`C/bin/calculator` doesn't work about pyvm

[ForceBru]

At address 0x531d there's a mov instruction whose source operand appears to be merely 0x4, which will read data at exactly that address of the code section! This instruction probably has an override?

	[0x531d]: 0x8b
mov ecx, 0x4=83e4f050
	[0x5320]: 0x8b
mov esi, 0x15994=50e80000
	[0x5324]: 0x29
sub esi=59472, ecx=1357964419
	[0x5326]: 0x89
mov 0x15998, ecx=83e4f050
	[0x532a]: 0x89
mov ecx, esi=cd0310af
	[0x532c]: 0xc1
sar ecx=-1357904947, imm8=31 (ecx := 4294967295)
	[0x532f]: 0x29
sub ebp=3899401284, edi=1
	[0x5331]: 0x19
sub edx=3988, ebx=0
	[0x5333]: 0x39
cmp ebp=3899401283, esi=2937062349
	[0x5335]: 0x19
sub edx=3988, ecx=4294967295
	[0x5337]: 0x7d
jmp rel8 0x538a
	[0x538a]: 0x8b
mov esi, 0x159b0=505b0100
	[0x538e]: 0x8b
mov ebp, 0x15998=83e4f050
	[0x5392]: 0x8b
mov ecx, 0x15994=50e80000
	[0x5396]: 0x89
mov 0x15bb8, ecx=50e80000
	[0x5399]: 0x8b
mov ecx, 0x15b7c=905c0100
	[0x539c]: 0x29
sub ecx=89232, ebp=1357964419
	[0x539e]: 0x89
mov edx, ecx=0d7810af
	[0x53a0]: 0xc1
sar edx=-1357875187, imm8=31 (edx := 4294967295)
	[0x53a3]: 0x01
add edi=1, ecx=2937092109
	[0x53a5]: 0x11
add ebx=0, edx=4294967295
	[0x53a7]: 0x89
mov 0x15bc4, edi=0e7810af
	[0x53aa]: 0x89
mov 0x15bc8, ebx=ffffffff
	[0x53ad]: 0xfb6

[ForceBru]

Nope, the instruction is mov ecx,DWORD PTR [esi+0x4], so esi must be zero at this point.

[ForceBru]

Apparently, __uflow at 0xb98 doesn't restore registers properly? At the beginning it pushes:

push ebx ; 0x00
push edi ; 0x00
push esi; 0x00015b50

And then pops different data:

pop esi ; 0x00 (!!)
pop edi ; 0x00
pop ebx ; 0x00

But it still returns to the correct address (0x5300). So, the top of the stack got corrupted somehow? Or is it intentional? Probably not, because these registers are supposed to be restored after a function call.

[ForceBru]

Basically, a bunch of stosd instructions (probably from address 0x7486) overwrite the first 3 bytes of esi that's stored on the stack. WTF?! Also, shouldn't there be some protection against overwriting the stack?

[ForceBru]

Looks like there was a bug in the C code