Issue 33682: Optimize the bytecode for float(0) ? (original) (raw)

Created on 2018-05-29 12:47 by matrixise, last changed 2022-04-11 14:59 by admin. This issue is now closed.

Messages (5)
msg318016 - (view)	Author: Stéphane Wirtel (matrixise) *	Date: 2018-05-29 12:47
Hi, Maybe already discussed with Victor but I think there is no optimization when we have this simple case for float(X) and int(X) Example: >>> import dis >>> dis.dis("x = float(0)") 1 0 LOAD_NAME 0 (float) 2 LOAD_CONST 0 (0) 4 CALL_FUNCTION 1 6 STORE_NAME 1 (x) 8 LOAD_CONST 1 (None) 10 RETURN_VALUE >>> dis.dis("x = 1 + 1") 1 0 LOAD_CONST 0 (2) 2 STORE_NAME 0 (x) 4 LOAD_CONST 1 (None) 6 RETURN_VALUE >>> dis.dis("x = int(0)") 1 0 LOAD_NAME 0 (int) 2 LOAD_CONST 0 (0) 4 CALL_FUNCTION 1 6 STORE_NAME 1 (x) 8 LOAD_CONST 1 (None) 10 RETURN_VALUE >>> There is an optim for x = 1 + 1
msg318017 - (view)	Author: Stéphane Wirtel (matrixise) *	Date: 2018-05-29 12:48
Victor, if you confirm, maybe you could help me for a mentoring for this issue?
msg318019 - (view)	Author: STINNER Victor (vstinner) *	Date: 2018-05-29 12:50
float and int names can be replaced in the current namespace, so you cannot implement such optimization :-( http://fatoptimizer.readthedocs.io/en/latest/optimizations.html#call-pure http://fatoptimizer.readthedocs.io/en/latest/semantics.html#builtin-functions-replaced-in-the-middle-of-a-function Example in the REPL: >>> float=bool >>> float(0) False >>> int=len >>> int("hello world!") 12 I suggest to close this issue as NOTABUG. You need to implement guards at runtime to implement such optimizations without breaking the Python semantics. It is exactly what I implemented in my FAT Python project: https://faster-cpython.readthedocs.io/fat_python.html
msg318020 - (view)	Author: Steven D'Aprano (steven.daprano) *	Date: 2018-05-29 12:54
I'm sorry, it isn't clear what optimizations for float(X) and int(X) you are referring to. I can only guess that you want to optimize: float(0) to use LOAD_CONST 0.0 instead of calling the float() function. If that is what you want, we can't do that because float() may have been shadowed or replaced. Optimizing 1+1 is safe because it involves only literals and operators, no name look-ups. If that is not what you want, please explain what optimization you are referring to.
msg318021 - (view)	Author: Stéphane Wirtel (matrixise) *	Date: 2018-05-29 13:01
@Victor, Thanks and you are right, int and float are not keywords of Python, in this case, we can override them. @Steven, in fact, the optimization was, when you see float/int (if they are keywords), don't call the function via the bytecode. Thanks and I close this issue.