erg/doc/EN/python/bytecode_specification.md

Python bytecode specification

Format

• 0~3 byte(u32): magic number (see common/bytecode.rs for details)
• 4~7 byte(u32): 0 padding
• 8~12 byte(u32): timestamp
• 13~16 byte(u32): 0 padding
• 17~ byte(PyCodeObject): code object

PyCodeObject

• 0 byte(u8): '0xe3' (prefix, this means code's 'c')
• 01~04 byte(u32): number of args (co_argcount)
• 05~08 byte(u32): number of position-only args (co_posonlyargcount)
• 09~12 byte(u32): number of keyword-only args (co_kwonlyargcount)
• 13~16 byte(u32): number of locals (co_nlocals)
• 17~20 byte(u32): stack size (co_stacksize)
• 21~24 byte(u32): flags (co_flags) ()
• ? byte: bytecode instructions, ends with '0x53', '0x0' (83, 0): RETURN_VALUE (co_code)
• ? byte(PyTuple): constants used in the code (co_consts)
• ? byte(PyTuple): names used in the code (co_names)
• ? byte(PyTuple): variable names defined in the code, include params (PyTuple) (co_varnames)
• ? byte(PyTuple): variables captured from the outer scope (co_freevars)
• ? byte(PyTuple): variables used in the inner closure (co_cellvars)
• ? byte(PyUnicode or PyShortAscii): file name, where it was loaded from (co_filename)
• ? byte(PyUnicode or PyShortAscii): the name of code itself, default is <module> (co_name)
• ?~?+3 byte(u32): number of first line (co_firstlineno)
• ? byte(bytes): line table, represented by PyStringObject? (co_lnotab)

PyTupleObject

• 0 byte: 0x29 (means ')')
• 01~04 byte(u32): number of tuple items
• ? byte(PyObject): items

PyStringObject

• If I use a character other than ascii, does it become PyUnicode?

• "あ", "𠮷", and "α" are PyUnicode (no longer used?)

• 0 byte: 0x73 (means 's')

• 1~4 byte: length of string

• 5~ byte: payload

PyUnicodeObject

• 0 byte: 0x75 (means 'u')
• 1~4 byte: length of string
• 5~ byte: payload

PyShortAsciiObject

• This is called short, but even if there are more than 100 characters, this will still short

• or rather, there is no ascii that is not short (is short a data type?)

• 0 byte: 0xFA (means 'z')

• 1~4 byte: length of string

• 5~ byte: payload

PyInternedObject

• interned objects are registered in a dedicated map and can be compared with is

• String, for example, can be compared in constant time regardless of its length

• 0 byte: 0x74 (means 't')

PyShortAsciiInternedObject

• 0 byte: 0xDA (means 'Z')
• 1~4 byte: length of string
• 5~ byte: payload