Change by Dennis Sweeney :
Removed file: https://bugs.python.org/file49672/twoway_demo.py
___
Python tracker
<https://bugs.python.org/issue41972>
___
___
Python-bug
Change by Dennis Sweeney :
Added file: https://bugs.python.org/file49674/twoway_demo.py
___
Python tracker
<https://bugs.python.org/issue41972>
___
___
Python-bugs-list m
Dennis Sweeney added the comment:
This is just how local/nonlocal/global/builtin variables work in Python.
When you assign to a name anywhere inside of a function, all occurrences of
that name refer by default to a local variable. So the line "ZeroDivisionError
= 1" tells the foo(
Dennis Sweeney added the comment:
To make it slightly more readable and future-proof so such things don't become
outdated again in the future, you could use the CodeType.replace() method.
See also https://bugs.python.org/issue42422
--
nosy: +Dennis Sw
Dennis Sweeney added the comment:
This is not a bug. See https://stackoverflow.com/a/10825998/11461120
--
nosy: +Dennis Sweeney
___
Python tracker
<https://bugs.python.org/issue42
Dennis Sweeney added the comment:
What commands did you enter? Are you following the instructions at
https://devguide.python.org/setup/ ?
--
nosy: +Dennis Sweeney
___
Python tracker
<https://bugs.python.org/issue42
Dennis Sweeney added the comment:
Were you able to run PCbuild\get_externals.bat ?
--
___
Python tracker
<https://bugs.python.org/issue42804>
___
___
Python-bug
New submission from Dennis Sweeney :
# common case
.\python.bat -m pyperf timeit "type(17)"
Master: 49.9 ns +- 1.8 ns ---> PR: 33.3 ns +- 1.4 ns
# uncommon case
.\python.bat -m pyperf timeit "type('A', (object, ), {})"
Master: 5.14 us +- 0.
Change by Dennis Sweeney :
--
keywords: +patch
pull_requests: +22892
stage: -> patch review
pull_request: https://github.com/python/cpython/pull/24058
___
Python tracker
<https://bugs.python.org/issu
Dennis Sweeney added the comment:
PR 22904 now adds a text document explaining how the Two-Way algorithm works in
much more detail.
I was looking at more benchmarking results, and I came to a couple of
conclusions about cutoffs.
There's a consistent benefit to using the two-way algo
Dennis Sweeney added the comment:
It looks like this was a duplicate of https://bugs.python.org/issue31356
--
___
Python tracker
<https://bugs.python.org/issue41
Dennis Sweeney added the comment:
https://bugs.python.org/issue41545 is a duplicate of this.
In that report, there's an example of something that can go wrong with the
save-a-boolean-per-context-manager approach even when threads are not used, but
when concurrent generators are used,
Dennis Sweeney added the comment:
I reproduced something similar in 3.9 with standard library modules:
>>> from datetime import datetime
>>> from zoneinfo import ZoneInfo
>>> zone = ZoneInfo(key='Etc/GMT-3')
>>> zone.tzname(datetime(2021, 1
Dennis Sweeney added the comment:
I think this is the expected behavior.
It is expected by IEEE 754 that nan != nan, so that behavior exists:
>>> nan = float('nan')
>>> nan2 = float('nan')
>>> assert nan != nan2
>>> assert nan != nan
Dennis Sweeney added the comment:
This is the expected behavior.
>From
>https://docs.python.org/3/tutorial/introduction.html#first-steps-towards-programming
"""
In Python, like in C, any non-zero integer value is true; zero is false. The
condition may also be a str
Dennis Sweeney added the comment:
You could express this as:
a = [input() for i in range(10)]
x = input()
print(x in a)
This is more clear IMO, because if you want to have something happen before
something else, it's clearest to put them on separate lines, one after the
other.
I
Dennis Sweeney added the comment:
This was a very intentional change from the commit
3bcbedc9f1471d957a30a90f9d1251516b422416
It's not safe to check `x is y` when x and y are strings.
You should always use `x == y` for strings instead.
In CPython, if the names x and y both refer to the
Dennis Sweeney added the comment:
There may be reason to re-open this.
With Python 3.9.0, I see the inconsistent behavior that Gary describes only
when using IDLE. So this is likely an IDLE issue.
I attached a screenshot of the difference.
--
Added file: https://bugs.python.org
Dennis Sweeney added the comment:
I think the strangeness is happening because sometimes, the warning is printed
to stderr, while other times, IDLE's parser notices the "is "
anti-pattern and raises a SyntaxError.
See the attached screenshot for the IDLE output versus the con
Dennis Sweeney added the comment:
I bisected the change to here:
https://github.com/python/cpython/pull/19881
commit 3dd2157febae5087cad24f69b6de9cbd13cd
Author: Raymond Hettinger
Date: Sun May 3 04:51:05 2020 -0700
Simplify set entry insertion logic. (GH-19881)
""&qu
Dennis Sweeney added the comment:
This seems like a very specific use case. Too specific IMO for a method on all
string objects for anyone using Python anywhere in the world. Why not just
write a function like this?
def my_join(strings, sep=", ", last_sep=", and &quo
Dennis Sweeney added the comment:
If I understand correctly, changing the -1 to a -2 does not actually make the
program "crash" -- you just only see one black circle.
The reason is that range(40, 0, -2) produces 40, 38, 36, etc., all of which are
even numbers, so rad % 2 is always
Dennis Sweeney added the comment:
Any chance PR 22904 can make it into 3.10 before the May 03 feature freeze? The
text document in that PR has some notes on how the algorithm works, so that may
be a good place to start reviewing if anyone is interested
Change by Dennis Sweeney :
--
pull_requests: +23457
pull_request: https://github.com/python/cpython/pull/24672
___
Python tracker
<https://bugs.python.org/issue41
Dennis Sweeney added the comment:
If the argument clinic is too disruptive, would it be okay to inline the
equivalent code like this?
diff --git a/Modules/_collectionsmodule.c b/Modules/_collectionsmodule.c
index 90bafb0ea8..d75388abc8 100644
--- a/Modules/_collectionsmodule.c
+++ b/Modules
Change by Dennis Sweeney :
--
pull_requests: +23564
pull_request: https://github.com/python/cpython/pull/24796
___
Python tracker
<https://bugs.python.org/issue41
Change by Dennis Sweeney :
--
keywords: +patch
nosy: +Dennis Sweeney
nosy_count: 2.0 -> 3.0
pull_requests: +19523
stage: -> patch review
pull_request: https://github.com/python/cpython/pull/20236
___
Python tracker
<https://bugs.p
Dennis Sweeney added the comment:
The attached PR isn't exactly what you requested, but it's a very minimal code
change that uses the existing __qualname__ functionality to change the message
to
TypeError: A.foo() takes 1 positional argument but 2 were given
Does that add
Dennis Sweeney added the comment:
While trying to write tests, I stumbled across something interesting:
_PyObject_FunctionString as discussed here ( https://bugs.python.org/issue37645
) returns a string that also includes the module name where applicable. For
example, the module name is
Dennis Sweeney added the comment:
I got this:
>>> class A:
... def f():
... pass
...
>>> A.f(1)
Traceback (most recent call last):
File "", line 1, in
TypeError: A.f() takes 0 positional arguments but 1 was given
>>> A.f(**{1:2})
Traceback (most rec
Dennis Sweeney added the comment:
Never mind; I think you're right, and
https://github.com/python/cpython/blob/master/Objects/call.c#L1009 is the line.
--
___
Python tracker
<https://bugs.python.org/is
Dennis Sweeney added the comment:
I just ran the entire test suite with:
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -4179,6 +4179,7 @@ _PyEval_EvalCode(PyThreadState *tstate,
Py_ssize_t j;
if (keyword == NULL || !PyUnicode_Check(keyword)) {
+printf("THIS
Dennis Sweeney added the comment:
Sure -- I'll file the issue.
--
___
Python tracker
<https://bugs.python.org/issue40679>
___
___
Python-bugs-list m
New submission from Dennis Sweeney :
When I was looking into https://bugs.python.org/issue40679, I couldn't come up
with a test case for the following block, so I added a print statement:
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -4179,6 +4179,7 @@ _PyEval_EvalCode(PyThreadState *t
Dennis Sweeney added the comment:
https://bugs.python.org/issue40706
--
___
Python tracker
<https://bugs.python.org/issue40679>
___
___
Python-bugs-list mailin
New submission from Dennis Sweeney :
One of the tests (test_ttk_guionly.test_variable_change) on the Ubuntu CI is
intermittently hanging on this code:
https://github.com/python/cpython/blob/e42b705188271da108de42b55d9344642170aa2b/Lib/tkinter/test/test_ttk/test_extensions.py#L147
Dennis Sweeney added the comment:
key_and_reverse.py employs the same strategy as winners.py, but uses lists as
the nodes of the tree rather than using Node instances. It also eliminates the
recursion of treeify, and adds (with neither much of a performance hit nor much
code duplication
Dennis Sweeney added the comment:
Wouldn't Floyd's or Brent's cycle detection algorithms be better here than the
allocation of a new set? I believe they might also eliminate the need to
fast-path the first 100 or however many.
As in: https://en.wikipedia.org/wiki/
Dennis Sweeney added the comment:
less_movement.py is my favorite so far. It still handles key and reverse,
but using instance attributes instead of the list indices I tried before.
It does this by only advancing the "key" and "leaf" attributes up toward
the root (where
Change by Dennis Sweeney :
--
pull_requests: +19782
stage: needs patch -> patch review
pull_request: https://github.com/python/cpython/pull/20539
___
Python tracker
<https://bugs.python.org/issu
Dennis Sweeney added the comment:
I believe PR 20539 solves the more general problem (using Floyd's Tortoise and
Hare Algorithm), and I would appreciate review / some more ideas for test cases.
--
___
Python tracker
<https://bugs.py
Dennis Sweeney added the comment:
> I it related to issue25782?
Yes -- I didn't see that issue. I'm a little confused about the resolution of
that issue though.
For clarification, the existing behavior on master:
When trying to raise the exception H,
F -> G -
Dennis Sweeney added the comment:
For clarification, the existing behavior on master:
When trying to raise the exception H,
F -> G -> H -> I -> NULL
becomes
H -> F -> G -> NULL
But when trying to set the exception A on top of
Change by Dennis Sweeney :
--
pull_requests: +19788
pull_request: https://github.com/python/cpython/pull/20539
___
Python tracker
<https://bugs.python.org/issue25
Change by Dennis Sweeney :
--
pull_requests: +19794
pull_request: https://github.com/python/cpython/pull/20550
___
Python tracker
<https://bugs.python.org/issue38
Dennis Sweeney added the comment:
PR 20550 uses a linked structure like what we've been talking about.
--
___
Python tracker
<https://bugs.python.org/is
Dennis Sweeney added the comment:
It looks like only the first 200 characters of the input string's repr are used
as the compiled pattern's repr for some reason:
https://github.com/python/cpython/blob/master/Modules/_sre.c#L1294
I don't know if there is a good reason, e
Dennis Sweeney added the comment:
I'd be happy to write a PR.
Method names could be "mapping", "target", "target_mapping", "target_dict",
"referent_dict", etc.
I like the choice of "target_mapping":
d = dict()
Dennis Sweeney added the comment:
I think this will also require typing.MappingProxyType to change a bit, since
it would make a proxy's underlying dict accessible:
>>> d = dict()
>>> proxy = MappingProxyType(d)
>>> type(proxy.items()) is ty
Change by Dennis Sweeney :
--
keywords: +patch
pull_requests: +19906
stage: -> patch review
pull_request: https://github.com/python/cpython/pull/20691
___
Python tracker
<https://bugs.python.org/issu
Dennis Sweeney added the comment:
Indeed, with PR 20691 applied, the following crashes:
>>> vars(str).items().mapping.clear()
>>> "uh oh"
--
___
Python tracker
<htt
Dennis Sweeney added the comment:
Would the best way to address this be adding new KeysProxy, ValuesProxy, and
ItemsProxy types?
--
___
Python tracker
<https://bugs.python.org/issue40
Dennis Sweeney added the comment:
What about returning another dict_items instead of a set? As in (using the
convention `d.items().mapping is d`):
dict_items = type({}.items())
def __xor__(self: dict_items, other):
if isinstance(other, dict_items):
new
Change by Dennis Sweeney :
--
keywords: +patch
pull_requests: +19928
stage: -> patch review
pull_request: https://github.com/python/cpython/pull/20718
___
Python tracker
<https://bugs.python.org/issu
Dennis Sweeney added the comment:
PR 20718 helps somewhat by only creating and hashing the tuples that wind up in
the final set. Here's a benchmark:
-m pyperf timeit -s "d1 = {i:i for i in range(100_000)}; d2 = {i:i|1 for i in
range(100_000)}" "d1.items() ^ d2.items()&
Dennis Sweeney added the comment:
A demo:
>>> class Int(int):
... hash_calls = 0
... def __hash__(self):
... Int.hash_calls += 1
... return super().__hash__()
...
>>> left = {Int(1): -1, Int(2): -2, Int(3): -3, Int(4): -4, Int(5): -5, Int(6):
Change by Dennis Sweeney :
--
keywords: +patch
pull_requests: +19954
pull_request: https://github.com/python/cpython/pull/20749
___
Python tracker
<https://bugs.python.org/issue40
Dennis Sweeney added the comment:
PR 20749 gives each dict view access to a mappingproxy for the original dict,
although I don't know if that defeats the original purpose.
It might be hard to sensibly make MappingProxy(d).items() return something
other than d.items(), since this is al
Dennis Sweeney added the comment:
Here's a workaround that's possible with PR 20749 applied:
>>> d = {"a":1, "b":2} # fill up the dict...
>>> DICT = object()
>>> d[DICT] = d
>>> items = d.items()
>>
Dennis Sweeney added the comment:
Would it be better to have a dictview.mapping() method rather than an
attribute, since it constructs a new object of a different type and since
that's what keys(), values(), and items() are?
--
___
Python tr
Change by Dennis Sweeney :
--
nosy: +Dennis Sweeney
nosy_count: 3.0 -> 4.0
pull_requests: +20038
pull_request: https://github.com/python/cpython/pull/20845
___
Python tracker
<https://bugs.python.org/issu
Dennis Sweeney added the comment:
I just added PR 20845, but I'm concerned about performance. I'm attaching the
results of a run of pyperformance before and after PR 20845.
--
Added file: https://bugs.python.org/file49228/pushpo
Change by Dennis Sweeney :
Added file: https://bugs.python.org/file49229/master_perf.txt
___
Python tracker
<https://bugs.python.org/issue40925>
___
___
Python-bug
Change by Dennis Sweeney :
Added file: https://bugs.python.org/file49230/perf_diff.txt
___
Python tracker
<https://bugs.python.org/issue40925>
___
___
Python-bugs-list m
Change by Dennis Sweeney :
Removed file: https://bugs.python.org/file49229/master_perf.txt
___
Python tracker
<https://bugs.python.org/issue40925>
___
___
Python-bug
Change by Dennis Sweeney :
Removed file: https://bugs.python.org/file49228/pushpop_perf.txt
___
Python tracker
<https://bugs.python.org/issue40925>
___
___
Python-bug
Change by Dennis Sweeney :
--
pull_requests: +20061
stage: resolved -> patch review
pull_request: https://github.com/python/cpython/pull/20873
___
Python tracker
<https://bugs.python.org/issu
Dennis Sweeney added the comment:
I think the word "attribute" is preferred over "value" because "value" can mean
just about anything, whereas, according to
https://docs.python.org/3/glossary.html?highlight=glossary , an attribute is
specifically:
""
Dennis Sweeney added the comment:
I reproduced something similar on Python 3.9.0b1, Windows 64-bit version:
py -m pyperf timeit -s "import threading; E = threading.Event()"
"E.wait()"
NUMBERMean +- std dev
---
0.0
New submission from Dennis Sweeney :
Benchmarked on my machine (Windows 10):
.\python.bat -m pyperf timeit -s "from collections import deque; x = [[], [1]]
* 1_000_000" "deque(map(bool, x), maxlen=0)"
--- Win32 build configuration ---
Master: 105
Change by Dennis Sweeney :
--
keywords: +patch
pull_requests: +20723
stage: -> patch review
pull_request: https://github.com/python/cpython/pull/21581
___
Python tracker
<https://bugs.python.org/issu
Dennis Sweeney added the comment:
More microbenchmarks:
pyperf timeit "bool()"
Before: 63.1 ns +- 0.7 ns
After: 51.7 ns +- 1.2 ns
pyperf timeit "bool(0)"
Before: 77.4 ns +- 1.9 ns
After: 67.2 ns +- 1.3 ns
pyperf timeit "bool(17)"
Befor
Dennis Sweeney added the comment:
This is exactly the motivation for context managers, no? I attached no_gc.py,
which works when nested and should additionally be thread-safe. Usage:
from no_gc import no_gc
with no_gc():
# collection disabled
with no_gc():
# collection is
Dennis Sweeney added the comment:
The save-a-boolean-for-each-context-manager approach has an issue if used with
concurrent generators, where the lifetimes of two generator objects might be
overlapping but not completely nested, as shown below. The same issue should
arise when using
Dennis Sweeney added the comment:
FWIW I forgot the gc.disable() line in the contextmanager, but what I said
still applies.
--
___
Python tracker
<https://bugs.python.org/issue41
Change by Dennis Sweeney :
--
keywords: +patch
nosy: +Dennis Sweeney
nosy_count: 3.0 -> 4.0
pull_requests: +21055
stage: -> patch review
pull_request: https://github.com/python/cpython/pull/21945
___
Python tracker
<https://bugs.p
Dennis Sweeney added the comment:
PR 21945 changes the signature:
- defaultdict(default_factory[, ...])
+ defaultdict(default_factory=None, /, [...])
--
___
Python tracker
<https://bugs.python.org/issue41
Dennis Sweeney added the comment:
If we were to do this, I think a better API might be to accept an arbitrary
iterable, then produce a sorted iterable:
def sorted_on_disk(iterable, key=None, reverse=False) -> Iterable:
...
It would sort chunks of the input and store them in files
Dennis Sweeney added the comment:
Attached is a proof of concept.
--
Added file: https://bugs.python.org/file49436/disksort.py
___
Python tracker
<https://bugs.python.org/issue41
Dennis Sweeney added the comment:
Here's a reproducer.
--
nosy: +Dennis Sweeney
Added file: https://bugs.python.org/file49447/reproducer.py
___
Python tracker
<https://bugs.python.org/is
Dennis Sweeney added the comment:
I believe the block stack is only for catching exceptions. Loops do not
interact with the block stack. Only SETUP_FINALLY adds to the block stack:
https://docs.python.org/3/library/dis.html#opcode-SETUP_FINALLY
Meanwhile, loops and conditionals and the like
New submission from Dennis Sweeney :
I got these benchmarks:
.\python.bat -m pyperf timeit "float(0)"
Before: Mean +- std dev: 79.0 ns +- 1.0 ns
After: Mean +- std dev: 51.5 ns +- 1.6 ns
--
components: Interpreter Core
messages: 377590
nosy: Dennis Sweeney
priority: norma
Change by Dennis Sweeney :
--
keywords: +patch
pull_requests: +21463
stage: -> patch review
pull_request: https://github.com/python/cpython/pull/22432
___
Python tracker
<https://bugs.python.org/issu
Dennis Sweeney added the comment:
Superseded by https://bugs.python.org/issue41870
--
nosy: -larry
resolution: -> works for me
stage: patch review -> resolved
status: open -> closed
___
Python tracker
<https://bugs.python.or
Dennis Sweeney added the comment:
Adding some hasty printf-debugging to fastsearch.h, I see this:
>>> with open('data.bin', 'rb') as f:
... s = f.read()
...
>>> base = 15403807 * b'\xff'
>>> longer = base + b
Dennis Sweeney added the comment:
Indeed, this is just a very unlucky case.
>>> n = len(longer)
>>> from collections import Counter
>>> Counter(s[:n])
Counter({0: 9056995, 255: 6346813})
>>> s[n-30:n+30].replace(b'
Dennis Sweeney added the comment:
I agree that skip could could do 1 better.
---
> I don't know whether it "should be" applied
I don't think I'm convinced: the second check fixes only the very specific case
when s[len(p):].startswith(p).
Perturbations of repro
Dennis Sweeney added the comment:
> Offhand do you know what the _best_ timing for two-way search is in a
> pattern-not-found case?
Looking at the glibc implementation, in the top-level "else" clause
(for when the string isn't completely periodic),
we have:
Dennis Sweeney added the comment:
Here is a C implementation of the two-way algorithm that should work as a
drop-in replacement for Objects/stringlib/fastsearch.h.
Benchmarking so far, it looks like it is a bit slower in a lot of cases. But
it's also a bit faster in a some other case
Change by Dennis Sweeney :
--
pull_requests: +21650
stage: -> patch review
pull_request: https://github.com/python/cpython/pull/22679
___
Python tracker
<https://bugs.python.org/issu
Dennis Sweeney added the comment:
PR 22679 is a draft that does the two-way algorithm but also adds both of the
tricks from Fredrik's implementation: a bit-set "bloom filter" and remembering
the skip-distance between some pair of characters.
--
Added file: https://
Dennis Sweeney added the comment:
I used random_bench.py to compare PR 22679 to Master, and the results are in
bench_results.txt. Results were varied. I suppose this depends on what cases we
want to optimize for.
--
Added file: https://bugs.python.org/file49512/random_bench.py
Dennis Sweeney added the comment:
bench_table.txt gives my results (`ref` is Master, `change` is with PR 22679).
The change gives 342 faster cases and 275 slower cases, and 9 cases with no
change.
I chose a random word of length 10**6 with a zipf character distribution for
the haystack
Dennis Sweeney added the comment:
Another algorithmic possibility: Instead of the bitset, we could have a
stack-allocated
uint8_t jump[32]; // maybe 64? Maybe uint16_t?
It would say this: If the last character lined up in the haystack is congruent
to i mod (1 << 8), then jump ah
Dennis Sweeney added the comment:
@Tim I got this again for that benchmark:
length=3442, value=ASXABCDHAB...: Mean +- std dev: 2.39 ms +- 0.01 ms
Unfortunately not a ghost.
--
___
Python tracker
<https://bugs.python.org/issue41
Dennis Sweeney added the comment:
That test needle happened to end with a G and not have another G until much
earlier. The status quo took advantage of that, but the PR only takes advantage
of the skip value for a certain middle character. Perhaps it could do both
Dennis Sweeney added the comment:
The most recent batch of commits added a jump table.
Between master and PR 22679 now, there are 151 cases slower than master and 463
that faster than master.
The slower cases are at most twice as slow, but the faster cases are often
10-20x faster.
I could
Dennis Sweeney added the comment:
I'm doing a couple more timing tests to try to understand exactly when the
cutoff should be applied (based on some combination of needle and haystack
lengths).
Can the rolling hash algorithm be made to go sublinear like O(n/m)? It looked
like it was p
Dennis Sweeney added the comment:
I added the cutoff for strings >= 10 characters, and I converted the PR from a
draft to "Ready to Review."
When running stringbench.py before and after the PR, I get these results:
Summary:
Unicode Before: 81.82 Bytes Before: 92.62
Dennis Sweeney added the comment:
> But there _also_ seem to be real (but much smaller) benefits
> for the "search backward" cases, which I don't recall seeing
> when I tried it. Do you have a guess as to why?
I did change `skip = mlast - 1;` to `skip = mlast;` as yo
401 - 500 of 516 matches
Mail list logo
