Use a `_PyStackRef` and defer the reference to `f_executable` when
possible. This avoids some reference count contention in the common case
of executing the same code object from multiple threads concurrently in
the free-threaded build.
* gh-116608: Apply style and compatibility changes from importlib_metadata.
* gh-121735: Ensure module-adjacent resources are loadable from a zipfile.
* gh-121735: Allow all modules to be processed by the ZipReader.
* Add blurb
* Remove update-zips script, unneeded.
* Remove unnecessary references to removed static fixtures.
* Remove zipdata fixtures, unused.
POSIX allows errno to be negative.
Even though all currently supported platforms have non-negative errno,
relying on a quirk like that would make Python less portable.
* Raise PicklingError instead of UnicodeEncodeError, ValueError
and AttributeError in both implementations.
* Chain the original exception to the pickle-specific one as __context__.
* Include the error message of ImportError and some AttributeError in
the PicklingError error message.
* Unify error messages between Python and C implementations.
* Refer to documented __reduce__ and __newobj__ callables instead of
internal methods (e.g. save_reduce()) or pickle opcodes (e.g. NEWOBJ).
* Include more details in error messages (what expected, what got).
* Avoid including a potentially long repr of an arbitrary object in
error messages.
This switches the main pyrepl event loop to always be non-blocking so that it
can listen to incoming interruptions from other threads.
This also resolves invalid display of exceptions from other threads
(gh-123178).
This also fixes freezes with pasting and an active input hook.
Improve import time of `socket` by writing `socket.errorTab`
as a constant and lazy import modules.
Co-authored-by: Pieter Eendebak <pieter.eendebak@gmail.com>
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
Co-authored-by: Gregory P. Smith <greg@krypto.org>
Co-authored-by: Hugo van Kemenade <1324225+hugovk@users.noreply.github.com>
Co-authored-by: Shantanu <12621235+hauntsaninja@users.noreply.github.com>
Increases the multiprocessing connection buffer size from 8k to 64k for efficiency, without overallocating.
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
Co-authored-by: Victor Stinner <vstinner@python.org>
Add PyConfig_Get(), PyConfig_GetInt(), PyConfig_Set() and
PyConfig_Names() functions to get and set the current runtime Python
configuration.
Add visibility and "sys spec" to config and preconfig specifications.
_PyConfig_AsDict() now converts PyConfig.xoptions as a dictionary.
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
* Remove backtracking when parsing tarfile headers
* Rewrite PAX header parsing to be stricter
* Optimize parsing of GNU extended sparse headers v0.0
Co-authored-by: Kirill Podoprigora <kirill.bast9@mail.ru>
Co-authored-by: Gregory P. Smith <greg@krypto.org>
* urljoin() with relative reference "?" sets empty query and removes fragment.
* Preserve empty components (authority, params, query, fragment) in urljoin().
* Preserve empty components (authority, params, query) in urldefrag().
Also refactor the code and get rid of double _coerce_args() and
_coerce_result() calls in urljoin(), urldefrag(), urlparse() and
urlunparse().
When checking if the registering browser is the "OS preferred browser", do not use a substring search - that makes no sense: one can have a preferred browser that looks like a super-string of a known browser, e.g. "firefox-nightly" vs "firefox".
https://github.com/python/cpython/issues/108172 explains in more detail, and lays out a potential better future enhancement for this case of just using xdg-open. We'll go with this for now.
---------
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
See https://github.com/python/cpython/issues/121313 for analysis, but this greatly reduces memory overallocation and overhead when multiprocessing is sending non-small data over its pipes between processes.
The `zip_next` function uses a common optimization technique for methods
that generate tuples. The iterator maintains an internal reference to
the returned tuple. When the method is called again, it checks if the
internal tuple's reference count is 1. If so, the tuple can be reused.
However, this approach is not safe under the free-threading build:
after checking the reference count, another thread may perform the same
check and also reuse the tuple. This can result in a double decref on
the items of the replaced tuple and a double incref (memory leak) on
the items of the tuple being set.
This adds a function, `_PyObject_IsUniquelyReferenced` that
encapsulates the stricter logic necessary for the free-threaded build:
the internal tuple must be owned by the current thread, have a local
refcount of one, and a shared refcount of zero.
* Make `weakref.WeakSet` safe against concurrent mutations while it is being iterated.
`_IterationGuard` is no longer used for `WeakSet`, it now relies on copying the underlying set which is an atomic operation while iterating so that it can be modified by other threads.
Per feedback from Paul Moore on GH-123158, it's better to defer making
`Path.delete()` public than ship it with under-designed error handling
capabilities.
We leave a remnant `_delete()` method, which is used by `move()`. Any
functionality not needed by `move()` is deleted.
These two methods accept an *existing* directory path, onto which we join
the source path's base name to form the final target path.
A possible alternative implementation is to check for directories in
`copy()` and `move()` and adjust the target path, which is done in several
`shutil` functions. This behaviour is helpful in a shell context, but
less so in a stored program that explicitly specifies destinations. For
example, a user that calls `Path('foo.py').copy('bar.py')` might not
imagine that `bar.py/foo.py` would be created, but under the alternative
implementation this will happen if `bar.py` is an existing directory.
When display lines above the cursor come from the cache, the first line
to not come from the cache may be a wrapped line, starting half way
through a logical line in the buffer. Detect and handle this case to
avoid accidentally drawing a stray prompt in the middle of a logical
line.
Add a `Path.move()` method that moves a file or directory tree, and returns a new `Path` instance pointing to the target.
This method is similar to `shutil.move()`, except that it doesn't accept a *copy_function* argument, and it doesn't check whether the destination is an existing directory.
* pass the original string error message from the ftplib error to URLError()
* Update request.py
Change error string for ftp error to be consistent with other errors reported for ftp
* Add NEWS entry for change to urllib.request for ftp errors.
* Track the change in the ftp error message in the test.
Check that the current default heap is initialized in
`_mi_os_get_aligned_hint` and `mi_os_claim_huge_pages`.
The mimalloc function `_mi_os_get_aligned_hint` assumes that there is an
initialized default heap. This is true for our main thread, but not for
background threads. The problematic code path is usually called during
initialization (i.e., `Py_Initialize`), but it may also be called if the
program allocates large amounts of memory in total.
The crash only affected the free-threaded build.
`Path.read_bytes()` is used to read a whole file. buffering /
BufferedIO is focused around making small, possibly interleaved,
read/write efficient which doesn't add value in this case.
On my Mac, running the benchmark:
```python
import pyperf
from pathlib import Path
def read_all(all_paths):
for p in all_paths:
p.read_bytes()
def read_file(path_obj):
path_obj.read_bytes()
all_rst = list(Path("Doc").glob("**/*.rst"))
all_py = list(Path(".").glob("**/*.py"))
assert all_rst, "Should have found rst files"
assert all_py, "Should have found python source files"
runner = pyperf.Runner()
runner.bench_func("read_file_small", read_file, Path("Doc/howto/clinic.rst"))
runner.bench_func("read_file_large", read_file, Path("Doc/c-api/typeobj.rst"))
```
before:
```python
.....................
read_file_small: Mean +- std dev: 6.80 us +- 0.07 us
.....................
read_file_large: Mean +- std dev: 10.8 us +- 0.2 us
````
after:
```python
.....................
read_file_small: Mean +- std dev: 5.67 us +- 0.05 us
.....................
read_file_large: Mean +- std dev: 9.77 us +- 0.52 us
```
Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com>
Co-authored-by: Hugo van Kemenade <1324225+hugovk@users.noreply.github.com>
This replaces the existing hashlib Blake2 module with a single implementation that uses HACL\*'s Blake2b/Blake2s implementations. We added support for all the modes exposed by the Python API, including tree hashing, leaf nodes, and so on. We ported and merged all of these changes upstream in HACL\*, added test vectors based on Python's existing implementation, and exposed everything needed for hashlib.
This was joint work done with @R1kM.
See the PR for much discussion and benchmarking details. TL;DR: On many systems, 8-50% faster (!) than `libb2`, on some systems it appeared 10-20% slower than `libb2`.
As of 529a160 (gh-118204), building with HAVE_DYNAMIC_LOADING stopped working. This is a minimal fix just to get builds working again. There are actually a number of long-standing deficiencies with HAVE_DYNAMIC_LOADING builds that need to be resolved separately.
Co-authored-by: Adam Turner <9087854+AA-Turner@users.noreply.github.com>
Co-authored-by: Hugo van Kemenade <1324225+hugovk@users.noreply.github.com>
Co-authored-by: Alyssa Coghlan <ncoghlan@gmail.com>
Rename `pathlib.Path.copy()` to `_copy_file()` (i.e. make it private.)
Rename `pathlib.Path.copytree()` to `copy()`, and add support for copying
non-directories. This simplifies the interface for users, and nicely
complements the upcoming `move()` and `delete()` methods (which will also
accept any type of file.)
Co-authored-by: Adam Turner <9087854+AA-Turner@users.noreply.github.com>
This reverts commit dcc028d924 and
commit 6c54e5d721.
Keep the deprecated logging warn() method in Python 3.13.
Co-authored-by: Gregory P. Smith <greg@krypto.org>
Co-authored-by: Hugo van Kemenade <1324225+hugovk@users.noreply.github.com>
We were not properly accounting for interpreter memory leaks at
shutdown and had two sources of leaks:
* Objects that use deferred reference counting and were reachable via
static types outlive the final GC. We now disable deferred reference
counting on all objects if we are calling the GC due to interpreter
shutdown.
* `_PyMem_FreeDelayed` did not properly check for interpreter shutdown
so we had some memory blocks that were enqueued to be freed, but
never actually freed.
* `_PyType_FinalizeIdPool` wasn't called at interpreter shutdown.
Return -1 and set an exception on error; return 0 if the iterator is
exhausted, and return 1 if the next item was fetched successfully.
Prefer this API to PyIter_Next(), which requires the caller to use
PyErr_Occurred() to differentiate between iterator exhaustion and errors.
Co-authered-by: Irit Katriel <iritkatriel@yahoo.com>
Fix _PyArg_UnpackKeywordsWithVararg for the case when argument for
positional-or-keyword parameter is passed by keyword.
There was only one such case in the stdlib -- the TypeVar constructor.
Frames of methods in code and codeop modules was show with non-default
sys.excepthook.
Save correct tracebacks in sys.last_traceback and update __traceback__
attribute of sys.last_value and sys.last_exc.
Rename `pathlib.Path.rmtree()` to `delete()`, and add support for deleting
non-directories. This simplifies the interface for users, and nicely
complements the upcoming `move()` and `copy()` methods (which will also
accept any type of file.)
Fix PyEval_GetLocals() to avoid SystemError ("bad argument to
internal function"). Don't redefine the 'ret' variable in the if
block.
Add an unit test on PyEval_GetLocals().
The free-threaded build partially stores heap type reference counts in
distributed manner in per-thread arrays. This avoids reference count
contention when creating or destroying instances.
Co-authored-by: Ken Jin <kenjin@python.org>
Modifies the handling of stdout/stderr redirection on Android to accomodate
the rate and buffer size limits imposed by Android's logging infrastructure.
Match statements in tooling require a more recent Python. Tools/cases_generator/*.py (and `Tools/jit/*.py` in 3.13+).
Co-authored-by: Erlend E. Aasland <erlend.aasland@protonmail.com>
Co-authored-by: Gregory P. Smith <greg@krypto.org>
As per C11 DR#471, ctanh (0 + i NaN) and ctanh (0 + i Inf) should return
0 + i NaN (with "invalid" exception in the second case). This has
corresponding implications for ctan(z), as its errors and special cases
are handled as if the operation is implemented by -i*ctanh(i*z).
This patch fixes cmath's code to do same.
Glibs patch: https://sourceware.org/git/?p=glibc.git;a=commitdiff;h=d15e83c5f5231d971472b5ffc9219d54056ca0f1
As per C11 DR#471 (adjusted resolution accepted for C17), cacosh (0 +
iNaN) should return NaN ± i pi/2, not NaN + iNaN. This patch
fixes cmath's code to do same.
The `PyStructSequence` destructor would crash if it was deallocated after
its type's dictionary was cleared by the GC, because it couldn't compute
the "real size" of the instance. This could occur with relatively
straightforward code in the free-threaded build or with a reference
cycle involving the type in the default build, due to differing orders
in which `tp_clear()` was called.
Account for the non-sequence fields in `tp_basicsize` and use that,
along with `Py_SIZE()`, to compute the "real" size of a
`PyStructSequence` in the dealloc function. This avoids the accesses to
the type's dictionary during dealloc, which were unsafe.
On recent versions of macOS (sometime between Catalina and Sonoma 14.5), the default Hovertip foreground color changed from black to white, thereby matching the background. This might be a matter of matching the white foreground of the dark-mode text. The unreadable result is shown here (#120083 (comment)).
The foreground and background colors were made parameters so we can pass different colors for future additional hovertips in IDLE.
---------
Co-authored-by: Terry Jan Reedy <tjreedy@udel.edu>
This flag was added as an escape hatch in gh-91401 and backported to
Python 3.10. The flag broke at some point between its addition and now.
As there is currently no publicly known environments that require this,
remove it rather than work on fixing it.
This leaves the flag in the subprocess module to not break code which
may have used / checked the flag itself.
discussion: https://discuss.python.org/t/subprocess-use-vfork-escape-hatch-broken-fix-or-remove/56915/2
Currently, idle-dev@python.org and idle-dev mailing list
serve to collect spam (90+%). Change About IDLE to direct
discussions to discuss.python.org. Users are already
doing so.
## Encode header parts that contain newlines
Per RFC 2047:
> [...] these encoding schemes allow the
> encoding of arbitrary octet values, mail readers that implement this
> decoding should also ensure that display of the decoded data on the
> recipient's terminal will not cause unwanted side-effects
It seems that the "quoted-word" scheme is a valid way to include
a newline character in a header value, just like we already allow
undecodable bytes or control characters.
They do need to be properly quoted when serialized to text, though.
## Verify that email headers are well-formed
This should fail for custom fold() implementations that aren't careful
about newlines.
Co-authored-by: Bas Bloemsaat <bas@bloemsaat.org>
Co-authored-by: Serhiy Storchaka <storchaka@gmail.com>
* Authenticate socket connection for `socket.socketpair()` fallback when the platform does not have a native `socketpair` C API. We authenticate in-process using `getsocketname` and `getpeername` (thanks to Nathaniel J Smith for that suggestion).
Co-authored-by: Gregory P. Smith <greg@krypto.org>
* Use compensated summation for complex sums with floating-point items.
This amends #121176.
* sum() specializations for floats and complexes now use
PyLong_AsDouble() instead of PyLong_AsLongAndOverflow() and
compensated summation as well.
Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com>
Co-authored-by: Hugo van Kemenade <1324225+hugovk@users.noreply.github.com>
Co-authored-by: Tomas R <tomas.roun8@gmail.com>
Co-authored-by: Scott Odle <scott@sjodle.com>
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
Co-authored-by: Petr Viktorin <encukou@gmail.com>
Serializing objects with complex __qualname__ (such as unbound methods and
nested classes) by name no longer involves serializing parent objects by value
in pickle protocols < 4.
Adds a --with-app-store-compliance configuration option that patches out code known to be an issue with App Store review processes. This option is applied automatically on iOS, and optionally on macOS.
Add a `Path.rmtree()` method that removes an entire directory tree, like
`shutil.rmtree()`. The signature of the optional *on_error* argument
matches the `Path.walk()` argument of the same name, but differs from the
*onexc* and *onerror* arguments to `shutil.rmtree()`. Consistency within
pathlib is probably more important.
In the private pathlib ABCs, we add an implementation based on `walk()`.
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
Problem occurred when attribute xyz could not be pickled.
Since this is not trivial to selectively fix, block all
attributes (other than 'width'). IDLE does not access them
and they are private implementation details.
* Switch PyUnicode_InternInPlace to _PyUnicode_InternMortal, clarify docs
* Document immortality in some functions that take `const char *`
This is PyUnicode_InternFromString;
PyDict_SetItemString, PyObject_SetAttrString;
PyObject_DelAttrString; PyUnicode_InternFromString;
and the PyModule_Add convenience functions.
Always point out a non-immortalizing alternative.
* Don't immortalize user-provided attr names in _ctypes
We should maintain the invariant that a zero `ob_tid` implies the
refcount fields are merged.
* Move the assignment in `_Py_MergeZeroLocalRefcount` to immediately
before the refcount merge.
* Update `_PyTrash_thread_destroy_chain` to set `ob_ref_shared` to
`_Py_REF_MERGED` when setting `ob_tid` to zero.
Also check this invariant with assertions in the GC in debug builds.
That uncovered a bug when running out of memory during GC.
They are alternate constructors which only accept numbers
(including objects with special methods __float__, __complex__
and __index__), but not strings.
It is our general practice to make new optional parameters keyword-only,
even if the existing parameters are all positional-or-keyword. Passing
this parameter as positional would look confusing and could be error-prone
if additional parameters are added in the future.
Performance improvement to `float.fromhex`: use a lookup table
for computing the hexadecimal value of a character, in place of the
previous switch-case construct. Patch by Bruno Lima.
On POSIX systems, excluding macOS framework installs, the lib directory
for the free-threaded build now includes a "t" suffix to avoid conflicts
with a co-located default build installation.
When builtin static types are initialized for a subinterpreter, various "tp" slots have already been inherited (for the main interpreter). This was interfering with the logic in add_operators() (in Objects/typeobject.c), causing a wrapper to get created when it shouldn't. This change fixes that by preserving the original data from the static type struct and checking that.
The `_PySeqLock_EndRead` function needs an acquire fence to ensure that
the load of the sequence happens after any loads within the read side
critical section. The missing fence can trigger bugs on macOS arm64.
Additionally, we need a release fence in `_PySeqLock_LockWrite` to
ensure that the sequence update is visible before any modifications to
the cache entry.
Make error message for index() methods consistent
Remove the repr of the searched value (which can be arbitrary large)
from ValueError messages for list.index(), range.index(), deque.index(),
deque.remove() and ShareableList.index(). Make the error messages
consistent with error messages for other index() and remove()
methods.
This reduces the system call count of a simple program[0] that reads all
the `.rst` files in Doc by over 10% (5706 -> 4734 system calls on my
linux system, 5813 -> 4875 on my macOS)
This reduces the number of `fstat()` calls always and seek calls most
the time. Stat was always called twice, once at open (to error early on
directories), and a second time to get the size of the file to be able
to read the whole file in one read. Now the size is cached with the
first call.
The code keeps an optimization that if the user had previously read a
lot of data, the current position is subtracted from the number of bytes
to read. That is somewhat expensive so only do it on larger files,
otherwise just try and read the extra bytes and resize the PyBytes as
needeed.
I built a little test program to validate the behavior + assumptions
around relative costs and then ran it under `strace` to get a log of the
system calls. Full samples below[1].
After the changes, this is everything in one `filename.read_text()`:
```python3
openat(AT_FDCWD, "cpython/Doc/howto/clinic.rst", O_RDONLY|O_CLOEXEC) = 3`
fstat(3, {st_mode=S_IFREG|0644, st_size=343, ...}) = 0`
ioctl(3, TCGETS, 0x7ffdfac04b40) = -1 ENOTTY (Inappropriate ioctl for device)
lseek(3, 0, SEEK_CUR) = 0
read(3, ":orphan:\n\n.. This page is retain"..., 344) = 343
read(3, "", 1) = 0
close(3) = 0
```
This does make some tradeoffs
1. If the file size changes between open() and readall(), this will
still get all the data but might have more read calls.
2. I experimented with avoiding the stat + cached result for small files
in general, but on my dev workstation at least that tended to reduce
performance compared to using the fstat().
[0]
```python3
from pathlib import Path
nlines = []
for filename in Path("cpython/Doc").glob("**/*.rst"):
nlines.append(len(filename.read_text()))
```
[1]
Before small file:
```
openat(AT_FDCWD, "cpython/Doc/howto/clinic.rst", O_RDONLY|O_CLOEXEC) = 3
fstat(3, {st_mode=S_IFREG|0644, st_size=343, ...}) = 0
ioctl(3, TCGETS, 0x7ffe52525930) = -1 ENOTTY (Inappropriate ioctl for device)
lseek(3, 0, SEEK_CUR) = 0
lseek(3, 0, SEEK_CUR) = 0
fstat(3, {st_mode=S_IFREG|0644, st_size=343, ...}) = 0
read(3, ":orphan:\n\n.. This page is retain"..., 344) = 343
read(3, "", 1) = 0
close(3) = 0
```
After small file:
```
openat(AT_FDCWD, "cpython/Doc/howto/clinic.rst", O_RDONLY|O_CLOEXEC) = 3
fstat(3, {st_mode=S_IFREG|0644, st_size=343, ...}) = 0
ioctl(3, TCGETS, 0x7ffdfac04b40) = -1 ENOTTY (Inappropriate ioctl for device)
lseek(3, 0, SEEK_CUR) = 0
read(3, ":orphan:\n\n.. This page is retain"..., 344) = 343
read(3, "", 1) = 0
close(3) = 0
```
Before large file:
```
openat(AT_FDCWD, "cpython/Doc/c-api/typeobj.rst", O_RDONLY|O_CLOEXEC) = 3
fstat(3, {st_mode=S_IFREG|0644, st_size=133104, ...}) = 0
ioctl(3, TCGETS, 0x7ffe52525930) = -1 ENOTTY (Inappropriate ioctl for device)
lseek(3, 0, SEEK_CUR) = 0
lseek(3, 0, SEEK_CUR) = 0
fstat(3, {st_mode=S_IFREG|0644, st_size=133104, ...}) = 0
read(3, ".. highlight:: c\n\n.. _type-struc"..., 133105) = 133104
read(3, "", 1) = 0
close(3) = 0
```
After large file:
```
openat(AT_FDCWD, "cpython/Doc/c-api/typeobj.rst", O_RDONLY|O_CLOEXEC) = 3
fstat(3, {st_mode=S_IFREG|0644, st_size=133104, ...}) = 0
ioctl(3, TCGETS, 0x7ffdfac04b40) = -1 ENOTTY (Inappropriate ioctl for device)
lseek(3, 0, SEEK_CUR) = 0
lseek(3, 0, SEEK_CUR) = 0
read(3, ".. highlight:: c\n\n.. _type-struc"..., 133105) = 133104
read(3, "", 1) = 0
close(3) = 0
```
Co-authored-by: Shantanu <12621235+hauntsaninja@users.noreply.github.com>
Co-authored-by: Erlend E. Aasland <erlend.aasland@protonmail.com>
Co-authored-by: Victor Stinner <vstinner@python.org>
Sam Gross
Anders Kaseorg
sobolevn
Jason R. Coombs
Hugo van Kemenade
Irit Katriel
Serhiy Storchaka
adang1345
Tushar Sadhwani
Victor Stinner
Jérôme Duval
algonell
Seth Michael Larson
Łukasz Langa
aorcajo
neonene
Furkan Onder
Peter Bierma
nkinnan
Savannah Ostrowski
Zachary Ware
Petr Viktorin
Bar Harel
Nice Zombies
Wulian
devdanzin
Sergey B Kirpichev
Yorik Hansen
Inada Naoki
Barney Gale
Bénédikt Tran
Donghee Na
Yoda
Wei-Hsiang (Matt) Wang
Oded Arbel
Alexander P.
Matt Wozniski
Bogdan Romanyuk
Alexandr Mitin
Pieter Eendebak
Kumar Aditya
Pablo Galindo Salgado
Kirill Podoprigora
Shantanu
Cody Maloney
blhsing
Mark Shannon
Jeremy Hylton
Pedro Lacerda
CF Bolz-Tereick
Mateusz Nowak
Brett Cannon
Eddie Elizondo
Nate Ohlson
Jonathan Protzenko
Eric Snow
Trey Hunner
Brandt Bucher
Jelle Zijlstra
Nico Mexis
Malcolm Smith
Miro Hrončok
Erlend E. Aasland
Lucas Esposito
Ali Hamdan
pukkandan
Pradyun Gedam
John Riggles
Terry Jan Reedy
Clinton
WilliamRoyNelson
Marc Mueller
Mikołaj Kuranowski
Dino Viehland
Xie Yanbo
Tian Gao
Sebastian Rittau
HarryLHW
Russell Keith-Magee
Rodrigo Oliveira
Nate Ohlson
Matthieu Caneill
Alex Waygood
Andreas Stocker
Ned Deily
Marta Gómez Macías
Rodrigo Girão Serrão
Dominic H
Samuel Thibault
Bruno Lima
Gregor
Bas Bloemsaat
satori1995