Make sql::Connection::Raze() more robust against corruptions.

Make sql::Connection::Raze() more robust against corruptions.

Corruptions resulting from the first page showing a different database
size than the filesystem shows cause very basic functions to fail.
This does fewer queries against the corrupt database, and also enables
a magic pragma to let it make better progress.

BUG=159490
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=167019

[Scott Hess - ex-Googler, 2012-11-08 01:19:26]

I'm so sad I'm only getting this up just before leaving for the day like 50
hours after I was basically done with it.  Sigh.

[Scott Hess - ex-Googler, 2012-11-09 02:00:26]

Way to test things, Scott!

Unfortunately, while the gist of this is right, getting the auto_vacuum setting
right is going to be annoying.  Android uses SQLITE_DEFAULT_AUTOVACUUM=1, which
means that there's both the default setting, and the setting that the caller
wishes to enable.  I suspect that I'll just change things so that it is only set
when the caller wishes to set it, otherwise it gets the default, and ignore the
case where SQLITE_DEFAULT_AUTOVACUUM=1 and the caller wishes to turn it off.

None of the current cases actually need any of this, more's the pity.

[Scott Hess - ex-Googler, 2012-11-09 02:13:21]

On 2012/11/09 02:00:26, shess wrote:
> Way to test things, Scott!
> 
> Unfortunately, while the gist of this is right, getting the auto_vacuum
setting
> right is going to be annoying.  Android uses SQLITE_DEFAULT_AUTOVACUUM=1,
which
> means that there's both the default setting, and the setting that the caller
> wishes to enable.  I suspect that I'll just change things so that it is only
set
> when the caller wishes to set it, otherwise it gets the default, and ignore
the
> case where SQLITE_DEFAULT_AUTOVACUUM=1 and the caller wishes to turn it off.
> 
> None of the current cases actually need any of this, more's the pity.

OK, punted on the auto_vacuum thing.  I think it would make sense to make it
like page_size, where if you want to set it you call an accessor before open. 
But it's not really necessary right now, so I just added a workaround comment
for anyone who runs across this in the meanwhile.  This should work for android,
too.

[erikwright (departed), 2012-11-09 03:44:03]

LGTM.

[Scott Hess - ex-Googler, 2012-11-09 21:16:07]

Yaron, you made the Android changes to sqlite.gyp, if this passes I'm just going
to commit, but I figured you should be looped in.

[Yaron, 2012-11-09 21:26:10]

Thanks for looping me in. Looks fine.

[commit-bot: I haz the power, 2012-11-09 21:48:45]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/shess@chromium.org/11369126/3012

[commit-bot: I haz the power, 2012-11-10 00:38:28]

Change committed as 167019