Bug 97149 – Fix IndexedDB unit tests

• Status: RESOLVED
• Resolution: FIXED
• Component: New Bugs

[David Grogan]

Fix IndexedDB unit tests

[David Grogan]

Created attachment 164804 [details]
Patch

[David Grogan]

Josh, could you take a look at this?

[Joshua Bell]

Comment on attachment 164804 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=164804&action=review

> Source/WebCore/Modules/indexeddb/IDBDatabaseBackendImpl.cpp:450
> +    if (m_version == NoStringVersion && m_intVersion == IDBDatabaseMetadata::NoIntVersion && m_transactionCoordinator) {

I'm uncomfortable with this. This is basically using m_transactionCoordinator as an "if null, I'm in test mode, so behave differently" flag which doesn't make for a good unit test; the behavior change has nothing to do with the coordinator itself (e.g. skipping a call, etc)

It looks like this introduces a way for tests to bypass the open/upgradeneeded pattern to simplify the test. Could the test call openConnectionWithVersion(cb1, cb2, 0) and get the behavior that way?

[David Grogan]

Comment on attachment 164804 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=164804&action=review

>> Source/WebCore/Modules/indexeddb/IDBDatabaseBackendImpl.cpp:450
>> +    if (m_version == NoStringVersion && m_intVersion == IDBDatabaseMetadata::NoIntVersion && m_transactionCoordinator) {
> 
> I'm uncomfortable with this. This is basically using m_transactionCoordinator as an "if null, I'm in test mode, so behave differently" flag which doesn't make for a good unit test; the behavior change has nothing to do with the coordinator itself (e.g. skipping a call, etc)
> 
> It looks like this introduces a way for tests to bypass the open/upgradeneeded pattern to simplify the test. Could the test call openConnectionWithVersion(cb1, cb2, 0) and get the behavior that way?

So, the behavior change was at least slightly related to the coordinator - it was the null transaction coordinator that caused an invalid memory access if an int version transaction was run.  But you're right about openConnectionWithVersion, it's possible to get the desired behavior by changing just test code, which seems a lot safer.

[David Grogan]

Created attachment 164818 [details]
Patch

[Joshua Bell]

(In reply to comment #4)
> 
> So, the behavior change was at least slightly related to the coordinator - it was the null transaction coordinator that caused an invalid memory access if an int version transaction was run. 

Thanks - I didn't trace it all the way through.

> But you're right about openConnectionWithVersion, it's possible to get the desired behavior by changing just test code, which seems a lot safer.

Cool - I wasn't sure. New patch LGTM (although the -1 magic number might catch someone's attention...)

[David Grogan]

Created attachment 164977 [details]
Patch

[David Grogan]

Nate, could you review this?  jsbell@ gave it the LGTM in c6

[David Grogan]

Nate, never mind, just noticed you're gardening.

Tony, could you review this?

[WebKit Review Bot]

Comment on attachment 164977 [details]
Patch

Clearing flags on attachment: 164977

Committed r129170: <http://trac.webkit.org/changeset/129170>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.