Hi,
> Sorry for the delay responding. I am not saying to do this (pseudocode):
>
> package.loaded = clone_table(originalPackageLoaded)
>
> Rather, I am suggesting this (pseudocode):
>
> clear_all_subkeys(package.loaded)
> merge_into_from(package.loaded, originalPackageLoaded)
>
> The first way would break identity with registry value _LOADED. The
second
> would not.*
>
> Referencing it through package.loaded is the right way because it is the
> documented way. The use of _LOADED in the registry is an undocumented
> implementation detail which might change.
Line 652 of lib/loadlib.c:
/* set field `loaded' */
luaL_findtable(L, LUA_REGISTRYINDEX, "_LOADED", 2);
lua_setfield(L, -2, "loaded");
package.loaded just references to the _LOADED in LUA_REGISTRYINDEX. We
could change package.loaded to whatever the hell we want, but internally
Lua just gets the _LOADED table in its REGISTRY (poke around loadlib.c
if you don't believe me). I ran a grep in the Lua source directory and
nowhere does Lua pull package data from package.loaded, it's ALWAYS from
_LOADED. Believe me, I've tried what you are suggesting; a long time
ago. My package.loaded would read correctly when I print it, but I
always get a "package already loaded" message because of the values in
_LOADED -- even if I reload the Lua startup script. I see no other way
to work around this other than changing Lua source code. That is why I
implemented it the way I did in the extension code, modifying _LOADED
directly. If you could do better, by all means please prove me wrong. I
am a novice C++ programmer after all ;) Maybe the way the extension code
re-inits the global scope could be done differently to preserve the
package table.
Here's my thoughts on what is happening.
1. When SciTE starts up, a new Lua state is created and _LOADED is
created in the REGISTRY. Then the extension saves the populated global
namespace for re-merging later when necessary and SciTE does its thing.
2. When different Lua scripts are loaded for different file extensions,
everything is still preserved. The memory addresses (as I have checked)
of the library tables (string, math, package, etc.) are intact. The
"shallow" copy of the global namespace functions as it should when it is
re-merged into the global namespace.
3. When the extension is *reloaded*, the global namespace is cleared and
*reinitialized*. Therefore all memory addresses are different (I have
checked and confirmed this) with the exception of _G naturally.
4. However, _LOADED knows nothing about this *reinitialization* because
the same Lua state is being used. Therefore it must be modified directly
in order to accurately reflect the status of loaded packages.
I will try to experiment more with different things, however I believe I
have already exhausted every conceivable thought as to getting it to
work without modifying _LOADED directly.
Take care,
-Mitchell;
>
> Cheers,
> Bruce
>
> * This assumes we're only making a shallow copy of the global
namespace, so
> the original package.loaded and the new package.loaded refer to the same
> instance. I believe that is the case; it was my intention at least.
If it
> is not working that way, I would call it a bug or at least a misfeature.
>
_______________________________________________
Scite-interest mailing list
[email protected]
http://mailman.lyra.org/mailman/listinfo/scite-interest
