On Sat, Sep 5, 2015 at 6:42 PM, Darin Adler <darin at apple.com> wrote:
> Michael is planning a workaround in WebKit that will call
> sqlite3_initialize manually exactly once before WebKit uses sqlite, using
> std::once to deal with the thread safety issue.
>
This reminds me ... I was recently working on a patch which used
sqlite3_config(), which needs to run before sqlite3_initialize(). Since we
were already calling sqlite3_initialize(), I put it in the obvious place.
Then I found a case where someone was making a sqlite3_*() call before
calling into the normal code path. That worked fine for lazy
initialization, but caused my sqlite3_config() to start failing depending
on what happened in which order.
It would be modestly interesting to have a macro SQLITE_REQUIRE_INIT (or
similar) which caused SQLite to throw errors if an API is being called
before sqlite3_initialize(). My pencil implementation would probably
piggyback on SQLITE_OMIT_AUTOINIT, changing those callsites from
sqlite3_initialize() to sqlite3Initialize(), then having that
implementation look something like:
#ifndef SQLITE_OMIT_AUTOINIT
int sqlite3Initialize() {
#ifdef SQLITE_REQUIRE_INIT
return sqlite3GlobalConfig.isInit ? SQLITE_OK : SQLITE_MISUSE;
#else
return sqlite3_initialize();
#endif
}
#endif
Mostly I'd want this in debug and testing builds, so that I could have
confidence in setting SQLITE_OMIT_AUTOINIT for release builds.
Thanks,
scott
