Re: Question regarding the XML parser binding in PIL

[Doug Currie <doug.currie@...>]

On Dec 17, 2008, at 1:24 AM, Sam Roberts wrote:
> On Sun, Dec 7, 2008 at 12:32 AM, Florian Weimer <fw@deneb.enyo.de>  
> wrote:
> > > When C is called from lua, it must use the L it was passed. If C  
> > > calls
> > > back into lua, after lua returns to it, it must use the L it was
> > > originally passed. Still. Even if lua had called back into C. This  
> > > may
> > > not be easy to arrange, and until coroutines are used creatively,  
> > > bugs
> > > may never be seen.
> > >
> > > What was the question?
> >
> > I wonder if there's a canonical way to avoid these problems, perhaps
> > involving lua_newthread.
>
> I've never seen any, and more threads are likely to make more of a  
> problem.

In lsqlite3, http://luaforge.net/projects/luasqlite/ --  which is not  
the subject of this thread (to overload that word even further), but  
which implements the same glue as the discussed wrapper lua-sqlite3,  
we use the following approach:

When the database's userdata handle is created we record the caller's  
L in the database's userdata handle.

When a Lua function (user function, busy handler, progress handler,  
trace callback) is provided to SQLite it is provided in the context of  
a database handle; that associated database handle is recorded in the  
function's userdata handle. The function's userdata handle is the  
(void *) data we give SQLite along with our C callback wrapper function.

When a callback is executed SQLite gives us the function's userdata  
handle we provided; that is used to get the associated database's  
userdata handle, and it's L is used to execute (lua_pcall) the Lua  
function.

So, the rule is: as long as the Lua thread that created the database  
handle remains active, the callback can be used from any Lua thread.

e