacy_parameters_to_solr(params)
                           query = query_for(model.Package)
                           results = query.run(params)
>>  params = convert_legacy_parameters_to_solr(params)
Module ckan.lib.search.query:38 in convert_legacy_parameters_to_solr
<<      for search_key in non_solr_params:
               value_obj = legacy_params[search_key]
               value = str(value_obj).replace('+', ' ')
               if search_key == 'all_fields':
                   if value:
>>  value = str(value_obj).replace('+', ' ')
UnicodeEncodeError: 'ascii' codec can't encode character u'\u2013' in position 30: ordinal not in range(128)

This problem affects CKAN 1.5 and 1.5.1 only.

You can reset your password (#1186) but you have to know the URI (/user/reset) - there is no link to this page!

Huge number of accumulated (and likely unnecessary) scripts in /bin directory.

I have just deny visitors the create-user permission:
sudo -u ckanstd paster --plugin=ckan roles deny reader create-user -c /etc/ckan/std/std.ini
sudo -u ckanstd paster --plugin=ckan roles deny anon_editor create-user -c /etc/ckan/std/std.ini

and after restarting, the register link is *not* hidden, but now when you access the register page, it shows you this message "Unauthorized to create a user" (when not logged in). But anyway that is an improvement.

Purging datasets (deleting them fully, not just changing the state to 'deleted') is important for users testing dataset creation over the API on a test CKAN instance.

Without this, they need to resort to more difficult methods such as:

• cleaning and reloading the database
• setting the test datasets to state 'deleted' and also appending a suffix '_00' and incrementing the number until there is no clash of names.

Requested for NHSIC.

Implementation

• This could slot into the Action API.
• Of course we would need to ensure the user's had been given the specific right to purge.
• I suggest we log the full details of the dataset being purged.

Now that ckanext-storage is back into the core (v1.6), CKAN documentation should probably contain info on ​Storate Metadata API.

Just like we display download counts we should display view counts.

You configure the extension with some RSS and/or Atom feeds, it automatically reads items from these feeds and makes them available in the template context, you write a custom template to e.g. display 'news' items from a Wordpress blog on your front page.

This extension might be simpler and less fragile than ckanext-wordpresser, and also more generally useful.

Details

• Mark Pilgrim's Universal Feed Parser might be useful for reading the feeds

• Feed items should probably be cached somewhere

Suggested

• The news item 'widget' should be wrapped in a known class so that it can be styled easily regardless of the format of any HTML entry.

• For non-HTML formatted items (Atom should tell you the content type of the entry) maybe we should have a template for rendering each item along with any enclosures that it might reference

• Caching is pretty crucial and should probably obey the ttl of the feed.

From Ira: Preview for google spreadsheets are not displaying correctly for me in Firefox v.10.0.02, fine in Chrome. ​http://i.imgur.com/KJaqz.png

When viewing a dataset, it would be nice to show a couple of 'Related Datasets'. i.e. ones that are similar.

SOLR has a feature for finding documents similar to a particular document, called 'More Like This'.

We would like this for DGU.