Stefan Sperling wrote:
> On Sat, Oct 29, 2011 at 06:45:50AM -0500, Jonathan Nieder wrote:
>> [[[
>> The actual ABI compatibility of sqlite3 doesn't depend on the patchlevel
>> (the x in 3.7.x), so stop being picky about the patchlevel when checking
>> the version number at runtime. This avoids spurious errors of the form
>> "svn: Couldn't perform atomic initialization" / "svn: SQLite compiled
>> for 3.7.4, but running with 3.7.3" when sqlite gets a minor update
>> without Subversion being rebuilt to match.
>
> Shouldn't the version numbers in your example be the other way around?
No, my description was just bogus. This is not about when sqlite gets
a minor update, but rather when either
A. sqlite gets downgraded in a minor way, or, more likely,
B. subversion is installed using pre-built binaries that were built
against a slightly newer version of sqlite.
>> --- subversion/libsvn_subr/sqlite.c (revision 1194866)
>> +++ subversion/libsvn_subr/sqlite.c (working copy)
>> @@ -606,7 +606,7 @@ static volatile svn_atomic_t sqlite_init_state = 0
>> static svn_error_t *
>> init_sqlite(void *baton, apr_pool_t *pool)
>> {
>> - if (sqlite3_libversion_number() < SQLITE_VERSION_NUMBER)
>> + if (sqlite3_libversion_number()/1000 < SQLITE_VERSION_NUMBER/1000)
This test was already a '<' rather than '!=' so minor sqlite updates
were already handled fine.
Thanks for the sanity check. I'll sleep on this for now, planning to
resend with a new log message and some other tweaks tomorrow.
Received on 2011-10-29 14:19:38 CEST