Difference in gas with new module

[kroggen]

When the name_service module is added to Lua on #192 some tests are failing:

=== RUN   TestGasPerFunction
{"level":"info","module":"statesql","size":1024,"time":"2023-09-06T23:57:44Z","message":"set max database size(MB)"}
    vm_dummy_pub_test.go:553: 
        	Error Trace:	/home/runner/work/aergo/aergo/contract/vm_dummy/vm_dummy_pub_test.go:553
        	Error:      	Not equal: 
        	            	expected: 143552
        	            	actual  : 148160
        	Test:       	TestGasPerFunction
        	Messages:   	wrong used gas for unary_ops
    vm_dummy_pub_test.go:553: 
        	Error Trace:	/home/runner/work/aergo/aergo/contract/vm_dummy/vm_dummy_pub_test.go:553
        	Error:      	Not equal: 
        	            	expected: 144482
        	            	actual  : 149090
        	Test:       	TestGasPerFunction
        	Messages:   	wrong used gas for table_ops
--- FAIL: TestGasPerFunction (0.18s)

The gas difference only happens for the above 2 cases

The additional 4608 gas happens on a a single call to lua_gasuse.

I used this code on lj_gas.c / lua_gasuse() to break on the debugger when gas is 4608:

  if (sz == 4608) {
    asm("int3");   // <-- debugger will stop here
  }

The call stack is this:

#0  lua_gasuse (L=0x7fffce807378, sz=4608) at lj_gas.c:54
#1  0x0000000000cc2d60 in lua_gasuse_mul (L=0x7fffce807378, sz=3, n=1536)
    at lj_gas.c:59
#2  0x0000000000cd2e6a in lj_mem_gas (L=0x7fffce807378, osz=0, nsz=1536)
    at lj_gc.c:847
#3  0x0000000000cd2f1a in lj_mem_realloc (L=0x7fffce807378, p=0x0, osz=0, 
    nsz=1536) at lj_gc.c:868
#4  0x0000000000cd4dc0 in newhpart (hbits=6, t=0x7fffce819b30, 
    L=0x7fffce807378) at lj_tab.c:64
#5  lj_tab_resize (L=0x7fffce807378, t=0x7fffce819b30, asize=0, hbits=6)
    at lj_tab.c:276
#6  0x0000000000cd557d in rehashtab (L=0x7fffce807378, t=0x7fffce819b30, 
    ek=0x7fffcdf65c40) at lj_tab.c:385
#7  0x0000000000cd59c6 in lj_tab_newkey (L=0x7fffce807378, t=0x7fffce819b30, 
    key=0x7fffcdf65c40) at lj_tab.c:469
#8  0x0000000000ccc9c3 in lj_BC_TSETS ()
#9  0x0000000000cb256e in lua_pcall (L=0x7fffce807378, nargs=0, nresults=0, 
    errfunc=0) at lj_api.c:1135
#10 0x0000000000ca02e0 in vm_loadcall (L=0x7fffce807378) at vm.c:126
#11 0x0000000000bcac1f in _cgo_f63def7740ae_Cfunc_vm_loadcall (v=0xc00049ef70)
    at /tmp/go-build/cgo-gcc-prolog:529
#12 0x000000000047b204 in runtime.asmcgocall ()
    at /usr/local/go/src/runtime/asm_amd64.s:848
#13 0x00007fffcdf65da0 in ?? ()
#14 0x000000000042086e in runtime.persistentalloc.func1 ()
    at /usr/local/go/src/runtime/malloc.go:1393
#15 0x0000000000479389 in runtime.systemstack ()
    at /usr/local/go/src/runtime/asm_amd64.s:496

It appears that the lj_tab_newkey is making the hash table to be re-hashed (rehashtab)

We need to discover if this is a bug or just an expected consequence of having an additional record on _LOADED table

[gosunuts]

it seems module append to global index table. luajit
i suspect when we just modify local hashtable, it can be affected by global table status.

[gosunuts]

If we can solve this problem, we can add function in modules without a hardfork. ( not change / delete )

[kroggen]

Adding functions to existing modules do not affect gas on V4. This was solved with these changes

[kroggen]

Well, it was tested with just 2 additional functions on #193. Not tested with many

[kroggen]

Adding functions to existing modules is failing again with #269 on V4
So the problem was not solved