Re: The life cycle of userdata values?

[Philipp Janda <siffiejoe@...>]

Am 26.07.2016 um 09:24 schröbte Marc Balmer:
> > Am 25.07.2016 um 21:03 schrieb Roberto Ierusalimschy <roberto@inf.puc-rio.br>:
> >
> > > > p = lua_newuserdate(L, sizeof int)
> > > > *p = 42
> > > >
> > > > /* Do a Lua API call that might error (longjmp) out */
> > >
> > > This has the risk of the memory being invalidated by the
> > > garbage collector.
> >
> > No, if the userdata is properly anchored (e.g, in the stack). Just
> > leave the the userdata in the stack (where lua_newuserdate put it)
> > and everything should work fine.
>
>
> Thanks to all that replied.  After I sprinkled some lua_checkstack()
> calls, the crashes went away.  Time for a release...

Testing with a Lua build that has assertions enabled and/or with a tool 
like valgrind helps to minimize the chances that you missed a place 
where a `luaL_checkstack()` would be needed.

> Out of curiosity, why does lua_newuserdata() not call lua_checkstack()
> internally?  When calling lua_newuserdata(), isn't it quite obvious
> that you want the result on the stack?

I'd say performance and consistency. For simple functions like 
`lua_pushnumber()` it's performance. For complex functions like 
`lua_newtable()` or `lua_newuserdata()` (which allocate memory anyway) 
it's consistency with the simple functions. (One exception to this is 
`lua_pushfstring()`, btw.)

> - mb

Philipp