We have two reasonably-supported methods for setting up a development environment:
- An instructor: ggbaker
- A system administrator: sumo
- An advisor: dzhao
- Some Grad students: 0aaagrad, 0bbbgrad, 0cccgrad...
- Some Undergrad students: 0aaa0... 0aaa19, 0bbb0...0bbb19...
Here's some stuff that we almost always tell our students about in the first week of Coursys work.
On any student-facing pages, we should display only students' preferred name. That is generated by either Person.name_pref() (as "John Smith") or Person.sortname_pref (as "Smith, John").
On pages that are staff- or instructor-facing, use the "First (Preferred)" style: Person.name_with_pref() ("Johnathan (John) Smith") or Person.sortname_pref() ("Smith, Johnathan (John)").
As we touch pages that use other styles (in particular, any usage that omits preferred name), they should be updated.
Often, a model can be in one of several states.
A form from our forms system, for example, can be Active, Inactive, Waiting, Stalled, Happy, Sad, Grumpy, or Sleepy.
Generally, statuses start as a list of tuples, like so:
NOTE_CATEGORIES = (
("EXC", "Exceptions"),
("WAI", "Waivers"),
("REQ", "Requirements"),
("TRA", "Transfers"),
("MIS", "Miscellaneous")
)
This list can be fed directly to a CharField in a Model, like so:
category = models.CharField(max_length=3, choices=NOTE_CATEGORIES)
When accessing the model's 'category' field, it will return the three-letter-code,
but you can - for example, in a template - return the 'pretty wording' using the
.get_foo_display property.
note.get_category_display
For every person in the system, there is a coredata.models Person object.
The Role object has a foreign key to the Person object. The Role object describes "one job" that is held by that Person. A Person can have many roles - in the live system, for example, I am one of just about everything, for the purposes of debugging and also occasional larceny.
It is possible to restrict access to a view to just people who have a specific role,
using the @requires_role decorator.
Most of the views you write should have one of these decorators.
@requires_role("ADVS")
def view_student_notes(...):
...
Anybody who attempts to trigger this view without having a Role object of the type "ADVS" foreign-keyed to their account will just get an Access Denied message.
Many things have a coredata.models "Unit" associated with them. This is a way
of subdividing parts of the system by school in the university - Units include
groups like "CMPT", "ENGI", "MSE", corresponding to different logical groups within SFU.
Roles must be created with a Unit. You can't just be an "Advisor", you have to be an "Advisor in CMPT" or an "Advisor in MSE".
The requires_role decorator automatically appends to the request object a
list of units that the current user has this role for, in request.units.
So, for example, if I'm a CMPT Advisor and MSE Advisor, and I'm logged in:
@requires_role("ADVS")
def view_student_notes(self, request):
print request.units
This request.units should contain the "CMPT" and "MSE" Unit model objects.
Only show users the data that they are allowed to see, considering their unit, as follows:
@requires_role("ADVS")
def view_student_notes(self, request):
notes = Notes.objects.filter(unit__in=request.units)
...
We want to keep our URLs as tidy and human-readable as possible.
Bad:
coursys.cs.sfu.ca/forms/923
Good:
coursys.cs.sfu.ca/forms/course_appeal_form
The django-autoslug package has got us covered - when creating a model, we can create an 'autoslug' field, point it at another field in the model, and from that point on the model has a 'slug' field that we can search on.
For more complicated slug logic, we can create an 'autoslug' function.
See /advisornotes/models.py for an example:
class Artifact(models.Model):
name = models.CharField(max_length=140, help_text='The name of the artifact', null=False, blank=False)
category = models.CharField(max_length=3, choices=ARTIFACT_CATEGORIES, null=False, blank=False)
def autoslug(self):
return make_slug(self.unit.label + '-' + self.name)
slug = AutoSlugField(populate_from=autoslug, null=False, editable=False, unique=True)
unit = models.ForeignKey(Unit, help_text='The academic unit that owns this artifact', null=False, blank=False, on_delete=models.PROTECT)
config = JSONField(null=False, blank=False, default=dict) # additional configuration stuff:
Here we're automatically making a slug out of the Artifact's unit (CMPT) and name - "cmpt-thingamajigger".
Never hardcode URLs. Use Django's built in reverse or the url template-tag,
instead.
When building models that may change in the future (pro-tip: this is most models) we often include an empty JSON field in the model.
config = JSONField(null=False, blank=False, default=dict)
When working with the config field, it's considered polite to include any fields that we might be storing in the config field as a comment.
# phonenumber - the user's phone number
# words - words words words
The config field can be treated as an object:
model_object.config["data"] = "harblar"
But before accessing config data, always protect against a KeyError or check if the data exists:
if "data" in model_object.config:
data = model_object.config["data"]
else:
data = "STEVE"
Hidden Fields
We try our very best to never delete anything. Instead, we set a 'hidden' flag, and filter out all 'hidden' variables every time that we pull data from the database.
Here's an example from the Grad subsystem:
hidden = models.BooleanField(null=False, default=False)
req = GradRequirement.objects.filter(program=grad.program, hidden=False)
All actions that modify the database should be logged with a log.models.LogEntry object.
Let's imagine that we have a Model:
CAMPUS_CHOICES = (
"BNBY", "Burnaby",
"SRRY", "Surrey",
"HRBR", "Harbour Centre"
)
class CampusRestaurant(models.Model):
location = models.CharField(max_length=100)
opened = models.DateField()
campus = models.CharField(max_length=4, choices=CAMPUS_CHOICES)
config = JSONField(null=False, blank=False, default=dict)
Now, if we want to take full advantage of Django's forms functionality, we'll have to create a Form:
class CampusRestaurantForm(forms.ModelForm):
location = forms.CharField(max_length=100)
opened = forms.DateField()
campus = ...
okay, I'm going to stop us right there and point out the obvious: we could just be pulling this information out of the Model, right? Right. See:
class CampusRestaurantForm(forms.ModelForm):
class Meta:
model = CampusRestaurant
Here, the ModelForm is converting the CampusRestaurantModel into a CampusRestaurantForm, with all of the validation and POST-processing logic that comes with a Form object. Hooray!
There's a pretty standard view for dealing with a form like this:
@login_required
def new_restaurant(request):
if request.method == 'POST':
form = CampusRestaurantForm(request.POST)
if form.is_valid():
restaurant = form.save(commit=False)
restaurant.save()
messages.add_message(request,
messages.SUCCESS,
u'Restaurant %s created.' % unicode(restaurant))
return HttpResponseRedirect(reverse('campus_eating.views.home'))
else:
form = CampusRestaurantForm()
return render(request, 'campus_eating/new_restaurant.html', {
'form':form})
And with that in place, we can render the form into HTML, inside new_restaurant.html,
using:
{% load form_display %}
{{ form|as_dl }}
Which is good, but when we visit that page, it contains the config field, which we don't want to show to users. And we'd like to handle our DateTimeField with a Calendar, rather than just a standard text field.
We can solve that first problem by adding editable=False as one of the arguments
to config in the Model:
config = JSONField(null=False, blank=False, editable=False, default=dict)
With that in place, config won't show up in any ModelFields.
As for the calendar, there's a Widget for that:
from coredata.widgets import CalendarWidget
class CampusRestaurantForm(forms.ModelForm):
class Meta:
model = CampusRestaurant
widgets = {'opened':CalendarWidget}
There are more useful widgets and fields in coredata.widgets - the OfferingField and PersonField can be used to easily build an Foreign Key to a CourseOffering or Person.
There are a lot of good examples of this in tacontracts/forms.py.
Logic belongs in models before views, and views before templates.
In a quote from Two Scoops of Django - http://twoscoopspress.com/products/two-scoops-of-django-1-6 :
When deciding where to put a piece of code, we like to follow the 'Fat Models, Helper Modules, Thin Views, Stupid Templates' approach.
We recommend that you err on the side of putting more logic into anything but views and templates.
The results are pleasing. The code becomes clearer, more self-documenting, less duplicated, and a lot more reusable.
As for template tags and filters, they should contain the minimum logic possible to function.
Wherever possible, test your application in yourapp/tests.py
These are expected of all code in the system:
- All actions that modify the database should be logged with a new
log.models.LogEntryobject. - All views should have an appropriate permissions decorator (like
@requires_course_staff_by_slug). If not, then be very careful when checking permission in the code. - Where possible, create unit tests in
app/tests.py. Check that the unit tests pass before committing changes. - Have a look at the Style Guide for Python and try to follow it as much as possible.
- If you like to use an IDE, PyCharm, which has free licenses for students/educational staff, it will check for PEP 0008 compliance for you
"Coursys" is the Brand Name of the system. Use it when describing the system. This is the displayed name of the system.
"courses" is the name of the codebase and core libraries.
So, you check out and build courses, which gives you a running instance of CourSys.
Here is a quick description of what is in the project:
- coredata: data on course offerings and enrolment; imported from goSFU or set by admins
- coredata/importer.py: program to do the import from goSFU
- dashboard: responsible for display of front page, main course pages, news items
- discipline: handling academic dishonesty cases
- grades: management of course activities and grades
- groups: handling groups so assignments can be group-based
- log: finer-grained logging for the system
- marking: finer-grained marking with a rubric
- submission: submission of work (for later grading)
- ta: management of TA applications and assignment
- tacontracts: management of TA contracts
- grad: grade student database
- ra: research assistant database and appointments
- advisornotes: advisor notepads for student advising
- discuss: discussion forum for course offerings
- pages: wiki-like course web pages
- onlineforms: configurable and fillable online forms
- visas: Visa management for grads and whoever else we want to manually manage
- oldcode: Dead/unused/unfinished code, including:
- alerts: automated student problem alerts (zombie code. Not used at all anymore)
- planning: planning of courses, assigning instructors, etc. (incomplete and disabled). This one will possibly get finished.
- techreq: collection of technical requirements for courses (incomplete and disabled)
- booking: booking system for advisors (incomplete and disabled)
- gpaconvert: GPA conversion code for grades from other places (incomplete and disabled)
- courselib: various library code for the system
- external: external code used by the system
- machines: Vagrant and Chef files to set up development, testing, and production environments
- media: static web files (CSS, JS, images, etc.)
- server-setup: description of how the production server is configured.
- tools: various scripts that have been used to manage the data as necessary
Other files are the standard Django setup.