Flattening an attribute value fails to quote it satisfactorily

[jonathanj]

Adjusting the value in the attribute tests to include a double quote in the rendered value.

--- a/nevow/test/test_newflat.py
+++ b/nevow/test/test_newflat.py
@@ -478,7 +478,7 @@ class FlattenTests(TestCase, FlattenMixin):
         which was returned by L{IRenderable.render} is replaced with the result
         of the named renderer on the L{IRenderable} which returned the L{Tag}.
         """
-        result = "quux"
+        result = 'qu"ux'
         renders = []
         request = object()
         tag = div(foo=invisible(render="bar"))

Yields this test failure:

twisted.trial.unittest.FailTest: '<div foo="qu"ux"></div>' != '<div foo="quux"></div>'

nevow.test.test_newflat.FlattenTests.test_renderAttributeNestedInAttributeValue

Resilient HTML parsers are able to parse this output but the results are mangled in the vein of <div foo="qu" ux="">. XHTML parsers (such as in use by Athena) refuse to parse this output altogether.

This issue became apparent when trying to render a value with double-quotes into the value attribute of an input element.

[jonathanj]

Brief investigation indicates that the inAttribute flag is not propagated (either from WovenContext or from nested _flatten calls) in the render-directive case, preventing the inAttribute case of escapedData from being used.

[wthie]

Just out of curiosity - can this be reproduced in Python 2? If not then the problem might be in one of the changes introduced by porting to Python 3.

[jonathanj]

@wthie Sorry I should have mentioned that this was discovered and reproduced on Python 2.

[wthie]

No problem, there was another flattener problem recently on the twisted mailing list, flattening still seems to be a tricky thing. I only have very simple use cases for flattening but will look into the problem when starting the chunking of the Python3 port.