Revert statement that is breaking sqlite on OSX.

[CyrusNajmabadi]

No description provided.

[CyrusNajmabadi]

Tagging @ericsink @Therzok @sharwell @AArnott

[CyrusNajmabadi]

Not sure what the OSX issue is here. On @Therzok to continue the investigation with @ericsink and Sqlitepcl.

[Therzok]

I think I figured it out, but not sure how to validate it works properly: ericsink/SQLitePCL.raw#407

[davidwengier]

Thank you @CyrusNajmabadi and @Therzok for tracking this down ❤️

[AArnott]

Sad, since this reverts a good change. I hope whatever the underlying bug is can be fixed and this workaround reverted soon.

[ghost]

Auto-approval

[sharwell]

Can you make the change conditional on the current operating system (i.e. avoid changing the behavior for Windows)?

[ericsink]

Just in case others have not seen it, ericsink/SQLitePCL.raw#407 contains a writeup of what's going on here.

[CyrusNajmabadi]

@sharwell done.

[ghost]

Auto-approval

[ericsink]

Discussion of this issue is happening on the sqlite forum at:

https://sqlite.org/forum/forumpost/de445cf241

Notable quote from DRH: "Trying to open an on-disk database using "mode=memory" doesn't make sense. Where were you trying to accomplish?"

He also reports that he was unable to repro the differing behavior, but I'm not yet sure why.

[ericsink]

Continuing to parrot things from the sqlite forum thread:

If "mode=memory" is removed, which DRH said does not make sense, then the behavior across Windows and other platforms no longer differs. In other words, it gives the error in all cases.

[CyrusNajmabadi]

@AArnott for info here. This was added in response to what the platform cache is doing.

[AArnott]

@ericsink Thanks for referencing the sqlite thread. I responded with a defense of what we're doing, pointing to the docs that explain why this is legit.

[CyrusNajmabadi]

Merging this in now. If we get resolution in https://sqlite.org/forum/forumpost/9babfc9cba?t=h as to what to do here, we can update accordingly.