no_redundant_jumps: Assertion `0' failed

[15r10nk]

Bug report

Bug description:

The following code causes a crash on cpython 3.13 branch (f5289c4) and the current main.

import ast

code="""
try:
    if name_4:
        pass
    else:
        pass
except* name_0:
    pass
else:
    name_4
"""

tree = ast.parse(code)
for node in ast.walk(tree):
    if hasattr(node,"lineno"):
         del   node.lineno
         del   node.end_lineno
         del   node.col_offset
         del   node.end_col_offset

compile(ast.fix_missing_locations(tree), "<file>", "exec")

output (Python 3.13.0a3+):

python: Python/flowgraph.c:511: no_redundant_jumps: Assertion `0' failed.

I bisected the problem down to 2091fb2.
@iritkatriel can you take a look at it?

CPython versions tested on:

3.13, CPython main branch

Operating systems tested on:

No response

Linked PRs

• gh-120367: fix removal of redundant NOPs and jumps after reordering hot-cold blocks #120425
• [3.13] gh-120367: fix removal of redundant NOPs and jumps after reordering hot-cold blocks (GH-120425) #120621
• gh-120367: fix bug where compiler detects redundant jump after pseudo op replacement #120714
• [3.13] gh-120367: fix bug where compiler detects redundant jump after pseudo op replacement (GH-120714) #120716

[15r10nk]

I have also several other code samples which trigger the same assertion like:

try:
    pass
except:
    pass
else:
    match name_2:
        case b'':
            pass
finally:
    something

let me know if you need more of them to fix the issue. (The good news is that I have only samples for this issue 😄)

[aisk]

I can re-produce the same crash with f5289c4 or current main branch(7dd8c37).

[picnixz]

On main (e16aed6), I can reproduce the first crash but not the one given in #120367 (comment). The second one only gives me NameError on name_2 (which is fine).

[15r10nk]

@picnixz sorry I that I did not post the complete example.

import ast

code="""
try:
    pass
except:
    pass
else:
    match name_2:
        case b'':
            pass
finally:
    something
"""

tree = ast.parse(code)
for node in ast.walk(tree):
    if hasattr(node,"lineno"):
         del   node.lineno
         del   node.end_lineno
         del   node.col_offset
         del   node.end_col_offset

compile(ast.fix_missing_locations(tree), "<file>", "exec")

This fails for me also on main

[picnixz]

Oh yes, now it works (I mean, it works in its failure).

[15r10nk]

@iritkatriel this fix solved most of the samples I found, but there are two samples which look similar and are still failing with the current 3.13 branch (71ad34d)

import ast
source='''
if name_2:
    pass
else:
    try:
        pass
    except:
        pass
~name_5
'''
tree = ast.parse(source)
for node in ast.walk(tree):
    if hasattr(node, 'lineno'):
        del node.lineno
ast.fix_missing_locations(tree)
compile(tree, '', 'exec')

output (Python 3.13.0b2+):

python: Python/flowgraph.c:511: no_redundant_jumps: Assertion `0' failed.

and the second:

import ast
source='''
for [] in b'':
    if name_0:
        pass
    else:
        try:
            pass
        except* name_0:
            pass
    name_2
'''
tree = ast.parse(source)
for node in ast.walk(tree):
    if hasattr(node, 'lineno'):
        del node.lineno
ast.fix_missing_locations(tree)
compile(tree, '', 'exec')

output (Python 3.13.0b2+):

python: Python/flowgraph.c:511: no_redundant_jumps: Assertion `0' failed.

Can you take a look at it?

[iritkatriel]

I see, what's happening is that we get:

-1: [EH=0 CLD=0 WRM=1 NO_FT=1 0x108866f20] used: 1, depth: 0, preds: 1 
  [00] line: 1, JUMP_NO_INTERRUPT (257)  target: 0x1088673a0 [21] jump 
6: [EH=0 CLD=0 WRM=1 NO_FT=0 0x1088670a0] used: 1, depth: 0, preds: 1 
  [00] line: 1, SETUP_FINALLY (261)  target: 0x108867340 [10] 
-1: [EH=0 CLD=0 WRM=0 NO_FT=0 0x108866fe0] used: 0, depth: 0, preds: 0 
21: [EH=0 CLD=0 WRM=1 NO_FT=1 0x1088673a0] used: 4, depth: 0, preds: 3 return 
  [00] line: 1, LOAD_NAME (92)  arg: 1 
  [01] line: 1, UNARY_INVERT (40)  
  [02] line: 1, POP_TOP (31)  
  [03] line: 1, RETURN_CONST (103)  arg: 0 

The jump is not identified as redundant because of the SETUP_FINALLY, but that is later removed and then the assertion trips. I'll think how to fix it.

(The jump would not be redundant if it, as well as the SETUP_FINALLY, were not artificially forced to appear to be on the same line as the next instruction).