Hi. I just looked over this and fixed one test failure in Five. In general all tests pass now, when run from the eggified Zope in: svn://svn.zope.org/repos/main/Zope2.buildout/trunk I guess that the zope.testbrowser problems are a mix of test interdependencies and maybe a differing mechanize version. The eggified Zope2 does not run tests for its dependencies but only for itself, so a lot less tests are run. Secondly the normal Zope2 trunk pulls in mechanize as a SVN checkout from the Zope repository. The eggified version has a normal egg dependency and pulls in the later mechanize 0.1.9. I have no idea what version that mechanize in our SVN is actually. Tres Seaver wrote:
Any chance you can get these worked out? I looked at the Products.Five tests today, with the following observations:
- Not as many tests fail if I run '-s Products.Five' as when I run the whole set: in particular, the first two in 'aqlegacy_ftest.txt' ('HTTPError', 'falcon.pt' not called) don't occur.
- The next 'aqlegacy_ftest.txt' failure is because the content provider is not being acquisition wrapped. The remaining failure is likely due to the same cause (it is on the following line).
This was caused by a change by Malthe, who forgot to update the tests. Fixed now.
- All the failing tests in Products.Five come from code which relies on zope.testbrowser, via the Products.Five.testbrowser shim.
- The first failing zope.testbrowser test (README.txt line 1162) seems to be due to a change somewhere which contradicts the comment just below the example: "Here, the body is left in place because it isn't form data."
- The testing requirements for zope.testbrowser itself are really snarled up: it pulls in basically all of Zope3.
As Benji figured out, these are really hard to trim down. I think it will cost us some more month of work before we have the zope.app dependency soup cleaned up, if we ever get there at all. Hanno