LOAD_GLOBAL instruction with wrong source position

[15r10nk]

BPO	47253
Nosy	@15r10nk
Files	test3.py: code which shows the problem

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2022-04-07.21:05:52.344>
labels = ['type-bug', '3.11']
title = 'LOAD_GLOBAL instruction with wrong source position'
updated_at = <Date 2022-04-07.21:05:52.344>
user = 'https://github.com/15r10nk'

bugs.python.org fields:

activity = <Date 2022-04-07.21:05:52.344>
actor = '15r10nk'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = []
creation = <Date 2022-04-07.21:05:52.344>
creator = '15r10nk'
dependencies = []
files = ['50727']
hgrepos = []
issue_num = 47253
keywords = []
message_count = 1.0
messages = ['416945']
nosy_count = 1.0
nosy_names = ['15r10nk']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'open'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue47253'
versions = ['Python 3.11']

[15r10nk]

The LOAD_GLOBAL instruction has a different/wrong source position if it refers the name of an imported module.

[terryjreedy]

You should have pasted in the actual position tuples you saw that show the presumed bug.
With 3.11.0a7, released 10 days ago, just before your report, the positions are identical (on Win 10).

>>> import dis
>>> dix=dis
>>> def f():
...     dis.b()
...     dix.b()
...
>>> for i in dis.Bytecode(f):
...     if i.opname == "LOAD_GLOBAL":
...         print(i.starts_lines, i.positions)
...   
2 Positions(lineno=2, end_lineno=2, col_offset=4, end_col_offset=7)
3 Positions(lineno=2, end_lineno=2, col_offset=4, end_col_offset=7)

I don't know why the Positions.lineno does not match starts_lines, but your report is about columns. Here they match.
I am presuming that you used an earlier alpha and that the OS does not matter and that this was recently discovered and fixed.
Reopen if I missed something

[15r10nk]

@terryjreedy The problem seems is be back in 3.11.0b4 (on ubuntu)

import dis

dix = 5


def f():
    dis.b()  # wrong source range
    dix.b()  # correct source range


d = [bc for bc in dis.Bytecode(f) if bc.opname == "LOAD_GLOBAL"]

print(d)

assert len(d) == 2

# LOAD_GLOBAL has a wrong end column if the global value is an imported module
assert d[0].positions[2:] == d[1].positions[2:] == (4, 7)

output:

[Instruction(opname='LOAD_GLOBAL', opcode=116, arg=1, argval='dis', argrepr='NULL + dis', offset=2, starts_line=7, is_jump_target=False, positions=Positions(lineno=7, end_lineno=7, col_offset=4, end_col_offset=9))
, Instruction(opname='LOAD_GLOBAL', opcode=116, arg=4, argval='dix', argrepr='dix', offset=40, starts_line=8, is_jump_target=False, positions=Positions(lineno=8, end_lineno=8, col_offset=4, end_col_offset=7))]
Traceback (most recent call last):
  File "/home/frank/projects/executing/test2.py", line 18, in <module>
    assert d[0].positions[2:] == d[1].positions[2:] == (4, 7)
AssertionError

can we open this issue again?

[terryjreedy]

You changed from 'global name is an imported module name' to 'global object is an imported module. I revised by example to cover both and ran with freshly updated and compiled 'main' and '3.11' and installed 3.11.0b4'.

import dis
dix = dis
dum = 5
def f(): dis, dix, dum

for i in dis.Bytecode(f):
    if i.opname == "LOAD_GLOBAL":
        print(i.positions)

Output on Win 10:

Positions(lineno=4, end_lineno=4, col_offset=9, end_col_offset=12)
Positions(lineno=4, end_lineno=4, col_offset=14, end_col_offset=17)
Positions(lineno=4, end_lineno=4, col_offset=19, end_col_offset=22)

For all 3 names, end - start == 3. Same with each on separate line.

Please run a pasted copy of the code above on your machine. If output is different, paste a copy.
If same but minimal editing gives anomaly, paste code and output.

[15r10nk]

I got the same results like you and there are no problems if we only lookup the global variable.
Sorry for my wording error. I hope this Example will make it clear what I mean.

The source range extends to the name of the called function if we call an function on a imported module.

import dis
dix = dis
dum = 5
def f():
    dis.a()
    dis.abc()
    dis.a
    dis
    dix.a()
    dum.abc()

for i in dis.Bytecode(f):
    if i.opname == "LOAD_GLOBAL":
        print(i.positions)

output:

Positions(lineno=5, end_lineno=5, col_offset=4, end_col_offset=9)
Positions(lineno=6, end_lineno=6, col_offset=4, end_col_offset=11)
Positions(lineno=7, end_lineno=7, col_offset=4, end_col_offset=7)
Positions(lineno=8, end_lineno=8, col_offset=4, end_col_offset=7)
Positions(lineno=9, end_lineno=9, col_offset=4, end_col_offset=7)
Positions(lineno=10, end_lineno=10, col_offset=4, end_col_offset=7)

you can see that the method name gets included in the range. This should be not the case in my opinion. Or are there any reasons for this Behaviour? There is also no Problem for attribute lookups like dis.a.

I also bisected the bahavior change down to 3011a097bd. Maybe this helps.

[terryjreedy]

@markshannon @pablogsal LOAD_GLOBAL of a name bound by import has an erroneous end_col_offset if followed by LOAD_METHOD and calls. OP bisected to #31933, 3011a097bd. (15... Thanks for doing that.)

Tests in additional to those already posted above:

import dis, abc
import ast as art
dix = dis
ast = art

def f():
    dix.a()  # Is first name affected if not bound name? (No.)
    dis.a()  # Is bound-file name affected if not first? (Yes.)
    abc.a()  # Are names other than 'dis' affected?  (Yes.)
    art.a()  # Is bound name not file name affected? (Yes.)
    ast.a()  # Is file name not bound name affected?  (No.)

for i in dis.Bytecode(f):
    if i.opname == "LOAD_GLOBAL":
        print(f'{i.opname:12s}', i.positions)

outputs

LOAD_GLOBAL  Positions(lineno=8, end_lineno=8, col_offset=4, end_col_offset=7)
LOAD_GLOBAL  Positions(lineno=9, end_lineno=9, col_offset=4, end_col_offset=9)
LOAD_GLOBAL  Positions(lineno=10, end_lineno=10, col_offset=4, end_col_offset=9)
LOAD_GLOBAL  Positions(lineno=11, end_lineno=11, col_offset=4, end_col_offset=9)
LOAD_GLOBAL  Positions(lineno=12, end_lineno=12, col_offset=4, end_col_offset=7)

[terryjreedy]

Thank you Brandt for the nice fix and tests.
@15r10nk Thank you for persisting through the fog so we fixed a real bug which would surely have bitten someone if released.