nose / TODO

Full commit



 sep. upload of index for 0.10 -- link from 0.9 index
 generate new plugin api docs, highlighting new and changed methods
 generate new docs for all wiki'd docs, flat html, interlinked, for 0.10 doc dir
 generate pydoctor(?) or pudge(?) api docs for 0.10 doc dir


add --plugins option displays loadable plugins, with version (if
  applicable), score, description for each. options for each also if
  verbosity >= 2

rename Failure and split into subclasses for Import and other, and make
it optionally include the name of the file being considered so that
Failure test logs can be more informative

when result prints final result, patch in error classes, eg,

FAILED (todo=10)
OK (skipped=2)

NOTE this can only be done in nose test runner

* loader
  - support module.callable in addition to module:callable names
  - loadTestsFromTestCase -- override to add plugin call

* result/proxy
  - don't use sep proxy obj for each test? -- see under PROFILE


-- when run vs spine suite, makeTest seemingly was called on an object
   that should not have passed the selector --
   sqlalchemy.sql._FunctionGateway, which doesn't match testmatch and
   isn't a testcase subclass -- need to look into it further.


all tests passing as of r203



 deprecated at module/module setup
 error at module/module setup
 plugin api -- comprehensive integration test that runs a suite with a plugin 
   that tracks calls and ensures that all calls are called with proper 
 coverage/doctest -- need tests for coverage/doctest interaction


need to profile -- on 250 tests with discovery, 0.10-dev is ~ 1 second
slower than 0.9. Profile and optimize.

one possiblity -- instead of instantiating a result proxy for every
test, tag each test with a weakref to the that wraps

older notes:
   not sure how this one can be made to work at the module level, since
   import and testrunning take place separately now -- mods a, b, c are all 
   imported before tests are run in a. probably best to have it work at the
   directory and context levels -- that will get pretty close to what it
   does in 0.9

   it can be made to work only in surrounding contexts that are generators:
      hook before
      load and yield suite
      hook after
   needed in 2 places -- loadTestsFromDir, loadTestsFromNames. 
   So loadTestsFromNames will have to become lazy -- what impact will that
     have on unittest compatibility, etc? can it just return a lazysuite?

     the problem with loadTestsFromNames is that context teardown requires
     that for a given context, all tests within the context are known before
     teardown of the first suite within the context. Say you have a package of 
     tests, a, with subpackages a.b and a.c. User wants to run tests a.b.test 
     and a.c.test:

     nosetests a.b.test a.c.test

     'a' and 'b' setup and 'b' teardown run naturally with the 1st test
     'c' setup and teardown run natually with the 2nd test. But when, or
     how, does 'a' teardown run? If all tests are known, then a.c.test can
     run a teardown, because it knows that no more tests in 'a' are going
     to run. But if the 2 tests are loaded lazily -- as is REQUIRED for 
     the isolation plugin -- then 1st test will mistakenly run 'a' teardown, 
     since it doesn't know as it is torn down that another test in the
     same context is about to load.

     can't fix this by a call at the end of the loading process, since then
     context teardowns may interleave -- 'a' and 'b' may be present at once

     can only fix it by forcing full context load & unload around each
     test WHEN USING loadTestFromNames -- this is inefficient, but unavoidable
     without controlling the *order* of tests passed to loadTestsFromNames.