1854873 - browser_invalidDbFolderLoad.js isn't failing. Should it?

Status: NEW

(Thunderbird :: General, task)

Description

[Ben Campbell]

I've got a rewrite of the folder parsing code (as part of my big mbox refactor), but it's causing a mochitest to fail.
The test is pretty small and simple:

browser_invalidDbFolderLoad.js

My new folder parsing is more strictly asynchronous that the current implementation.
I have a suspicion that the test is relying on the current implementation being a little more synchronous than it should be...

The be_in_folder() call triggers a folder.updateFolder() call, which in turn kicks off a folder reparse (which is supposed to be fully asynchronous).

In the existing code, be_in_folder returns after the folder parse is complete.

But with my changes, the folder parse is still running when be_in_folder returns, and the test then fails in assert_messages_in_view - reading folder.msgDatabase fails with an NS_MSG_ERROR_FOLDER_SUMMARY_OUT_OF_DATE (which indicates the parse is still running).

So.
Does be_in_folder have some special mechanism to ensure it waits until the folder parse triggered by UpdateFolder() has completed?
Or is the test actually wrong, and it's just a lucky quirk of the existing code which means the folder parse completes before be_in_folder returns?

Comment 1

[Geoff Lankow (:darktrojan)]

(In reply to Ben Campbell from comment #0)

Does be_in_folder have some special mechanism to ensure it waits until the folder parse triggered by UpdateFolder() has completed?
Or is the test actually wrong, and it's just a lucky quirk of the existing code which means the folder parse completes before be_in_folder returns?

Looks like it. This code calls updateFolder then nothing happens until we get notified about it finishing. This doesn't really answer the question, I got a bit sidetracked.

Comment 2

[Geoff Lankow (:darktrojan)]

I think that it's just luck. be_in_folder waits for folderURIChanged which would happen after updateFolder is called, but not necessarily before it calls back to say it's finished. assert_messages_in_view should fail if it's not finished.

I tried raising the number of messages used by the test to slow it down but didn't see any change in behaviour. Maybe the current code really is sync.

Comment 3

[Geoff Lankow (:darktrojan)]

Oh here we go: Uncaught exception in test bound test_load_folder_with_invalidDB - [Exception... "Component returned failure code: 0x80550005 [nsIMsgFolder.msgDatabase]". The test doesn't like 30000 messages.

Comment 4

[Ben Campbell]

(In reply to Geoff Lankow (:darktrojan) from comment #3)

Oh here we go: Uncaught exception in test bound test_load_folder_with_invalidDB - [Exception... "Component returned failure code: 0x80550005 [nsIMsgFolder.msgDatabase]". The test doesn't like 30000 messages.

Cool! That's exactly the error I get when I have my new folder-parsing code in place, which has a fastidiously async folder ParseFolder() implementation.