Re: `require()` semantics?

[Philipp Janda <siffiejoe@...>]

Hi!

Am 25.06.2014 02:59 schröbte Pierre-Yves Gérardy:
> I think I've found a bug in the C loader/seracher of Lua 5.1:
>
> With the current version of require51 found at
> https://gist.github.com/pygy/7d7d0209767a2d1c593a/f10fdaff5f391b0a6ccfde0a6d52408582176051,
> if the sentinel is a table, it ends up being populated by the C loader
> as if it were the module.

The sentinel is not a table but a lightuserdata, so maybe the bug is in 
the current version of `require51` ...
E.g. both `luaL_register` and the `module` function assume that the 
sentinel is _not_ a table (the C searcher has nothing to do with it, 
neither does the C loader unless it calls `luaL_register` with a 
non-NULL library name). Even with a non-table sentinel, `require51` 
doesn't work correctly with a module that uses `module` or 
`luaL_register` and doesn't return a value.

Also, the searcher can be any callable object, but the loader has to be 
a function.

> For eample, requiring lpeg twice triggers the cycle detection code. If
> the sentinel has another type, the error does not occur.
>
> However it is nowhere passed as a parameter to anything... I suppose
> it is left on the stack and coopted by the loader.
>
> —Pierre-Yves

Philipp