Implements DatabaseManager::OpenDatabase

Adds Database2MetaData helper class which manages database2 files and their associated version metadata.

[Dimitri, 2008-05-26 14:13:28]

This is exciting! The module is starting really come together.

http://codereview.appspot.com/1145/diff/1/3
File database2/connection.cc (right):

http://codereview.appspot.com/1145/diff/1/3#newcode148
Line 148: 
Should we make the poisoning/marking-as-corrupt bit into its own method? This
would be checked for after a call to any SQLite function that returns a result
code.

http://codereview.appspot.com/1145/diff/1/7
File database2/database2_metadata.cc (right):

http://codereview.appspot.com/1145/diff/1/7#newcode61
Line 61: return false;
Can BuildSqliteErrorString from sqlite_wrapper be used here?

http://codereview.appspot.com/1145/diff/1/7#newcode264
Line 264: int new_version_cookie = version.empty() ? 0 : 1;
Why change the value here? When the database is created, the user_version will
be 0 regardless.

http://codereview.appspot.com/1145/diff/1/2
File test/testcases/database2_tests.js (right):

http://codereview.appspot.com/1145/diff/1/2#newcode63
Line 63: assertError(function() {
I think that's logical. A user should probably do changeVersion on an
unversioned database if you want to assigned it a version.

[Aaron, 2008-05-26 19:27:52]

Ready for another look.

http://codereview.appspot.com/1145/diff/1/3
File database2/connection.cc (right):

http://codereview.appspot.com/1145/diff/1/3#newcode148
Line 148: 
On 2008/05/26 14:13:29, Dimitri wrote:
> Should we make the poisoning/marking-as-corrupt bit into its own method? This
> would be checked for after a call to any SQLite function that returns a result
> code.

Poisoning and marking corrupt are currently separate in database1. Poisoning is
done after every sqlite function, but marking corrupt is only done after open.
I'm not sure what the reasoning is here, and whether we should maintain it.
Let's start here since this is what database1 did.

http://codereview.appspot.com/1145/diff/1/7
File database2/database2_metadata.cc (right):

http://codereview.appspot.com/1145/diff/1/7#newcode61
Line 61: return false;
On 2008/05/26 14:13:29, Dimitri wrote:
> Can BuildSqliteErrorString from sqlite_wrapper be used here?

Done.

http://codereview.appspot.com/1145/diff/1/7#newcode264
Line 264: int new_version_cookie = version.empty() ? 0 : 1;
On 2008/05/26 14:13:29, Dimitri wrote:
> Why change the value here? When the database is created, the user_version will
> be 0 regardless.

Ok, fair enough, there is no need to distinguish these two choices.

However, you say "when the database is created, the user_version will be...". I
think that ideally, we should create the initial user version with whatever this
code returns, just so that we only have it in one place.

http://codereview.appspot.com/1145/diff/1/2
File test/testcases/database2_tests.js (right):

http://codereview.appspot.com/1145/diff/1/2#newcode63
Line 63: assertError(function() {
On 2008/05/26 14:13:29, Dimitri wrote:
> I think that's logical. A user should probably do changeVersion on an
> unversioned database if you want to assigned it a version.

Ok, I've raised this on whatwg@. Will leave the TODO until I hear back.

[Dimitri, 2008-05-26 21:27:10]

LGTM to non-build-merge-related changes.