Step 0: updating code to CubicWeb 3.9 / cubicweb-file 1.9

CubicWeb 3.9 brings several improvements that we'll want to use, and the 1.9 version of the file cube has a major change: the Image type has been dropped in favor of an IImage adapter that makes code globally much cleaner (although this is not directly visible here). So the first thing to do is to upgrade our cube to the 3.9 API. As CubicWeb releases are mostly backward compatible, this is not mandatory but it's easier to follow changes as they come than having a huge upgrade to do at some point. Also, this remove deprecation warnings which are a bit tedious...

Since we only have very few lines of code, this step is pretty simple. Actually the main thing we have to do is to upgrade our schema, to remove occurrences of the Image type or replace them by the File type. Here is the (striped) diff:

 class comments(RelationDefinition):
     subject = 'Comment'
-    object = ('File', 'Image')
+    object = 'File'
     cardinality = '1*'
     composite = 'object'

 class tags(RelationDefinition):
     subject = 'Tag'
-    object = ('File', 'Image')
+    object = 'File'

 class displayed_on(RelationDefinition):
     subject = 'Person'
-    object = 'Image'
+    object = 'File'

 class situated_in(RelationDefinition):
-    subject = 'Image'
+    subject = 'File'
     object = 'Zone'

 class filed_under(RelationDefinition):
-    subject = ('File', 'Image')
+    subject = 'File'
     object = 'Folder'

 class visibility(RelationDefinition):
-    subject = ('Folder', 'File', 'Image', 'Comment')
+    subject = ('Folder', 'File', 'Comment')
     object = 'String'
     constraints = [StaticVocabularyConstraint(('public', 'authenticated',
                                                'restricted', 'parent'))]

 class may_be_readen_by(RelationDefinition):
-    subject = ('Folder', 'File', 'Image', 'Comment',)
+    subject = ('Folder', 'File', 'Comment',)
     object = 'CWUser'


-from cubes.file.schema import File, Image
+from cubes.file.schema import File

 File.__permissions__ = VISIBILITY_PERMISSIONS
-Image.__permissions__ = VISIBILITY_PERMISSIONS

Now, let's set the dependency in the __pkginfo__ file. As 3.8 simplifies this file, we can merge __depends_cubes__ (as introduced in the first blog of this series) with __depends__ to get the following result:

__depends__ = {'cubicweb': '>= 3.9.0',
               'cubicweb-file': '>= 1.9.0',
               'cubicweb-folder': None,
               'cubicweb-person': None,
               'cubicweb-zone': None,
               'cubicweb-comment': None,
               'cubicweb-tag': None,
               }

If your cube is packaged for debian, it's a good idea to update the debian/control file at the same time, so you won't forget it.

That's it for the API update, CubicWeb and cubicweb-file will handle other stuff for us. Easy, no?

We can now start some more fun stuff...

Step 1: let's improve site's usability for our visitors

The first thing I've noticed is that people to whom I send links to photos with some login/password authentication get lost, because they don't grasp they have to login by clicking on the 'authenticate' link. That's probably because they only get a 404 when trying to access an unauthorized folder, and the site doesn't make clear that 1. you're not authenticated, 2. you could get more content by authenticating yourself.

So, to improve this situation, I decided that I should:

  • make a login box appears for anonymous, so they see at a first glance a place to put the login / password information I provided
  • customize the 404 page, proposing to login to anonymous.

Here is the code, samples from my cube's views.py file:

from cubicweb.selectors import is_instance
from cubicweb.web import box
from cubicweb.web.views import basetemplates, error

class FourOhFour(error.FourOhFour):
    __select__ = error.FourOhFour.__select__ & anonymous_user()

    def call(self):
        self.w(u"<h1>%s</h1>" % self._cw._('this resource does not exist'))
        self.w(u"<p>%s</p>" % self._cw._('have you tried to login?'))

class LoginBox(box.BoxTemplate, basetemplates.LogFormView):
    """display a box containing links to all startup views"""
    __regid__ = 'sytweb.loginbox'
    __select__ = box.BoxTemplate.__select__ & anonymous_user()

    title = _('Authenticate yourself')
    order = 70

    def call(self, **kwargs):
        self.w(u'<div class="sideBoxTitle"><span>%s</span></div>' % self.title)
        self.w(u'<div class="sideBox"><div class="sideBoxBody">')
        self.login_form('loginBox')
        self.w(u'</div></div>')

The first class provides a new specific implementation of the default page you get on a 404 error, to display an explicit message for anonymous users.

Note

Thanks to the selection mechanism, it will be selected for anonymous users, since the additional anonymous_user() selector gives it a higher score than the default, and not for authenticated since this selector will return 0 otherwise (hence the object won't be selectable).

The second class defines a simple box, that will be displayed by default with boxes in the left column, thanks to default box.BoxTemplate'selector. The HTML is written to match default CubicWeb boxes style. To get the actual login form, we inherit from the LogFormView view which provides a login_form method (handling some stuff under the cover for us, hence the multiple inheritance), that we simply have to call to get the form's HTML.

login box / 404 screenshot

The login box and the custom 404 page for an anonymous visitor (translated in french)

Step 2: providing a custom index page

Another thing we can easily do to improve the site is... A nicer index page (e.g. the first page you get when accessing the web site)! The default one is quite intimidating (that should change in a near future). I will provide a much simpler index page that simply list available folders (e.g. photo albums in that site).

from cubicweb.web.views import startup

class IndexView(startup.IndexView):
    def call(self, **kwargs):
        self.w(u'<div>\n')
        if self._cw.cnx.anonymous_connection:
            self.w(u'<h4>%s</h4>\n' % self._cw._('Public Albums'))
        else:
            self.w(u'<h4>%s</h4>\n' % self._cw._('Albums for %s') % self._cw.user.login)
        self._cw.vreg['views'].select('tree', self._cw).render(w=self.w)
        self.w(u'</div>\n')

def registration_callback(vreg):
    vreg.register_all(globals().values(), __name__, (IndexView,))
    vreg.register_and_replace(IndexView, startup.IndexView)

As you can see, we override the default index view found in cubicweb.web.views.startup, getting back nothing but its identifier and selector since we override the top level view's call method.

Note

In that case, we want our index view to replace the existing one. We implement the registration_callback function, in which we code a registeration of everything in the module but our IndexView, then we register it instead of the former index view.

Also, we added a title that tries to make it more evident that the visitor is authenticated, or not. Hopefully people will get it now!

default index page screenshot

The default index page

new index page screenshot

Our simpler, less intimidating, index page (still translated in french)

Step 3: more navigation improvements

There are still a few problems I want to solve...

  • Images in a folder are displayed in a somewhat random order. I would like to have them ordered by file's name (which will usually, inside a given folder, also result ordering photo by their date and time)
  • When clicking a photo from an album view, you've to get back to the gallery view to go to the next photo. This is pretty annoying...
  • Also, when viewing an image, there is no clue about the folder to which this image belongs to.

I will first try to explain the ordering problem. By default, when accessing related entities by using the ORM's API, you should get them ordered according to the target's class fetch_order. If we take a look at the file cube's schema, we can see:

class File(AnyEntity):
    """customized class for File entities"""
    __regid__ = 'File'
    fetch_attrs, fetch_order = fetch_config(['data_name', 'title'])

By default, fetch_config will return a fetch_order method that will order on the first attribute in the list. We could expect to get files ordered by their name. But we don't. What's up doc ?

The problem is that files are related to folder using the filed_under relation. And that relation is ambiguous, eg it can lead to File entities, but also to Folder entities. In such a case, since both entity types don't share the attribute on which we want to sort, we'll get linked entities sorted on a common attribute (usually modification_date).

To fix this, we have to help the ORM. We'll do this in the method from the ITree folder's adapter, used in the folder's primary view to display the folder's content. Here's the code that I've put in our cube's entities.py file, since it's more logical stuff than view stuff:

from cubes.folder import entities as folder

class FolderITreeAdapter(folder.FolderITreeAdapter):

    def different_type_children(self, entities=True):
        rql = self.entity.cw_related_rql(self.tree_relation,
                                         self.parent_role, ('File',))
        rset = self._cw.execute(rql, {'x': self.entity.eid})
        if entities:
            return list(rset.entities())
        return rset

def registration_callback(vreg):
    vreg.register_and_replace(FolderITreeAdapter, folder.FolderITreeAdapter)

As you can see, we simply inherit from the adapter defined in the folder cube, then we override the different_type_children method to give a clue to the ORM's cw_related_rql method, that will generate the rql to get entities related to the folder by the filed_under relation (the value of the tree_relation attribute). The clue is that we only want to consider the File target entity type. By doing this, we remove the ambiguity and get back a RQL query that correctly orders files by their data_name attribute.

Note

  • Adapters have been introduced in CubicWeb 3.9 / cubicweb-folder 1.8.
  • As seen earlier, we want to replace the folder's ITree adapter by our implementation, hence the custom registration_callback method.

Ouf. That one was tricky...

Now the easier parts. Let's start by adding some links on the file's primary view to see the previous / next image in the same folder. CubicWeb provides a component that do exactly that. To make it appear, it has to be adaptable to the IPrevNext interface. Here is the related code sample, extracted from our cube's views.py file:

from cubicweb.selectors import is_instance
from cubicweb.web.views import navigation


class FileIPrevNextAdapter(navigation.IPrevNextAdapter):
    __select__ = is_instance('File')

    def previous_entity(self):
        rset = self._cw.execute('File F ORDERBY FDN DESC LIMIT 1 WHERE '
                                'X filed_under FOLDER, F filed_under FOLDER, '
                                'F data_name FDN, X data_name > FDN, X eid %(x)s',
                                {'x': self.entity.eid})
        if rset:
            return rset.get_entity(0, 0)

    def next_entity(self):
        rset = self._cw.execute('File F ORDERBY FDN ASC LIMIT 1 WHERE '
                                'X filed_under FOLDER, F filed_under FOLDER, '
                                'F data_name FDN, X data_name < FDN, X eid %(x)s',
                                {'x': self.entity.eid})
        if rset:
            return rset.get_entity(0, 0)

The IPrevNext interface implemented by the adapter simply consist of the previous_entity / next_entity methods, that should respectively return the previous / next entity or None. We make an RQL query to get files in the same folder, ordered similarly (eg by their data_name attribute). We set ascendant/descendant ordering and a strict comparison with current file's name (the "X" variable representing the current file).

Note

  • Former implements selector should be replaced by is_instance or adaptable selector with CubicWeb >= 3.9. In our case, is_instance is used to tell our adapter to get File entities.

Notice that this query supposes we wont have two files of the same name in the same folder. Fixing this is out of the scope of this blog. And as I would like to have at some point a smarter, context sensitive previous/next entity, I'll probably never fix this query (though if I had to, I would probably choose to add a constraint in the schema so that we can't add two files of the same name in a folder).

One more thing: by default, the component will be displayed below the content zone (the one with the white background). You can change this in the site's properties through the ui, but you can also change the default value in the code by modifying the context attribute of the component:

navigation.NextPrevNavigationComponent.context = 'navcontentbottom'

Note

context may be one of 'navtop', 'navbottom', 'navcontenttop' or 'navcontentbottom'; the first two being outside the main content zone, the two others inside it.

screenshot of the previous/next entity component

The previous/next entity component, at the bottom of the main content zone.

Now, the only remaining stuff in my todo list is to see the file's folder. I'll use the standard breadcrumb component to do so. Similarly as what we've seen before, this component is controlled by the IBreadCrumbs interface, so we'll have to provide a custom adapter for File entity, telling the a file's parent entity is its folder:

from cubicweb.web.views import ibreadcrumbs

class FileIBreadCrumbsAdapter(ibreadcrumbs.IBreadCrumbsAdapter):
    __select__ = is_instance('File')

    def parent_entity(self):
        if self.entity.filed_under:
            return self.entity.filed_under[0]

In this case, we simply use the attribute notation provided by the ORM to get the folder in which the current file (e.g. self.entity) is located.

Note

The IBreadCrumbs interface is a breadcrumbs method, but the default IBreadCrumbsAdapter provides a default implementation for it that will look at the value returned by its parent_entity method. It also provides a default implementation for this method for entities adapting to the ITree interface, but as our File doesn't, we've to provide a custom adapter.

screenshot of the breadcrumb component

The breadcrumb component when on a file entity, now displaying parent folder.

Step 4: preparing the release and migrating the instance

Now that greatly enhanced our cube, it's time to release it and to upgrade production site. I'll probably detail that process later, but I currently simply transfer the new code to the server running the web site.

However, there's some commands to get things done properly... First, as I've added some translatable string, I have to run:

$ cubicweb-ctl i18ncube sytweb

To update the cube's gettext catalogs (the '.po' files under the cube's i18n directory). Once the above command is executed, I'll then update translations.

To see if everything is ok on my test instance, I do:

$ cubicweb-ctl i18ninstance sytweb
$ cubicweb-ctl start -D sytweb

The first command compile i18n catalogs (e.g. generates '.mo' files) for my test instance. The second command starts it in debug mode, so I can open my browser and navigate through the web site to see if everything is ok...

Note

In the 'cubicweb-ctl i18ncube' command, sytweb refers to the cube, while in the two other, it refers to the instance (if you can't see the difference, reread CubicWeb's concept chapter !).

Once I've checked it's ok, I simply have to bump the version number in the __pkginfo__ module to trigger a migration once I'll have updated the code on the production site. I can check the migration is also going fine, by first restoring a dump from the production site, then upgrading my test instance.

To generate a dump from the production site:

$ cubicweb-ctl db-dump sytweb
pg_dump -Fc --username=syt --no-owner --file /home/syt/etc/cubicweb.d/sytweb/backup/tmpYIN0YI/system sytweb
-> backup file /home/syt/etc/cubicweb.d/sytweb/backup/sytweb-2010-07-13_10-22-40.tar.gz

I can now get back the dump file ('sytweb-2010-07-13_10-22-40.tar.gz') to my test machine (using scp for instance) to restore it and start migration:

$ cubicweb-ctl db-restore sytweb sytweb-2010-07-13_10-22-40.tar.gz
$ cubicweb-ctl upgrade sytweb

You'll have to answer some questions, as we've seen in an earlier post.

Now that everything is tested, I can transfer the new code to the production server, apt-get upgrade cubicweb 3.9 and its dependencies, and eventually upgrade the production instance.

Conclusion

This is a somewhat long post that starts showing you the way CubicWeb provides a highly configurable user interface, as well as powerful and reusable components. And there are a lot of others like those!

So see you next time for part V, where we'll probably want to do more ui stuff!

blog entry of

CubicWeb Blog