"gettext()" vs "gettext_lazy()" in Django

I have a question about using ugettext and gettext_lazy() for translations. I learned that in models I should use gettext_lazy(), while in views ugettext. But are there any other places, where I should use gettext_lazy() too? What about form definitions? Are there any performance diffrences between them?

Edit: And one more thing. Sometimes, instead of gettext_lazy(), gettext_noop() is used. As documentation says, gettext_noop() strings are only marked for translation and translated at the latest possible momment before displaying them to the user, but I'm little confused here, isn't that similar to what gettext_lazy() do? It's still hard for me to decide, which should I use in my models and forms.

gettext() vs. gettext_lazy()

In definitions like forms or models you should use gettext_lazy because the code of this definitions is only executed once (mostly on django's startup); gettext_lazy translates the strings in a lazy fashion, which means, eg. every time you access the name of an attribute on a model the string will be newly translated-which totally makes sense because you might be looking at this model in different languages since django was started!

In views and similar function calls you can use gettext without problems, because everytime the view is called gettext will be newly executed, so you will always get the right translation fitting the request!

Regarding gettext_noop()

As Bryce pointed out in his answer, this function marks a string as extractable for translation but does return the untranslated string. This is useful for using the string in two places – translated and untranslated. See the following example:

import logging
from django.http import HttpResponse
from django.utils.translation import gettext as _, gettext_noop as _noop

def view(request):
    msg = _noop("An error has occurred")
    logging.error(msg)
    return HttpResponse(_(msg))