On Linux where the `subprocess` module can use the `vfork` syscall for
faster spawning, prevent the parent process from blocking other threads
by dropping the GIL while it waits for the vfork'ed child process `exec`
outcome. This prevents spawning a binary from a slow filesystem from
blocking the rest of the application.
Fixes#104372.
This matches the GIL releasing behavior of our existing `_hashopenssl`
module, extending it to the HACL* built-ins.
Includes adding comments to better describe the ENTER/LEAVE macros
purpose and explain the lock strategy in both existing and new code.
* Support for conversion specifiers o (octal) and X (uppercase hexadecimal).
* Support for length modifiers j (intmax_t) and t (ptrdiff_t).
* Length modifiers are now applied to all integer conversions.
* Support for wchar_t C strings (%ls and %lV).
* Support for variable width and precision (*).
* Support for flag - (left alignment).
This PR updates `math.nextafter` to add a new `steps` argument. The behaviour is as though `math.nextafter` had been called `steps` times in succession.
---------
Co-authored-by: Mark Dickinson <mdickinson@enthought.com>
This adds a number of PRIO_DARWIN_* constants to the os module for use with os.setpriority.
---------
Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com>
Move all of the Python C API calls into the parent process up front
instead of doing PyLong_AsLong and PyErr_Occurred and PyTuple_GET from
the post-fork/vfork child process.
Much of this was long overdue. We shouldn't have been using PyTuple and
PyLong APIs within all of these low level functions anyways.
This implements PEP 695, Type Parameter Syntax. It adds support for:
- Generic functions (def func[T](): ...)
- Generic classes (class X[T](): ...)
- Type aliases (type X = ...)
- New scoping when the new syntax is used within a class body
- Compiler and interpreter changes to support the new syntax and scoping rules
Co-authored-by: Marc Mueller <30130371+cdce8p@users.noreply.github.com>
Co-authored-by: Eric Traut <eric@traut.com>
Co-authored-by: Larry Hastings <larry@hastings.org>
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
When preparing the _io extension module for isolation, many methods were
adapted to Argument Clinic. Some of these used the '*args: object'
signature, which is incorrect. These are now corrected to an exact
signature, and marked unused, since they are stub methods.
This PR removes `_Py_dg_stdnan` and `_Py_dg_infinity` in favour of
using the standard `NAN` and `INFINITY` macros provided by C99.
This change has the side-effect of fixing a bug on MIPS where the
hard-coded value used by `_Py_dg_stdnan` gave a signalling NaN
rather than a quiet NaN.
---------
Co-authored-by: Mark Dickinson <dickinsm@gmail.com>
- Replace query with parameter in bufferediobase_unsupported()
- Replace query with parameter in iobase_unsupported()
- Hide delegate: Add method wrapper for _PyIOBase_check_seekable
- Hide delegate: Add method wraper for _PyIOBase_check_readable
- Hide delegate: Add method wraper for _PyIOBase_check_writable
- Replace query with parameter in _PyIOBase_check_seekable()
- Replace query with parameter in _PyIOBase_check_readable()
- Replace query with parameter in _PyIOBase_check_writable()
* socket.getnameinfo should release the GIL
* 📜🤖 Added by blurb_it.
---------
Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com>
Replaces our built-in SHA3 implementation with a verified one from the HACL* project.
This implementation is used when OpenSSL does not provide SHA3 or is not present.
3.11 shiped with a very slow tiny sha3 implementation to get off of the <=3.10 reference implementation that wound up having serious bugs. This brings us back to a reasonably performing built-in implementation consistent with what we've just replaced our other guaranteed available standard hash algorithms with: code from the HACL* project.
---------
Co-authored-by: Gregory P. Smith <greg@krypto.org>
We also add PyInterpreterState.ceval.own_gil to record if the interpreter actually has its own GIL.
Note that for now we don't actually respect own_gil; all interpreters still share the one GIL. However, PyInterpreterState.ceval.own_gil does reflect PyInterpreterConfig.own_gil. That lie is a temporary one that we will fix when the GIL really becomes per-interpreter.
Here we are doing no more than adding the value for Py_mod_multiple_interpreters and using it for stdlib modules. We will start checking for it in gh-104206 (once PyInterpreterState.ceval.own_gil is added in gh-104204).
In preparation for a per-interpreter GIL, we add PyInterpreterState.ceval.gil, set it to the shared GIL for each interpreter, and use that rather than using _PyRuntime.ceval.gil directly. Note that _PyRuntime.ceval.gil is still the actual GIL.
We also expose PyInterpreterConfig. This is part of the PEP 684 (per-interpreter GIL) implementation. We will add docs as soon as we can.
FYI, I'm adding the new config field for per-interpreter GIL in gh-99114.
his involves moving tp_dict, tp_bases, and tp_mro to PyInterpreterState, in the same way we did for tp_subclasses. Those three fields are effectively const for builtin static types (unlike tp_subclasses). In theory we only need to make their values immortal, along with their contents. However, that isn't such a simple proposition. (See gh-103823.) In the meantime the simplest solution is to move the fields into the interpreter.
One alternative is to statically allocate the values, but that's its own can of worms.
The default task name is "Task-<counter>" (if no name is passed in during Task creation).
This is initialized in `Task.__init__` (C impl) using string formatting, which can be quite slow.
Actually using the task name in real world code is not very common, so this is wasted init.
Let's defer this string formatting to the first time the name is read (in `get_name` impl),
so we don't need to pay the string formatting cost if the task name is never read.
We don't change the order in which tasks are assigned numbers (if they are) --
the number is set on task creation, as a PyLong instead of a formatted string.
Co-authored-by: Łukasz Langa <lukasz@langa.pl>
Using `datetime.datetime.utcnow()` and `datetime.datetime.utcfromtimestamp()` will now raise a `DeprecationWarning`.
We also have removed our internal uses of these functions and documented the change.
This is strictly about moving the "obmalloc" runtime state from
`_PyRuntimeState` to `PyInterpreterState`. Doing so improves isolation
between interpreters, specifically most of the memory (incl. objects)
allocated for each interpreter's use. This is important for a
per-interpreter GIL, but such isolation is valuable even without it.
FWIW, a per-interpreter obmalloc is the proverbial
canary-in-the-coalmine when it comes to the isolation of objects between
interpreters. Any object that leaks (unintentionally) to another
interpreter is highly likely to cause a crash (on debug builds at
least). That's a useful thing to know, relative to interpreter
isolation.
This is the implementation of PEP683
Motivation:
The PR introduces the ability to immortalize instances in CPython which bypasses reference counting. Tagging objects as immortal allows up to skip certain operations when we know that the object will be around for the entire execution of the runtime.
Note that this by itself will bring a performance regression to the runtime due to the extra reference count checks. However, this brings the ability of having truly immutable objects that are useful in other contexts such as immutable data sharing between sub-interpreters.
The function is like Py_AtExit() but for a single interpreter. This is a companion to the atexit module's register() function, taking a C callback instead of a Python one.
We also update the _xxinterpchannels module to use _Py_AtExit(), which is the motivating case. (This is inspired by pain points felt while working on gh-101660.)
Fix a (correct) warning about potential uses of uninitialized memory in
_xxsubinterpreter. Unlike newly allocated PyObject structs or global
structs, stack-allocated structs are not initialised, and a few places in
the code expect the _sharedexception struct data to be either NULL or
initialised.
Prior to https://github.com/python/cpython/pull/25300, the
make_ssl_data.py script used various tables, exposed in _ssl, to update
the error list.
After that PR, this is no longer used. Moreover, the err_names_to_codes
map isn't used at all. Clean those up. This gets them out of the way if,
in the future, OpenSSL provides an API to do what the code here is doing
directly. (https://github.com/openssl/openssl/issues/19848)
In PEM, we need to parse until error and then suppress `PEM_R_NO_START_LINE`, because PEM allows arbitrary leading and trailing data. DER, however, does not. Parsing until error and suppressing `ASN1_R_HEADER_TOO_LONG` doesn't quite work because that error also covers some cases that should be rejected.
Instead, check `BIO_eof` early and stop the loop that way.
Automerge-Triggered-By: GH:Yhg1s
* Eliminate all remaining uses of Py_SIZE and Py_SET_SIZE on PyLongObject, adding asserts.
* Change layout of size/sign bits in longobject to support future addition of immortal ints and tagged medium ints.
* Add functions to hide some internals of long object, and for setting sign and digit count.
* Replace uses of IS_MEDIUM_VALUE macro with _PyLong_IsCompact().
GH-25309 enabled SSL_OP_IGNORE_UNEXPECTED_EOF by default, with a comment
that it restores OpenSSL 1.1.1 behavior, but this wasn't quite right.
That option causes OpenSSL to treat transport EOF as the same as
close_notify (i.e. SSL_ERROR_ZERO_RETURN), whereas Python actually has
distinct SSLEOFError and SSLZeroReturnError exceptions. (The latter is
usually mapped to a zero return from read.) In OpenSSL 1.1.1, the ssl
module would raise them for transport EOF and close_notify,
respectively. In OpenSSL 3.0, both act like close_notify.
Fix this by, instead, just detecting SSL_R_UNEXPECTED_EOF_WHILE_READING
and mapping that to the other exception type.
There doesn't seem to have been any unit test of this error, so fill in
the missing one. This had to be done with the BIO path because it's
actually slightly tricky to simulate a transport EOF with Python's fd
based APIs. (If you instruct the server to close the socket, it gets
confused, probably because the server's SSL object is still referencing
the now dead fd?)
Aside from sys and builtins, _io is the only core builtin module that hasn't been ported to multi-phase init. We may do so later (e.g. gh-101948), but in the meantime we must at least take care of the module's static types properly. (This came up while working on gh-101660.)
https://github.com/python/cpython/issues/94673
Prior to this change, errors in _Py_NewInterpreterFromConfig() were always fatal. Instead, callers should be able to handle such errors and keep going. That's what this change supports. (This was an oversight in the original implementation of _Py_NewInterpreterFromConfig().) Note that the existing [fatal] behavior of the public Py_NewInterpreter() is preserved.
https://github.com/python/cpython/issues/98608
This deprecates `st_ctime` fields on Windows, with the intent to change them to contain the correct value in 3.14. For now, they should keep returning the creation time as they always have.
Add `MS_WINDOWS_DESKTOP`, `MS_WINDOWS_APPS`, `MS_WINDOWS_SYSTEM` and `MS_WINDOWS_GAMES` preprocessor definitions to allow switching off functionality missing from particular API partitions ("partitions" are used in Windows to identify overlapping subsets of APIs).
CPython only officially supports `MS_WINDOWS_DESKTOP` and `MS_WINDOWS_SYSTEM` (APPS is included by normal desktop builds, but APPS without DESKTOP is not covered). Other configurations are a convenience for people building their own runtimes.
`MS_WINDOWS_GAMES` is for the Xbox subset of the Windows API, which is also available on client OS, but is restricted compared to `MS_WINDOWS_DESKTOP`. These restrictions may change over time, as they relate to the build headers rather than the OS support, and so we assume that Xbox builds will use the latest available version of the GDK.
Specific changes:
* move the import lock to PyInterpreterState
* move the "find_and_load" diagnostic state to PyInterpreterState
Note that the import lock exists to keep multiple imports of the same module in the same interpreter (but in different threads) from stomping on each other. Independently, we use a distinct global lock to protect globally shared import state, especially related to loaded extension modules. For now we can rely on the GIL as that lock but with a per-interpreter GIL we'll need a new global lock.
The remaining state in _PyRuntimeState.imports will (probably) continue being global.
https://github.com/python/cpython/issues/100227
This is related to fixing the refleaks introduced by commit 096d009. I haven't been able to find the leak yet, but these changes are a consequence of that effort. This includes some cleanup, some tweaks to the existing tests, and a bunch of new test cases. The only change here that might have impact outside the tests in question is in imp.py, where I update imp.load_dynamic() to use spec_from_file_location() instead of creating a ModuleSpec directly.
Also note that I've updated the tests to only skip if we're checking for refleaks (regrtest's --huntrleaks), whereas in gh-101969 I had skipped the tests entirely. The tests will be useful for some upcoming work and I'd rather the refleaks not hold that up. (It isn't clear how quickly we'll be able to fix the leaking code, though it will certainly be done in the short term.)
https://github.com/python/cpython/issues/102251
lzma.LZMADecompressor and bz2.BZ2Decompressor objects caused
segfaults when their `__init__()` methods were not called.
lzma.LZMADecompressor, lzma.LZMACompressor, bz2.BZ2Compressor,
and bz2.BZ2Decompressor objects would leak locks and internal buffers
when their `__init__()` methods were called multiple times.
https://bugs.python.org/issue23224
- partial tests for cosh/sinh overflows (L535 and L771). I doubt
both ||-ed conditions could be tested.
- removed inaccessible case in sqrt (L832): ax=ay=0 is handled
above (L823) because fabs() is exact. Also added test (checked
with mpmath and gmpy2) for second condition on that line.
- some trivial tests for isclose (cover all conditions on L1217-1218)
- add comment for uncovered L1018
Co-authored-by: Mark Dickinson <dickinsm@gmail.com>
* fileutils: handle non-blocking pipe IO on Windows
Handle erroring operations on non-blocking pipes by reading the _doserrno code.
Limit writes on non-blocking pipes that are too large.
* Support blocking functions on Windows
Use the GetNamedPipeHandleState and SetNamedPipeHandleState Win32 API functions to add support for os.get_blocking and os.set_blocking.
This merges their code. They're backed by the same single HACL* static library, having them be a single module simplifies maintenance.
This should unbreak the wasm enscripten builds that currently fail due to linking in --whole-archive mode and the HACL* library appearing twice.
Long unnoticed error fixed: _sha512.SHA384Type was doubly assigned and was actually SHA512Type. Nobody depends on those internal names.
Also rename LIBHACL_ make vars to LIBHACL_SHA2_ in preperation for other future HACL things.
Enforcing (optionally) the restriction set by PEP 489 makes sense. Furthermore, this sets the stage for a potential restriction related to a per-interpreter GIL.
This change includes the following:
* add tests for extension module subinterpreter compatibility
* add _PyInterpreterConfig.check_multi_interp_extensions
* add Py_RTFLAGS_MULTI_INTERP_EXTENSIONS
* add _PyImport_CheckSubinterpIncompatibleExtensionAllowed()
* fail iff the module does not implement multi-phase init and the current interpreter is configured to check
https://github.com/python/cpython/issues/98627
This builds HACL* as a library in one place.
A followup to #101707 which broke some WASM builds. This fixes 2/4 of them, but the enscripten toolchain in the others don't deduplicate linker arguments and error out. A follow-on PR will address those.
The new test exercises the most important variants for single-phase init extension modules. We also add some explanation about those variants to import.c.
https://github.com/python/cpython/issues/101758
Replace the builtin hashlib implementations of SHA2-384 and SHA2-512
originally from LibTomCrypt with formally verified, side-channel resistant
code from the [HACL*](https://github.com/hacl-star/hacl-star/) project.
The builtins remain a fallback only used when OpenSSL does not provide them.
Previously, we checked exclusively for `__GLIBC__` (AND'd with some other
conditions). Checking for `__linux__` instead should be fine.
This fixes using e.g. `os.listxattr()` on systems using musl libc.
Bug: https://bugs.gentoo.org/894130
Co-authored-by: Gregory P. Smith <greg@krypto.org>
`socket.getaddrinfo()` no longer raises `OverflowError` based on the **port** argument. Error reporting (or not) for its value is left up to the underlying C library `getaddrinfo()` implementation.
`math_1_to_whatever()` is no longer useful, since all existing uses of it convert to `float`.
Earlier versions of Python used `math_1_to_whatever` with an integer target; see
gh-16991 for the PR where that use was removed.
* Make sure that the current exception is always normalized.
* Remove redundant type and traceback fields for the current exception.
* Add new API functions: PyErr_GetRaisedException, PyErr_SetRaisedException
* Add new API functions: PyException_GetArgs, PyException_SetArgs
replacing hashlib primitives (for the non-OpenSSL case) with verified implementations from HACL*. This is the first PR in the series, and focuses specifically on SHA2-256 and SHA2-224.
This PR imports Hacl_Streaming_SHA2 into the Python tree. This is the HACL* implementation of SHA2, which combines a core implementation of SHA2 along with a layer of buffer management that allows updating the digest with any number of bytes. This supersedes the previous implementation in the tree.
@franziskuskiefer was kind enough to benchmark the changes: in addition to being verified (thus providing significant safety and security improvements), this implementation also provides a sizeable performance boost!
```
---------------------------------------------------------------
Benchmark Time CPU Iterations
---------------------------------------------------------------
Sha2_256_Streaming 3163 ns 3160 ns 219353 // this PR
LibTomCrypt_Sha2_256 5057 ns 5056 ns 136234 // library used by Python currently
```
The changes in this PR are as follows:
- import the subset of HACL* that covers SHA2-256/224 into `Modules/_hacl`
- rewire sha256module.c to use the HACL* implementation
Co-authored-by: Gregory P. Smith [Google LLC] <greg@krypto.org>
Co-authored-by: Erlend E. Aasland <erlend.aasland@protonmail.com>
This PR fixes the buildbot failures introduced by the merge of #5561, by restricting the relevant tests to something that should work on both 32-bit and 64-bit platforms. It also silences some compiler warnings introduced in that PR.
The summary of this diff is that it:
* adds a `_ctypes_alloc_format_padding` function to append strings like `37x` to a format string to indicate 37 padding bytes
* removes the branches that amount to "give up on producing a valid format string if the struct is packed"
* combines the resulting adjacent `if (isStruct) {`s now that neither is `if (isStruct && !isPacked) {`
* invokes `_ctypes_alloc_format_padding` to add padding between structure fields, and after the last structure field. The computation used for the total size is unchanged from ctypes already used.
This patch does not affect any existing aligment computation; all it does is use subtraction to deduce the amount of paddnig introduced by the existing code.
---
Without this fix, it would never include padding bytes - an assumption that was only
valid in the case when `_pack_` was set - and this case was explicitly not implemented.
This should allow conversion from ctypes structs to numpy structs
Fixes https://github.com/numpy/numpy/issues/10528
A PyThreadState can be in one of many states in its lifecycle, represented by some status value. Those statuses haven't been particularly clear, so we're addressing that here. Specifically:
* made the distinct lifecycle statuses clear on PyThreadState
* identified expectations of how various lifecycle-related functions relate to status
* noted the various places where those expectations don't match the actual behavior
At some point we'll need to address the mismatches.
(This change also includes some cleanup.)
https://github.com/python/cpython/issues/59956
To use this, ensure that clang support was selected in Visual Studio Installer, then set the PlatformToolset environment variable to "ClangCL" and build as normal from the command line.
It remains unsupported, but at least is possible now for experimentation.
Fixes a reference counting issue with `ctypes.Structure` when a `from_param()` method call is used and the structure size is larger than a C pointer `sizeof(void*)`.
This problem existed for a very long time, but became more apparent in 3.8+ by change likely due to garbage collection cleanup timing changes.
When getaddrinfo returns an error, the output pointer is in an unknown state
Don't call freeaddrinfo on it. See the issue for discussion and details with
links to reasoning. _Most_ libc getaddrinfo implementations never modify the
output pointer unless they are returning success.
Co-authored-by: Sergey G. Brester <github@sebres.de>
Co-authored-by: Oleg Iarygin <dralife@yandex.ru>
When testing element truth values, emit a DeprecationWarning in all implementations.
This had emitted a FutureWarning in the rarely used python-only implementation since ~2.7 and has always been documented as a behavior not to rely on.
Matching an element in a tree search but having it test False can be unexpected. Raising the warning enables making the choice to finally raise an exception for this ambiguous behavior in the future.
The objective of this change is to help make the GILState-related code easier to understand. This mostly involves moving code around and some semantically equivalent refactors. However, there are a also a small number of slight changes in structure and behavior:
* tstate_current is moved out of _PyRuntimeState.gilstate
* autoTSSkey is moved out of _PyRuntimeState.gilstate
* autoTSSkey is initialized earlier
* autoTSSkey is re-initialized (after fork) earlier
https://github.com/python/cpython/issues/59956
Have _posixsubprocess.c stop using boolean flags to say if gid and uid values were supplied and action is required. Such an implicit "either initialized or look somewhere else" confused both the reader (another mental connection to constantly track between functions) and a compiler (warnings on potentially uninitialized variables being passed). Instead, we can utilize a special group/user id as a flag value -1 defined by POSIX but used nowhere else. Namely:
gid: call_setgid = False → gid = -1
uid: call_setuid = False → uid = -1
groups: call_setgroups = False → groups = NULL (obtained with (groups_list != Py_None) ? groups : NULL)
This PR is required for #94519.
This PR removes the `volatile` qualifier on various intermediate quantities
in the `math.fsum` implementation, and updates the notes preceding the
algorithm accordingly (as well as fixing some of the exsting notes). See
the linked issue #100833 for discussion.
Not comprehensive, best effort warning. There are cases when threads exist on some platforms that this code cannot detect. macOS when API permissions allow and Linux with a readable /proc procfs present are the currently supported cases where a warning should show up reliably.
Starting with a DeprecationWarning for now, it is less disruptive than something like RuntimeWarning and most likely to only be seen in people's CI tests - a good place to start with this messaging.
The itemsize returned in a memoryview of a ctypes array is now computed from the item type, instead of dividing the total size by the length and assuming that the length is not zero.
