All:
I noticed the following when using sqlite to timestamp flags in an
embedded system. I will lay out the tests performed and the results I
got.
First, an overview. I am working with a linux 2.6.26 kernel and
sqlite v3.5.0. For the handling of timezones, I use the zoneinfo
files. /etc/localtime
is a symbolic link to /var/tmp/localtime. At boot the proper timezone
file is copied to /var/tmp/localtime (the reason for this setup is
/etc resides
in a read only partition. I timestamp each flag using 'Insert into
flags (timestamp, FLAGDATA...) values (datetime('now','localtime),
FLAGDATA)'.
Now on to my issue.
--Scenario A--
Boot System
Update /var/localtime with correct zoneinfo file
Write Flag(s)
------------------------------------------------------------------
Result.... All time stamps match my localtime.
--Scenario B--
Boot System
Write 1 Flag
Update /var/localtime with correct zoneinfo file
Write a few flags (5-10)
....Some time elapses....
Write some more flags
-------------------------------------------------------------------
Result...
The first flag ,written before /var/localtime, is written in UTC. I
expect this since the symbolic link does not yet point to a valid
zoneinfo file.
The next few flags (5-10 depending on what the system is doing) are
also in UTC, these flags are the same flags written
in Scenario A after the zoneinfo update.
The system may briefly stop writing flags, then write some more due to
a button press or something.
Eventually the flags receive the correct timestamp for my timezone
settings. I think, but cant say for certain, that the
timestamp becomes 'correct' after a small amount of time elapses.
So this email isn't blaming SQLite for my improper timestamps even
after I set the correct zoneinfo file. I am just trying to
understand what is going on. Should I assume that my observation of
'Set zoneinfo first, then use datetime functions' is
a valid fix? Or am I fooling myself?
I just wanted to know the mechanism at work here. Of course I do
fully acknowledge this is could be an artifact in the
GLIBC library. If someone knows the particular reason this is
happening, even if it is GLIBC fault, I would love to know.
My last bit of information which may be relevant is all flags are
written in individual transactions.
Thanks in advance,
Rich
_______________________________________________
sqlite-users mailing list
sqlite-users@sqlite.org
http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
