spell3 is broken on windows

[Richard Boulton]

Spell3 is now skipped on windows because it tries to create two writable databases simultaneously at the same path, deleting but not closing the first before creating the second. The delete fails on windows due to the differing file semantics. However, the test shouldn't be trying to do this anyway. I've just added the following explanatory comment to the testcase code, reproduced here in case it gets separated from the source code in future:

FIXME: the following two lines create two writable databases, but at the same path. The first database is deleted, but kept open, before the second is opened. This isn't really supported behaviour (in Olly's words: "it'll end in tears if it tries to flush"), and doesn't work on windows (hence the SKIP_TEST for WIN32 above) so we need to fix this. This probably involves making the backendmanager support a "name" parameter for get_writable_database() which specifies the path to create the database at. This would also make it easier to get a readonly handle on a previously created writable database.

[Olly Betts]

I've overhauled BackendManager, so this should now be fairly easy to sort out.

[Olly Betts]

Marking for 1.0.5.

[Olly Betts]

Should now be fixed, but I can't test currently as the win32 buildbots aren't available currently.

[Olly Betts]

The windows buildbot doesn't seem to be working still, so Cc-ing Charlie.

Charlie: Does "apitest spell3" work on MS Windows with recent SVN HEAD?

[Charlie Hull]

No, it gives the following error:

Running test: spell3....flint\dbw\flintlock - The process cannot access the file

because it is being used by another process.

.flint\dbw\postlist.DB - The process cannot access the file because it is being used by another process. .flint\dbw\record.DB - The process cannot access the file because it is being us ed by another process. .flint\dbw\termlist.DB - The process cannot access the file because it is being used by another process.

DatabaseOpeningError

apitest completed test run: 0 tests passed, 1 failed.

In fact (partly due to the new test harness) there are around 40 test failures on SVN_HEAD at the moment, I'm going to take a look soon.

[Olly Betts]

OK, stupid mistake - spell3 should work now.

I don't know about the rest...

[Charlie Hull]

Yep, spell3 works now. Still looking at the others.