Django: Odd mark_safe behaviour?

I wrote this little function for writing out HTML tags:

def html_tag(tag, content=None, close=True, attrs={}):
    lst = ['<',tag]
    for key, val in attrs.iteritems():
        lst.append(' %s="%s"' % (key, escape_html(val)))
    if close:
        if content is None: lst.append(' />')
        else: lst.extend(['>', content, '</', tag, '>'])
    else:
        lst.append('>')
    return mark_safe(''.join(lst))

Which worked great, but then I read this article on efficient string concatenation (I know it doesn't really matter for this, but I wanted consistency) and decided to update my script:

def html_tag(tag, body=None, close=True, attrs={}):
    s = StringIO()
    s.write('<%s'%tag)
    for key, val in attrs.iteritems():
        s.write(' %s="%s"' % (key, escape_html(val)))
    if close:
        if body is None: s.write(' />')
        else: s.write('>%s</%s>' % (body, tag))
    else:
        s.write('>')
    return mark_safe(s.getvalue())

But now my HTML get escaped when I try to render it from my template. Everything else is exactly the same. It works properly if I replace the last line with return mark_safe(unicode(s.getvalue())). I checked the return type of s.getvalue(). It should be a str, just like the first function, so why is this failing??

Also fails with SafeString(s.getvalue()) but succeeds with SafeUnicode(s.getvalue()).

---

I'd also like to point out that I used return mark_safe(s.getvalue()) in a different function with no odd behavior.

---

The "call stack" looks like this:

class Input(Widget):
    def render(self):
        return html_tag('input', attrs={'type':self.itype, 'id':self.id,
            'name':self.name, 'value':self.value, 'class':self.itype})
class Field:
    def __unicode__(self):
        return mark_safe(self.widget.render())

And then {{myfield}} is in the template. So it does get mark_safed'd twice, which I thought might have been the problem, but I tried removing that too..... I really have no idea what's causing t开发者_如何学JAVAhis, but it's not too hard to work around, so I guess I won't fret about it.

---

The render method of your widget is called by the BoundField.__unicode__ function, which returns SafeString instead (of a subclass of unicode.)

Many places in Django (e.g. django.template.VariableNode.render) will actually call force_unicode on the field instance itself. This will have the effect of doing unicode(instance.__unicode__()), so even though instance.__unicode__() returns a SafeString object, it will become a regular unicode object.

To illustrate, have a look at the snippet below:

from django.utils.encoding import force_unicode
from django.utils.safestring import mark_safe

class Foo(object):
    def __unicode__(self):
        return mark_safe("foo")

foo = Foo()
print "foo =", type(foo)

ufoo = unicode(foo)
print "ufoo =", type(ufoo)

forced_foo = force_unicode(foo)
print "forced_foo =", type(forced_foo)


bar = mark_safe("bar")
print "bar =", type(bar)

forced_bar = force_unicode(bar)
print "forced_bar =", type(forced_bar)

Output:

foo = <class 'testt.Foo'>
ufoo = <type 'unicode'>
forced_foo = <type 'unicode'>
bar = <class 'django.utils.safestring.SafeString'>
forced_bar = <class 'django.utils.safestring.SafeUnicode'>