These are cleanups I've pulled out of gh-118116. Mostly, this change moves code around to align with some future changes and to improve clarity a little. There is one very small change in behavior: we now add the module to the per-interpreter caches after updating the global state, rather than before.
This is a collection of very basic cleanups I've pulled out of gh-118116. It is mostly renaming variables and moving a couple bits of code in functionally equivalent ways.
Makes sys.settrace, sys.setprofile, and monitoring generally thread-safe.
Mostly uses a stop-the-world approach and synchronization around the code object's _co_instrumentation_version. There may be a little bit of extra synchronization around the monitoring data that's required to be TSAN clean.
We were under-counting calls in `_PyEvalFramePushAndInit`
because the `CALL_STAT_INC` macro was redefined to a no-op
for the Tier 2 interpreter. The fix is not to `#undef` it at all.
This results in ~37% more "Frames pushed" reported
under "Call stats".
Quiet erroneous TSAN reports of data races in `_PySeqLock`
TSAN reports a couple of data races between the compare/exchange in
`_PySeqLock_LockWrite` and the non-atomic loads in `_PySeqLock_{Abandon,Unlock}Write`.
This is another instance of TSAN incorrectly modeling failed compare/exchange
as a write instead of a load.
TSAN erroneously reports a data race between the `_Py_atomic_compare_exchange_int`
on `tstate->state` in `tstate_try_attach()` and the non-atomic load of
`tstate->state` in `start_the_world`. The `_Py_atomic_compare_exchange_int` fails,
but TSAN erroneously treats it as a store.
This is similar to the situation with threading._DummyThread. The methods (incl. __del__()) of interpreters.Interpreter objects must be careful with interpreters not created by interpreters.create(). The simplest thing to start with is to disable any method that modifies or runs in the interpreter. As part of this, the runtime keeps track of where an interpreter was created. We also handle interpreter "refcounts" properly.
The free-threaded build does not currently support the combination of
single-phase init modules and non-isolated subinterpreters. Ensure that
`check_multi_interp_extensions` is always `True` for subinterpreters in
the free-threaded build so that importing these modules raises an
`ImportError`.
This keeps track of the per-thread total reference count operations in
PyThreadState in the free-threaded builds. The count is merged into the
interpreter's total when the thread exits.
Most mutable data is protected by a striped lock that is keyed on the
referenced object's address. The weakref's hash is protected using the
weakref's per-object lock.
Note that this only affects free-threaded builds. Apart from some minor
refactoring, the added code is all either gated by `ifdef`s or is a no-op
(e.g. `Py_BEGIN_CRITICAL_SECTION`).
Introduce a unified 16-bit backoff counter type (``_Py_BackoffCounter``),
shared between the Tier 1 adaptive specializer and the Tier 2 optimizer. The
API used for adaptive specialization counters is changed but the behavior is
(supposed to be) identical.
The behavior of the Tier 2 counters is changed:
- There are no longer dynamic thresholds (we never varied these).
- All counters now use the same exponential backoff.
- The counter for ``JUMP_BACKWARD`` starts counting down from 16.
- The ``temperature`` in side exits starts counting down from 64.
This merges all `_CHECK_STACK_SPACE` uops in a trace into a single `_CHECK_STACK_SPACE_OPERAND` uop that checks whether there is enough stack space for all calls included in the entire trace.
I had meant to switch everything to InterpreterError when I added it a while back. At the time I missed a few key spots.
As part of this, I've added print-the-exception to _PyXI_InitTypes() and fixed an error case in `_PyStaticType_InitBuiltin().
These helpers make it easier to customize and inspect the config used to initialize interpreters. This is especially valuable in our tests. I found inspiration from the PyConfig API for the PyInterpreterConfig dict conversion stuff. As part of this PR I've also added a bunch of tests.
This fixes a crash in `test_threading.test_reinit_tls_after_fork()` when
running with the GIL disabled. We already properly handle the case where
the thread state is `_Py_THREAD_ATTACHED` in `tstate_delete_common()` --
we just need to remove an assertion.
Keeping the thread attached means that a stop-the-world pause, such as
for a `fork()`, won't commence until we remove our thread state from the
interpreter's linked list. This prevents a crash when the child process
tries to clean up the dead thread states.
This adds a stop the world pause to make the two functions thread-safe
when the GIL is disabled in the free-threaded build.
Additionally, the main test thread may call `sys._current_exceptions()` as
soon as `g_raised.set()` is called. The background thread may not yet reach
the `leave_g.wait()` line.
The free-threaded GC sometimes sees objects with zero refcount. This can
happen due to the delay in merging biased reference counting fields,
and, in the future, due to deferred reference counting. We should not
untrack these objects or they will never be collected.
This fixes the refleaks in the free-threaded build.
Mark the swap operations as critical sections.
Add an internal Py_BEGIN_CRITICAL_SECTION_MUT API that takes a PyMutex
pointer instead of a PyObject pointer.
Co-authored-by: Hugo van Kemenade <1324225+hugovk@users.noreply.github.com>
Co-authored-by: Malcolm Smith <smith@chaquo.com>
Co-authored-by: Ned Deily <nad@python.org>
Change old space bit of young objects from 0 to gcstate->visited_space.
This ensures that any object created *and* collected during cycle GC has the bit set correctly.
---------
Co-authored-by: Peter Lazorchak <lazorchakp@gmail.com>
Co-authored-by: Guido van Rossum <gvanrossum@users.noreply.github.com>
Co-authored-by: Guido van Rossum <gvanrossum@gmail.com>
When I added _PyInterpreterState_IsRunningMain() and friends last year, I tried to accommodate applications that embed Python but don't call _PyInterpreterState_SetRunningMain() (not that they're expected to). That mostly worked fine until my recent changes in gh-117049, where the subtleties with the fallback code led to failures; the change ended up breaking test_tools.test_freeze, which exercises a basic embedding situation.
The simplest fix is to drop the fallback code I originally added to _PyInterpreterState_IsRunningMain() (and later to _PyThreadState_IsRunningMain()). I've kept the fallback in the _xxsubinterpreters module though. I've also updated Py_FrozenMain() to call _PyInterpreterState_SetRunningMain().
Changes to the function version cache:
- In addition to the function object, also store the code object,
and allow the latter to be retrieved even if the function has been evicted.
- Stop assigning new function versions after a critical attribute (e.g. `__code__`)
has been modified; the version is permanently reset to zero in this case.
- Changes to `__annotations__` are no longer considered critical. (This fixes gh-109998.)
Changes to the Tier 2 optimization machinery:
- If we cannot map a function version to a function, but it is still mapped to a code object,
we continue projecting the trace.
The operand of the `_PUSH_FRAME` and `_POP_FRAME` opcodes can be either NULL,
a function object, or a code object with the lowest bit set.
This allows us to trace through code that calls an ephemeral function,
i.e., a function that may not be alive when we are constructing the executor,
e.g. a generator expression or certain nested functions.
We will lose globals removal inside such functions,
but we can still do other peephole operations
(and even possibly [call inlining](https://github.com/python/cpython/pull/116290),
if we decide to do it), which only need the code object.
As before, if we cannot retrieve the code object from the cache, we stop projecting.
Split `_PyThreadState_DeleteExcept` into two functions:
- `_PyThreadState_RemoveExcept` removes all thread states other than one
passed as an argument. It returns the removed thread states as a
linked list.
- `_PyThreadState_DeleteList` deletes those dead thread states. It may
call destructors, so we want to "start the world" before calling
`_PyThreadState_DeleteList` to avoid potential deadlocks.
I added it quite a while ago as a strategy for managing interpreter lifetimes relative to the PEP 554 (now 734) implementation. Relatively recently I refactored that implementation to no longer rely on InterpreterID objects. Thus now I'm removing it.
Add Py_GetConstant() and Py_GetConstantBorrowed() functions.
In the limited C API version 3.13, getting Py_None, Py_False,
Py_True, Py_Ellipsis and Py_NotImplemented singletons is now
implemented as function calls at the stable ABI level to hide
implementation details. Getting these constants still return borrowed
references.
Add _testlimitedcapi/object.c and test_capi/test_object.py to test
Py_GetConstant() and Py_GetConstantBorrowed() functions.
Mostly we unify the two different implementations of the conversion code (from PyObject * to int64_t. We also drop the PyArg_ParseTuple()-style converter function, as well as rename and move PyInterpreterID_LookUp().
This changes the free-threaded build to perform a stop-the-world pause
before deleting other thread states when forking and during shutdown.
This fixes some crashes when using multiprocessing and during shutdown
when running with `PYTHON_GIL=0`.
This also changes `PyOS_BeforeFork` to acquire the runtime lock
(i.e., `HEAD_LOCK(&_PyRuntime)`) before forking to ensure that data
protected by the runtime lock (and not just the GIL or stop-the-world)
is in a consistent state before forking.
These writes to `pending->calls_to_do` need to be atomic, because other threads
can read (atomically) from `calls_to_do` without holding `pending->mutex`.
* document equivalent command-line options for all environment variables
* document equivalent environment variables for all command-line options
* reduce the size of variable and option descriptions to minimum
* remove the ending period in single-sentence descriptions
Co-authored-by: Éric <merwok@netwok.org>
Co-authored-by: Hugo van Kemenade <1324225+hugovk@users.noreply.github.com>
Keep Tools/build/deepfreeze.py around (we may repurpose it for deepfreezing non-code objects),
and keep basic "clean" targets that remove the output of former deep-freeze activities,
to keep the build directories of current devs clean.
Somehow we ended up with two separate counter variables tracking "the next function version".
Most likely this was a historical accident where an old branch was updated incorrectly.
This PR merges the two counters into a single one: `interp->func_state.next_version`.
There is a race between when `Thread._tstate_lock` is released[^1] in `Thread._wait_for_tstate_lock()`
and when `Thread._stop()` asserts[^2] that it is unlocked. Consider the following execution
involving threads A, B, and C:
1. A starts.
2. B joins A, blocking on its `_tstate_lock`.
3. C joins A, blocking on its `_tstate_lock`.
4. A finishes and releases its `_tstate_lock`.
5. B acquires A's `_tstate_lock` in `_wait_for_tstate_lock()`, releases it, but is swapped
out before calling `_stop()`.
6. C is scheduled, acquires A's `_tstate_lock` in `_wait_for_tstate_lock()` but is swapped
out before releasing it.
7. B is scheduled, calls `_stop()`, which asserts that A's `_tstate_lock` is not held.
However, C holds it, so the assertion fails.
The race can be reproduced[^3] by inserting sleeps at the appropriate points in
the threading code. To do so, run the `repro_join_race.py` from the linked repo.
There are two main parts to this PR:
1. `_tstate_lock` is replaced with an event that is attached to `PyThreadState`.
The event is set by the runtime prior to the thread being cleared (in the same
place that `_tstate_lock` was released). `Thread.join()` blocks waiting for the
event to be set.
2. `_PyInterpreterState_WaitForThreads()` provides the ability to wait for all
non-daemon threads to exit. To do so, an `is_daemon` predicate was added to
`PyThreadState`. This field is set each time a thread is created. `threading._shutdown()`
now calls into `_PyInterpreterState_WaitForThreads()` instead of waiting on
`_tstate_lock`s.
[^1]: 441affc9e7/Lib/threading.py (L1201)
[^2]: 441affc9e7/Lib/threading.py (L1115)
[^3]: 8194653279
---------
Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com>
Co-authored-by: Antoine Pitrou <antoine@python.org>
On Windows, time.monotonic() now uses the QueryPerformanceCounter()
clock to have a resolution better than 1 us, instead of the
gGetTickCount64() clock which has a resolution of 15.6 ms.
There are now at least two bytecodes that may attempt to optimize,
JUMP_BACK, and more recently, COLD_EXIT.
Only the JUMP_BACK was counting the attempt in the stats.
This moves that counter to uop_optimize itself so it should
always happen no matter where it is called from.
This isn't strictly necessary because the implementation of `gc_should_collect`
already checks `gcstate->enabled` in the free-threaded build, but it seems
like a good idea until the common pieces of gc.c and gc_free_threading.c are
refactored out.
This moves `current_fast_clear()` up so that the current thread state is
`NULL` while running `tstate_delete_common()`.
This doesn't fix any bugs, but it means that we are more consistent that
`_PyThreadState_GET() != NULL` means that the thread is "attached".
Return 0 on success. Set an exception and return -1 on error.
Fix os.timerfd_settime(): properly report exceptions on
_PyTime_FromSecondsDouble() failure.
No longer export _PyTime_FromSecondsDouble().
In free-threaded builds, running with `PYTHON_GIL=0` will now disable the
GIL. Follow-up issues track work to re-enable the GIL when loading an
incompatible extension, and to disable the GIL by default.
In order to support re-enabling the GIL at runtime, all GIL-related data
structures are initialized as usual, and disabling the GIL simply sets a flag
that causes `take_gil()` and `drop_gil()` to return early.
In general, when `_PyThreadState_GET()` is non-NULL then the current
thread is "attached", but there is a small window during
`PyThreadState_DeleteCurrent()` where that's not true:
tstate_delete_common() is called when the thread is detached, but before
current_fast_clear().
Co-authored-by: Eric Snow <ericsnowcurrently@gmail.com>
If a thread blocks while waiting on the `shared->mutex` lock, the array
of QSBR states may be reallocated. The `tstate->qsbr` values before the
lock is acquired may not be the same as the value after the lock is acquired.
This implements the delayed reuse of mimalloc pages that contain Python
objects in the free-threaded build.
Allocations of the same size class are grouped in data structures called
pages. These are different from operating system pages. For thread-safety, we
want to ensure that memory used to store PyObjects remains valid as long as
there may be concurrent lock-free readers; we want to delay using it for
other size classes, in other heaps, or returning it to the operating system.
When a mimalloc page becomes empty, instead of immediately freeing it, we tag
it with a QSBR goal and insert it into a per-thread state linked list of
pages to be freed. When mimalloc needs a fresh page, we process the queue and
free any still empty pages that are now deemed safe to be freed. Pages
waiting to be freed are still available for allocations of the same size
class and allocating from a page prevent it from being freed. There is
additional logic to handle abandoned pages when threads exit.
A previous commit introduced a bug to `interpreter_clear()`: it set
`interp->ceval.instrumentation_version` to 0, without making the corresponding
change to `tstate->eval_breaker` (which holds a thread-local copy of the
version). After this happens, Python code can still run due to object finalizers
during a GC, and the version check in bytecodes.c will see a different result
than the one in instrumentation.c causing an infinite loop.
The fix itself is straightforward: clear `tstate->eval_breaker` when clearing
`interp->ceval.instrumentation_version`.
Make `_thread.ThreadHandle` thread-safe in free-threaded builds
We protect the mutable state of `ThreadHandle` using a `_PyOnceFlag`.
Concurrent operations (i.e. `join` or `detach`) on `ThreadHandle` block
until it is their turn to execute or an earlier operation succeeds.
Once an operation has been applied successfully all future operations
complete immediately.
The `join()` method is now idempotent. It may be called multiple times
but the underlying OS thread will only be joined once. After `join()`
succeeds, any future calls to `join()` will succeed immediately.
The internal thread handle `detach()` method has been removed.
This changes the `sym_set_...()` functions to return a `bool` which is `false`
when the symbol is `bottom` after the operation.
All calls to such functions now check this result and go to `hit_bottom`,
a special error label that prints a different message and then reports
that it wasn't able to optimize the trace. No executor will be produced
in this case.
This undoes the *temporary* default disabling of the T2 optimizer pass in gh-115860.
- Add a new test that reproduces Brandt's example from gh-115859; it indeed crashes before gh-116028 with PYTHONUOPSOPTIMIZE=1
- Re-enable the optimizer pass in T2, stop checking PYTHONUOPSOPTIMIZE
- Rename the env var to disable T2 entirely to PYTHON_UOPS_OPTIMIZE (must be explicitly set to 0 to disable)
- Fix skipIf conditions on tests in test_opt.py accordingly
- Export sym_is_bottom() (for debugging)
- Fix various things in the `_BINARY_OP_` specializations in the abstract interpreter:
- DECREF(temp)
- out-of-space check after sym_new_const()
- add sym_matches_type() checks, so even if we somehow reach a binary op with symbolic constants of the wrong type on the stack we won't trigger the type assert
- Any `sym_set_...` call that attempts to set conflicting information
cause the symbol to become `bottom` (contradiction).
- All `sym_is...` and similar calls return false or NULL for `bottom`.
- Everything's tested.
- The tests still pass with `PYTHONUOPSOPTIMIZE=1`.
* Rename _Py_UOpsAbstractInterpContext to _Py_UOpsContext and _Py_UOpsSymType to _Py_UopsSymbol.
* #define shortened form of _Py_uop_... names for improved readability.
The theory is that even if we saw a jump go in the same direction the
last 16 times we got there, we shouldn't be overly confident that it's
still going to go the same way in the future. This PR makes it so that
in the extreme cases, the confidence is multiplied by 0.9 instead of
remaining unchanged. For unpredictable jumps, there is no difference
(still 0.5). For somewhat predictable jumps, we interpolate.
PyTime_t no longer uses an arbitrary unit, it's always a number of
nanoseconds (64-bit signed integer).
* Rename _PyTime_FromNanosecondsObject() to _PyTime_FromLong().
* Rename _PyTime_AsNanosecondsObject() to _PyTime_AsLong().
* Remove pytime_from_nanoseconds().
* Remove pytime_as_nanoseconds().
* Remove _PyTime_FromNanoseconds().
Remove references to the old names _PyTime_MIN
and _PyTime_MAX, now that PyTime_MIN and
PyTime_MAX are public.
Replace also _PyTime_MIN with PyTime_MIN.
* Rename `_testinternalcapi.get_{uop,counter}_optimizer` to `new_*_optimizer`
* Use `_PyUOpName()` instead of` _PyOpcode_uop_name[]`
* Add `target` to executor iterator items -- `list(ex)` now returns `(opcode, oparg, target, operand)` quadruples
* Add executor methods `get_opcode()` and `get_oparg()` to get `vmdata.opcode`, `vmdata.oparg`
* Define a helper for printing uops, and unify various places where they are printed
* Add a hack to summarize_stats.py to fix legacy uop names (e.g. `POP_TOP` -> `_POP_TOP`)
* Define helpers in `test_opt.py` for accessing the set or list of opnames of an executor
This adds `_PyMem_FreeDelayed()` and supporting functions. The
`_PyMem_FreeDelayed()` function frees memory with the same allocator as
`PyMem_Free()`, but after some delay to ensure that concurrent lock-free
readers have finished.
<pycore_time.h> include is no longer needed to get the PyTime_t type
in internal header files. This type is now provided by <Python.h>
include. Add <pycore_time.h> includes to C files instead.
This avoids filling the memory occupied by ob_tid, ob_ref_local, and
ob_ref_shared with debug bytes (e.g., 0xDD) in mimalloc in the
free-threaded build.
This change adds an `eval_breaker` field to `PyThreadState`. The primary
motivation is for performance in free-threaded builds: with thread-local eval
breakers, we can stop a specific thread (e.g., for an async exception) without
interrupting other threads.
The source of truth for the global instrumentation version is stored in the
`instrumentation_version` field in PyInterpreterState. Threads usually read the
version from their local `eval_breaker`, where it continues to be colocated
with the eval breaker bits.
This adds a safe memory reclamation scheme based on FreeBSD's "GUS" and
quiescent state based reclamation (QSBR). The API provides a mechanism
for callers to detect when it is safe to free memory that may be
concurrently accessed by readers.
The GC keeps track of the number of allocations (less deallocations)
since the last GC. This buffers the count in thread-local state and uses
atomic operations to modify the per-interpreter count. The thread-local
buffering avoids contention on shared state.
A consequence is that the GC scheduling is not as precise, so
"test_sneaky_frame_object" is skipped because it requires that the GC be
run exactly after allocating a frame object.
Makes _PyType_Lookup thread safe, including:
Thread safety of the underlying cache.
Make mutation of mro and type members thread safe
Also _PyType_GetMRO and _PyType_GetBases are currently returning borrowed references which aren't safe.
This is a temporary fix to unblock embedders that do not call Py_Main().
_PyInterpreterState_IsRunningMain() will always return true for the main interpreter, even in corner cases where it technically should not. The (future) full solution will do the right thing in those corner cases.
For the most part, these changes make is substantially easier to backport subinterpreter-related code to 3.12, especially the related modules (e.g. _xxsubinterpreters). The main motivation is to support releasing a PyPI package with the 3.13 capabilities compiled for 3.12.
A lot of the changes here involve either hiding details behind macros/functions or splitting up some files.
Setters for members with an unsigned integer type now support
the same range of valid values for objects that has a __index__()
method as for int.
Previously, Py_T_UINT, Py_T_ULONG and Py_T_ULLONG did not support
objects that has a __index__() method larger than LONG_MAX.
Py_T_ULLONG did not support negative ints. Now it supports them and
emits a RuntimeWarning.
Biased reference counting maintains two refcount fields in each object:
`ob_ref_local` and `ob_ref_shared`. The true refcount is the sum of these two
fields. In some cases, when refcounting operations are split across threads,
the ob_ref_shared field can be negative (although the total refcount must be
at least zero). In this case, the thread that decremented the refcount
requests that the owning thread give up ownership and merge the refcount
fields.
This changes a number of internal usages of `PyDict_SetDefault` to use `PyDict_SetDefaultRef`.
Co-authored-by: Erlend E. Aasland <erlend.aasland@protonmail.com>
This marks dead ThreadHandles as non-joinable earlier in
`PyOS_AfterFork_Child()` before we execute any Python code. The handles
are stored in a global linked list in `_PyRuntimeState` because `fork()`
affects the entire process.
Fix race between `_PyParkingLot_Park` and `_PyParkingLot_UnparkAll` when handling interrupts
There is a potential race when `_PyParkingLot_UnparkAll` is executing in
one thread and another thread is unblocked because of an interrupt in
`_PyParkingLot_Park`. Consider the following scenario:
1. Thread T0 is blocked[^1] in `_PyParkingLot_Park` on address `A`.
2. Thread T1 executes `_PyParkingLot_UnparkAll` on address `A`. It
finds the `wait_entry` for `T0` and unlinks[^2] its list node.
3. Immediately after (2), T0 is woken up due to an interrupt. It
then segfaults trying to unlink[^3] the node that was previously
unlinked in (2).
To fix this we mark each waiter as unparking before releasing the bucket
lock. `_PyParkingLot_Park` will wait to handle the coming wakeup, and not
attempt to unlink the node, when this field is set. `_PyParkingLot_Unpark`
does this already, presumably to handle this case.
* Fix a RuntimeWarning emitted when assign an integer-like value that
is not an instance of int to an attribute that corresponds to a C
struct member of type T_UINT and T_ULONG.
* Fix a double RuntimeWarning emitted when assign a negative integer value
to an attribute that corresponds to a C struct member of type T_UINT.
* gh-112529: Remove PyGC_Head from object pre-header in free-threaded build
This avoids allocating space for PyGC_Head in the free-threaded build.
The GC implementation for free-threaded CPython does not use the
PyGC_Head structure.
* The trashcan mechanism uses the `ob_tid` field instead of `_gc_prev`
in the free-threaded build.
* The GDB libpython.py file now determines the offset of the managed
dict field based on whether the running process is a free-threaded
build. Those are identified by the `ob_ref_local` field in PyObject.
* Fixes `_PySys_GetSizeOf()` which incorrectly incorrectly included the
size of `PyGC_Head` in the size of static `PyTypeObject`.
The free-threaded build's GC implementation is non-generational, but was
scheduled as if it were collecting a young generation leading to
quadratic behavior. This increases the minimum threshold and scales it
to the number of live objects as we do for the old generation in the
default build.
Note that the scheduling is still not thread-safe without the GIL. Those
changes will come in later PRs.
A few tests, like "test_sneaky_frame_object" rely on prompt scheduling
of the GC. For now, to keep that test passing, we disable the scaled
threshold after calls like `gc.set_threshold(1, 0, 0)`.
Add a configure define for HAVE_PTHREAD_COND_TIMEDWAIT_RELATIVE_NP and
replaces pthread_cond_timedwait() with pthread_cond_timedwait_relative_np()
for relative time when supported in semaphore waiting logic.
Mark Shannon
neonene
Tian Gao
Eric Snow
Nice Zombies
Irit Katriel
Guido van Rossum
mpage
Joe Jevnik
Jelle Zijlstra
Dino Viehland
Russell Keith-Magee
Donghee Na
Victor Stinner
Kirill Podoprigora
Jeff Glass
Sam Gross
Serhiy Storchaka
Brandt Bucher
Erlend E. Aasland
Michael Droettboom
Peter Lazorchak
Steve (Gadget) Barnes
Ken Jin
Brett Simmers
Bogdan Romanyuk
guangwu
Pablo Galindo Salgado
Nikita Sobolev
Steve Dower
Sergii K
Benjamin Peterson
wookie184
David Hewitt
Petr Viktorin
Shantanu
Carl Meyer
Jason Zhang
Andrew Rogers
Anders Kaseorg
Matt Prodani