- Unify `get_unicode` and `get_string` in a single function.
- Allow to retrieve the underlying `object` attribute, its
size, and the adjusted 'start' and 'end', all at once.
Add a new `_PyUnicodeError_GetParams` internal function for this.
(In `exceptions.c`, it's somewhat common to not need all the attributes,
but the compiler has opportunity to inline the function and optimize
unneeded work away. Outside that file, we'll usually need all or
most of them at once.)
- Use a common implementation for the following functions:
- `PyUnicode{Decode,Encode}Error_GetEncoding`
- `PyUnicode{Decode,Encode,Translate}Error_GetObject`
- `PyUnicode{Decode,Encode,Translate}Error_{Get,Set}Reason`
- `PyUnicode{Decode,Encode,Translate}Error_{Get,Set}{Start,End}`
There was a data race on the utf8 field between `PyUnicode_SET_UTF8` and
`_PyUnicode_CheckConsistency`. Use the `_PyUnicode_UTF8()` accessor,
which uses an atomic load internally, to avoid the data race.
Methods (functions defined in class scope) are likely to be cleaned
up by the GC anyway.
Add a new code flag, `CO_METHOD`, that is set for functions defined
in a class scope. Use that when deciding to defer functions.
* Add `_PyDictKeys_StringLookupSplit` which does locking on dict keys and
use in place of `_PyDictKeys_StringLookup`.
* Change `_PyObject_TryGetInstanceAttribute` to use that function
in the case of split keys.
* Add `unicodekeys_lookup_split` helper which allows code sharing
between `_Py_dict_lookup` and `_PyDictKeys_StringLookupSplit`.
* Fix locking for `STORE_ATTR_INSTANCE_VALUE`. Create
`_GUARD_TYPE_VERSION_AND_LOCK` uop so that object stays locked and
`tp_version_tag` cannot change.
* Pass `tp_version_tag` to `specialize_dict_access()`, ensuring
the version we store on the cache is the correct one (in case of
it changing during the specalize analysis).
* Split `analyze_descriptor` into `analyze_descriptor_load` and
`analyze_descriptor_store` since those don't share much logic.
Add `descriptor_is_class` helper function.
* In `specialize_dict_access`, double check `_PyObject_GetManagedDict()`
in case we race and dict was materialized before the lock.
* Avoid borrowed references in `_Py_Specialize_StoreAttr()`.
* Use `specialize()` and `unspecialize()` helpers.
* Add unit tests to ensure specializing happens as expected in FT builds.
* Add unit tests to attempt to trigger data races (useful for running under TSAN).
* Add `has_split_table` function to `_testinternalcapi`.
The `PyWeakref_IsDead()` function tests if a weak reference is dead
without any side effects. Although you can also detect if a weak
reference is dead using `PyWeakref_GetRef()`, that function returns a
strong reference that must be `Py_DECREF()`'d, which can introduce side
effects if the last reference is concurrently dropped (at least in the
free threading build).
In some cases, previously computed as (nan+nanj), we could recover
meaningful component values in the result, see e.g. the C11, Annex
G.5.1, routine _Cmultd():
>>> z = 1e300+1j
>>> z*(nan+infj) # was (nan+nanj)
(-inf+infj)
That also fix some complex powers for small integer exponents, computed
with optimized algorithm (by squaring):
>>> z**5 # was (nan+nanj)
Traceback (most recent call last):
File "<python-input-1>", line 1, in <module>
z**5
~^^~
OverflowError: complex exponentiation
Objects may be temporarily "resurrected" in destructors when calling
finalizers or watcher callbacks. We previously undid the resurrection
by decrementing the reference count using `Py_SET_REFCNT`. This was not
thread-safe because other threads might be accessing the object
(modifying its reference count) if it was exposed by the finalizer,
watcher callback, or temporarily accessed by a racy dictionary or list
access.
This adds internal-only thread-safe functions for temporary object
resurrection during destructors.
The CALL family of instructions were mostly thread-safe already and only required a small number of changes, which are documented below.
A few changes were needed to make CALL_ALLOC_AND_ENTER_INIT thread-safe:
Added _PyType_LookupRefAndVersion, which returns the type version corresponding to the returned ref.
Added _PyType_CacheInitForSpecialization, which takes an init method and the corresponding type version and only populates the specialization cache if the current type version matches the supplied version. This prevents potentially caching a stale value in free-threaded builds if we race with an update to __init__.
Only cache __init__ functions that are deferred in free-threaded builds. This ensures that the reference to __init__ that is stored in the specialization cache is valid if the type version guard in _CHECK_AND_ALLOCATE_OBJECT passes.
Fix a bug in _CREATE_INIT_FRAME where the frame is pushed to the stack on failure.
A few other miscellaneous changes were also needed:
Use {LOCK,UNLOCK}_OBJECT in LIST_APPEND. This ensures that the list's per-object lock is held while we are appending to it.
Add missing co_tlbc for _Py_InitCleanup.
Stop/start the world around setting the eval frame hook. This allows us to read interp->eval_frame non-atomically and preserves the behavior of _CHECK_PEP_523 documented below.
* Replace uses of `PyCell_GET` and `PyCell_SET`. These macros are not
safe to use in the free-threaded build. Use `PyCell_GetRef()` and
`PyCell_SetTakeRef()` instead.
* Since `PyCell_GetRef()` returns a strong rather than borrowed ref, some
code restructuring was required, e.g. `frame_get_var()` returns a strong
ref now.
* Add critical sections to `PyCell_GET` and `PyCell_SET`.
* Move critical_section.h earlier in the Python.h file.
* Add `PyCell_GET` to the free-threading howto table of APIs that return
borrowed refs.
* Add additional unit tests for free-threading.
Use existing helpers to atomically modify the bytecode. Add unit tests
to ensure specializing is happening as expected. Add test_specialize.py
that can be used with ThreadSanitizer to detect data races.
Fix thread safety issue with cell_set_contents().
In the free threading build, if a non-owning thread resizes a list,
it must use QSBR to free the old list array because there may be a
concurrent access (without a lock) from the owning thread.
To match the pattern in dictobject.c, we just mark the list as "shared"
before resizing if it's from a non-owning thread and not already marked
as shared.
"Generally, mixed-mode arithmetic combining real and complex variables should
be performed directly, not by first coercing the real to complex, lest the sign
of zero be rendered uninformative; the same goes for combinations of pure
imaginary quantities with complex variables." (c) Kahan, W: Branch cuts for
complex elementary functions.
This patch implements mixed-mode arithmetic rules, combining real and
complex variables as specified by C standards since C99 (in particular,
there is no special version for the true division with real lhs
operand). Most C compilers implementing C99+ Annex G have only these
special rules (without support for imaginary type, which is going to be
deprecated in C2y).
Enable specialization of LOAD_GLOBAL in free-threaded builds.
Thread-safety of specialization in free-threaded builds is provided by the following:
A critical section is held on both the globals and builtins objects during specialization. This ensures we get an atomic view of both builtins and globals during specialization.
Generation of new keys versions is made atomic in free-threaded builds.
Existing helpers are used to atomically modify the opcode.
Thread-safety of specialized instructions in free-threaded builds is provided by the following:
Relaxed atomics are used when loading and storing dict keys versions. This avoids potential data races as the dict keys versions are read without holding the dictionary's per-object lock in version guards.
Dicts keys objects are passed from keys version guards to the downstream uops. This ensures that we are loading from the correct offset in the keys object. Once a unicode key has been stored in a keys object for a combined dictionary in free-threaded builds, the offset that it is stored in will never be reused for a different key. Once the version guard passes, we know that we are reading from the correct offset.
The dictionary read fast-path is used to read values from the dictionary once we know the correct offset.
This is a precursor to the actual fix for gh-114940, where we will change these macros to use the new lock. This change is almost entirely mechanical; the exceptions are the loops in codeobject.c and ceval.c, which now hold the "head" lock. Note that almost all of the uses of _Py_FOR_EACH_TSTATE_UNLOCKED() here will change to _Py_FOR_EACH_TSTATE_BEGIN() once we add the new per-interpreter lock.
* Mark almost all reachable objects before doing collection phase
* Add stats for objects marked
* Visit new frames before each increment
* Remove lazy dict tracking
* Update docs
* Clearer calculation of work to do.
Move creation of a tuple for var-positional parameter out of
_PyArg_UnpackKeywordsWithVararg().
Merge _PyArg_UnpackKeywordsWithVararg() with _PyArg_UnpackKeywords().
Add a new parameter in _PyArg_UnpackKeywords().
The "parameters" and "converters" attributes of ParseArgsCodeGen no
longer contain the var-positional parameter. It is now available as the
"varpos" attribute. Optimize code generation for var-positional
parameter and reuse the same generating code for functions with and without
keyword parameters.
Add special converters for var-positional parameter. "tuple" represents it as
a Python tuple and "array" represents it as a continuous array of PyObject*.
"object" is a temporary alias of "tuple".
* gh-126298: Don't deduplicated slice constants based on equality
* NULL check for PySlice_New
* Fix refcounting
* Fix refcounting some more
* Fix refcounting
* Make tests more complete
* Fix tests
Each thread specializes a thread-local copy of the bytecode, created on the first RESUME, in free-threaded builds. All copies of the bytecode for a code object are stored in the co_tlbc array on the code object. Threads reserve a globally unique index identifying its copy of the bytecode in all co_tlbc arrays at thread creation and release the index at thread destruction. The first entry in every co_tlbc array always points to the "main" copy of the bytecode that is stored at the end of the code object. This ensures that no bytecode is copied for programs that do not use threads.
Thread-local bytecode can be disabled at runtime by providing either -X tlbc=0 or PYTHON_TLBC=0. Disabling thread-local bytecode also disables specialization.
Concurrent modifications to the bytecode made by the specializing interpreter and instrumentation use atomics, with specialization taking care not to overwrite an instruction that was instrumented concurrently.
Avoid temporary tuple creation when all arguments either positional-only
or vararg.
Objects/setobject.c and Modules/gcmodule.c adapted. This fixes slight
performance regression for set methods, introduced by gh-115112.
* Remove `@suppress_immortalization` decorator
* Make suppression flag per-thread instead of per-interpreter
* Suppress immortalization in `eval()` to avoid refleaks in three tests
(test_datetime.test_roundtrip, test_logging.test_config8_ok, and
test_random.test_after_fork).
* frozenset() is constant, but not a singleton. When run multiple times,
the test could fail due to constant interning.
This replaces `_PyEval_BuiltinsFromGlobals` with
`_PyDict_LoadBuiltinsFromGlobals`, which returns a new reference
instead of a borrowed reference. Internally, the new function uses
per-thread reference counting when possible to avoid contention on the
refcount fields on the builtins module.
They used to be shared, before 3.12. Returning to sharing them resolves a failure on Py_TRACE_REFS builds.
Co-authored-by: Petr Viktorin <encukou@gmail.com>
Dictionary watchers on an object's attributes dictionary
(`object.__dict__`) were not triggered when the managed dictionary used
the object's inline values.
This is essentially a cleanup, moving a handful of API declarations to the header files where they fit best, creating new ones when needed.
We do the following:
* add pycore_debug_offsets.h and move _Py_DebugOffsets, etc. there
* inline struct _getargs_runtime_state and struct _gilstate_runtime_state in _PyRuntimeState
* move struct _reftracer_runtime_state to the existing pycore_object_state.h
* add pycore_audit.h and move to it _Py_AuditHookEntry , _PySys_Audit(), and _PySys_ClearAuditHooks
* add audit.h and cpython/audit.h and move the existing audit-related API there
*move the perfmap/trampoline API from cpython/sysmodule.h to cpython/ceval.h, and remove the now-empty cpython/sysmodule.h
Use per-thread refcounting for the reference from function objects to
their corresponding code object. This can be a source of contention when
frequently creating nested functions. Deferred refcounting alone isn't a
great fit here because these references are on the heap and may be
modified by other libraries.
Replace the private _PyUnicodeWriter with the public PyUnicodeWriter.
Replace PyObject_Repr() + _PyUnicodeWriter_WriteStr()
with PyUnicodeWriter_WriteRepr().
Fix a crash caused by immortal interned strings being shared between
sub-interpreters that use basic single-phase init. In that case, the string
can be used by an interpreter that outlives the interpreter that created and
interned it. For interpreters that share obmalloc state, also share the
interned dict with the main interpreter.
This is an un-revert of gh-124646 that then addresses the Py_TRACE_REFS
failures identified by gh-124785.
* Replace unicode_compare_eq() with unicode_eq().
* Use unicode_eq() in setobject.c.
* Replace _PyUnicode_EQ() with _PyUnicode_Equal().
* Remove unicode_compare_eq() and _PyUnicode_EQ().
* Make slices marshallable
* Emit slices as constants
* Update Python/marshal.c
Co-authored-by: Peter Bierma <zintensitydev@gmail.com>
* Refactor codegen_slice into two functions so it
always has the same net effect
* Fix for free-threaded builds
* Simplify marshal loading of slices
* Only return SUCCESS/ERROR from codegen_slice
---------
Co-authored-by: Mark Shannon <mark@hotpy.org>
Co-authored-by: Peter Bierma <zintensitydev@gmail.com>
Stop the world when invalidating function versions
The tier1 interpreter specializes `CALL` instructions based on the values
of certain function attributes (e.g. `__code__`, `__defaults__`). The tier1
interpreter uses function versions to verify that the attributes of a function
during execution of a specialization match those seen during specialization.
A function's version is initialized in `MAKE_FUNCTION` and is invalidated when
any of the critical function attributes are changed. The tier1 interpreter stores
the function version in the inline cache during specialization. A guard is used by
the specialized instruction to verify that the version of the function on the operand
stack matches the cached version (and therefore has all of the expected attributes).
It is assumed that once the guard passes, all attributes will remain unchanged
while executing the rest of the specialized instruction.
Stopping the world when invalidating function versions ensures that all critical
function attributes will remain unchanged after the function version guard passes
in free-threaded builds. It's important to note that this is only true if the remainder
of the specialized instruction does not enter and exit a stop-the-world point.
We will stop the world the first time any of the following function attributes
are mutated:
- defaults
- vectorcall
- kwdefaults
- closure
- code
This should happen rarely and only happens once per function, so the performance
impact on majority of code should be minimal.
Additionally, refactor the API for manipulating function versions to more clearly
match the stated semantics.
Currently, we only use per-thread reference counting for heap type objects and
the naming reflects that. We will extend it to a few additional types in an
upcoming change to avoid scaling bottlenecks when creating nested functions.
Rename some of the files and functions in preparation for this change.
Instead of be limited just by the size of addressable memory (2**63
bytes), Python integers are now also limited by the number of bits, so
the number of bit now always fit in a 64-bit integer.
Both limits are much larger than what might be available in practice,
so it doesn't affect users.
_PyLong_NumBits() and _PyLong_Frexp() are now always successful.
* Setting the __module__ attribute for a class now removes the
__firstlineno__ item from the type's dict.
* The _collections_abc and _pydecimal modules now completely replace the
collections.abc and decimal modules after importing them. This
allows to get the source of classes and functions defined in these
modules.
* inspect.findsource() now checks whether the first line number for a
class is out of bound.
Use a `_PyStackRef` and defer the reference to `f_funcobj` 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.
We need to return immediately if there's an error during dictionary
lookup.
Also avoid the conditional-if operator. MSVC versions through v19.27 miscompile
compound literals with side effects within a conditional operator. This caused
crashes in the Windows10 buildbot.
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.
In gh-121602, I applied a fix to a builtin types initialization bug.
That fix made sense in the context of some broader future changes,
but introduced a little bit of extra complexity. That fix has turned
out to be incomplete for some of the builtin types we haven't
been testing. I found that out while improving the tests.
A while back, @markshannon suggested a simpler fix that doesn't
have that problem, which I've already applied to 3.12 and 3.13.
I'm switching to that here. Given the potential long-term
benefits of the more complex (but still incomplete) approach,
I'll circle back to it in the future, particularly after I've improved
the tests so no corner cases slip through the cracks.
(This is effectively a "forward-port" of 716c677 from 3.13.)
Switch more _Py_IsImmortal(...) assertions to _Py_IsImmortalLoose(...)
The remaining calls to _Py_IsImmortal are in free-threaded-only code,
initialization of core objects, tests, and guards that fall back to
code that works with mortal objects.
* Replace PyLong_AS_LONG() with PyLong_AsLong().
* Call PyLong_AsLong() only once per the replacement code.
* Use PyMapping_GetOptionalItem() instead of PyObject_GetItem().
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.
There were a still a number of gaps in the tests, including not looking
at all the builtin types and not checking wrappers in subinterpreters
that weren't in the main interpreter. This fixes all that.
I considered incorporating the names of the PyTypeObject fields
(a la gh-122866), but figured doing so doesn't add much value.
This replaces `_PyList_FromArraySteal` with `_PyList_FromStackRefSteal`.
It's functionally equivalent, but takes a `_PyStackRef` array instead of
an array of `PyObject` pointers.
Co-authored-by: Ken Jin <kenjin@python.org>
* Parameters after the var-positional parameter are now keyword-only
instead of positional-or-keyword.
* Correctly calculate min_kw_only.
* Raise errors for invalid combinations of the var-positional parameter
with "*", "/" and deprecation markers.
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>
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>
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.
* gh-120974: Make _asyncio._leave_task atomic in the free-threaded build
Update `_PyDict_DelItemIf` to allow for an argument to be passed to the
predicate.
This refactors asyncio to use the common freelist helper functions and
macros. As a side effect, the freelist for _asyncio.Future is now
re-enabled in the free-threaded build.
This combines and updates our freelist handling to use a consistent
implementation. Objects in the freelist are linked together using the
first word of memory block.
If configured with freelists disabled, these operations are essentially
no-ops.
compare_unicode_generic(), compare_unicode_unicode() and
compare_generic() are callbacks used by do_lookup(). When enabling
assertions, it's not possible to inline these functions.
* 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.
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.
* The result has type Py_ssize_t, not intptr_t.
* Type cast between unsigned and signdet integer types should be explicit.
* Downcasting should be explicit.
* Fix integer overflow check in sum().
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 `used` field must be written using atomic stores because `set_len`
and iterators may access the field concurrently without holding the
per-object lock.
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.
Refactor the fast Unicode hash check into `_PyObject_HashFast` and use relaxed
atomic loads in the free-threaded build.
After this change, the TSAN doesn't report data races for this method.
In some cases, previously computed as (nan+nanj), we could
recover meaningful component values in the result, see
e.g. the C11, Annex G.5.2, routine _Cdivd().
Fix warnings when using -Wimplicit-fallthrough compiler flag.
Annotate explicitly "fall through" switch cases with a new
_Py_FALLTHROUGH macro which uses __attribute__((fallthrough)) if
available. Replace "fall through" comments with _Py_FALLTHROUGH.
Add _Py__has_attribute() macro. No longer define __has_attribute()
macro if it's not defined. Move also _Py__has_builtin() at the top
of pyport.h.
Co-Authored-By: Nikita Sobolev <mail@sobolevn.me>
This PR sets up tagged pointers for CPython.
The general idea is to create a separate struct _PyStackRef for everything on the evaluation stack to store the bits. This forces the C compiler to warn us if we try to cast things or pull things out of the struct directly.
Only for free threading: We tag the low bit if something is deferred - that means we skip incref and decref operations on it. This behavior may change in the future if Mark's plans to defer all objects in the interpreter loop pans out.
This implies a strict stack reference discipline is required. ALL incref and decref operations on stackrefs must use the stackref variants. It is unsafe to untag something then do normal incref/decref ops on it.
The new incref and decref variants are called dup and close. They mimic a "handle" API operating on these stackrefs.
Please read Include/internal/pycore_stackref.h for more information!
---------
Co-authored-by: Mark Shannon <9448417+markshannon@users.noreply.github.com>
Remove the const qualifier of the argument of functions:
* _PyLong_IsCompact()
* _PyLong_CompactValue()
Py_TYPE() argument is not const.
Fix the compiler warning:
Include/cpython/longintrepr.h: In function ‘_PyLong_CompactValue’:
Include/pyport.h:19:31: error: cast discards ‘const’ qualifier from
pointer target type [-Werror=cast-qual]
(...)
Include/cpython/longintrepr.h:133:30: note: in expansion of macro
‘Py_TYPE’
assert(PyType_HasFeature(Py_TYPE(op), Py_TPFLAGS_LONG_SUBCLASS));
Bénédikt Tran