Re: [ptxdist] [PATCH] sqlite: version bump and improved build options

[Clemens Gruber <clemens.gruber@pqgruber.com>]

On Wed, Jun 15, 2016 at 08:46:37AM +0200, Marc Kleine-Budde wrote:
> On 06/15/2016 07:08 AM, Michael Olbrich wrote:
> > On Wed, Jun 15, 2016 at 12:21:19AM +0200, Clemens Gruber wrote:
> >> - Bump the version to 3.13.0: http://sqlite.org/releaselog/3_13_0.html
> >> - Using FTS4 also enables FTS3, they share most of their code and FTS4
> >>   is recommended for new applications. (Better query performance)
> >> - Include the JSON1 functions
> >> - The lookaside memory allocator should not be disabled, it is the
> >>   default allocator since 3.6.1 and is much quicker. 50KB per database
> >>   connection are used for the memory pool but this can be changed at
> >>   runtime, if a smaller memory footprint is required.
> >> - SECURE_DELETE should not be the default, those who need it can always
> >>   enable it at runtime with the secure_delete pragma. That build option
> >>   only controls the default value.
> > 
> > Marc, you introduced these options. Can you comment?
> 
> Sorry, I don't remember why I switched on these options.

Maybe it was a usecase with extremely low memory footprint and security
critical data stored in the database?
Or the lookaside allocator was deemed too new and not stable enough back
then.

Today, I see no reason to disable the lookaside allocator. It causes
less latency and is only a little heavier on memory. Also if the slots
are full or for larger allocations, SQLite falls back to malloc anyway.

Secure delete has a significant performance penalty. Users can always
opt-in with PRAGMA secure_delete = 1; if they need to zero out deleted
content.
Also, due to the limited number of erase cycles on flash storage, I
would rather encrypt the data than zero it out. (SQLCipher extension!)
Then I'd only have to throw away the key. 

Clemens

_______________________________________________
ptxdist mailing list
ptxdist@pengutronix.de