 |
lua-l archive |
|
|
||
|
|
> > ===============
> > local a; > > a = function() end; > > =============== > > The code generated was: > > =============== > > 1 [1] LOADNIL 0 0 > > 2 [2] CLOSURE 1 0 ; 0x80049078 > > 3 [2] MOVE 0 1 > > 4 [2] RETURN 0 1 > > =============== > > It seemed that we can not store the closure directly to local var "a". > > Why not keep the expdesc as "VRELOCABLE" and patch the target register > > later? > > The bytecode does store the function in "a", only the comment does > not help you to know that. > > Local variables are mere aliases for VM registers. > > You should use "luac -l -l". That shows you the constants, locals, > upvalues and function prototypes too, not only the disassembly. > > You'll see that "a" is register 0. The closure is generated in > register 1, so the MOVE instruction does store the closure in "a". > Perhaps I didn't describe my question clearly. I knew the relationship between register and local var. What I want to know is the code generation policy at lparser.c line 520. I think that we can eliminate the "MOVE" instruction by removing this "luaK_exp2nextreg" call, and let "luaK_storevar" function to fill register 0 to "CLOSURE" instruction directly. |