Now that our int<->str conversions are size limited and we have the
_pylong module handling larger integers, we don't need to limit
everything just to avoid wasting time in the quadratic time DoS-like
case while fuzzing.
We can tweak these further after seeing how this goes.
In very rare circumstances the JUMP opcode could be confused with the
argument of the opcode in the "then" part which doesn't end with the
JUMP opcode. This led to incorrect detection of the final JUMP opcode
and incorrect calculation of the size of the subexpression.
NOTE: Changed return value of functions _validate_inner() and
_validate_charset() in Modules/_sre/sre.c. Now they return 0 on success,
-1 on failure, and 1 if the last op is JUMP (which usually is a failure).
Previously they returned 1 on success and 0 on failure.
Previously, the optional restrictions on subinterpreters were: disallow fork, subprocess, and threads. By default, we were disallowing all three for "isolated" interpreters. We always allowed all three for the main interpreter and those created through the legacy `Py_NewInterpreter()` API.
Those settings were a bit conservative, so here we've adjusted the optional restrictions to: fork, exec, threads, and daemon threads. The default for "isolated" interpreters disables fork, exec, and daemon threads. Regular threads are allowed by default. We continue always allowing everything For the main interpreter and the legacy API.
In the code, we add `_PyInterpreterConfig.allow_exec` and `_PyInterpreterConfig.allow_daemon_threads`. We also add `Py_RTFLAGS_DAEMON_THREADS` and `Py_RTFLAGS_EXEC`.
(see https://github.com/python/cpython/issues/98608)
This change does the following:
1. change the argument to a new `_PyInterpreterConfig` struct
2. rename the function to `_Py_NewInterpreterFromConfig()`, inspired by `Py_InitializeFromConfig()` (takes a `_PyInterpreterConfig` instead of `isolated_subinterpreter`)
3. split up the boolean `isolated_subinterpreter` into the corresponding multiple granular settings
* allow_fork
* allow_subprocess
* allow_threads
4. add `PyInterpreterState.feature_flags` to store those settings
5. add a function for checking if a feature is enabled on an opaque `PyInterpreterState *`
6. drop `PyConfig._isolated_interpreter`
The existing default (see `Py_NewInterpeter()` and `Py_Initialize*()`) allows fork, subprocess, and threads and the optional "isolated" interpreter (see the `_xxsubinterpreters` module) disables all three. None of that changes here; the defaults are preserved.
Note that the given `_PyInterpreterConfig` will not be used outside `_Py_NewInterpreterFromConfig()`, nor preserved. This contrasts with how `PyConfig` is currently preserved, used, and even modified outside `Py_InitializeFromConfig()`. I'd rather just avoid that mess from the start for `_PyInterpreterConfig`. We can preserve it later if we find an actual need.
This change allows us to follow up with a number of improvements (e.g. stop disallowing subprocess and support disallowing exec instead).
(Note that this PR adds "private" symbols. We'll probably make them public, and add docs, in a separate change.)
Functions re.sub() and re.subn() and corresponding re.Pattern methods
are now 2-3 times faster for replacement strings containing group references.
Closes#91524
Primarily authored by serhiy-storchaka Serhiy Storchaka
Minor-cleanups-by: Gregory P. Smith [Google] <greg@krypto.org>
Added os.setns and os.unshare to easily switch between namespaces
on Linux.
Co-authored-by: Christian Heimes <christian@python.org>
Co-authored-by: CAM Gerlach <CAM.Gerlach@Gerlach.CAM>
Co-authored-by: Victor Stinner <vstinner@python.org>
The os module and the PyUnicode_FSDecoder() function no longer accept
bytes-like paths, like bytearray and memoryview types: only the exact
bytes type is accepted for bytes strings.
Change summary:
+ There is now a `gzip.READ_BUFFER_SIZE` constant that is 128KB. Other programs that read in 128KB chunks: pigz and cat. So this seems best practice among good programs. Also it is faster than 8 kb chunks.
+ a zlib._ZlibDecompressor was added. This is the _bz2.BZ2Decompressor ported to zlib. Since the zlib.Decompress object is better for in-memory decompression, the _ZlibDecompressor is hidden. It only makes sense in file decompression, and that is already implemented now in the gzip library. No need to bother the users with this.
+ The ZlibDecompressor uses the older Cpython arrange_output_buffer functions, as those are faster and more appropriate for the use case.
+ GzipFile.read has been optimized. There is no longer a `unconsumed_tail` member to write back to padded file. This is instead handled by the ZlibDecompressor itself, which has an internal buffer. `_add_read_data` has been inlined, as it was just two calls.
EDIT: While I am adding improvements anyway, I figured I could add another one-liner optimization now to the python -m gzip application. That read chunks in io.DEFAULT_BUFFER_SIZE previously, but has been updated now to use READ_BUFFER_SIZE chunks.
On macOS, fix a crash in syslog.syslog() in multi-threaded
applications. On macOS, the libc syslog() function is not
thread-safe, so syslog.syslog() no longer releases the GIL to call
it.
Signal wakeup fd errors are now logged with
_PyErr_WriteUnraisableMsg(), rather than PySys_WriteStderr() and
PyErr_WriteUnraisable(), to pass the error message to
sys.unraisablehook. By default, it's still written into stderr (unless
sys.unraisablehook is overriden).
This seems pretty straightforward. The issue mentions other calls in mmapmodule that we could release the GIL on, but those are in methods where we'd need to be careful to ensure that something sensible happens if those are called concurrently. In prior art, note that #12073 released the GIL for munmap. In a toy benchmark, I see the speedup you'd expect from doing this.
Automerge-Triggered-By: GH:gvanrossum
* gh-96821: Fix undefined behaviour in `audioop.c`
Left-shifting negative numbers is undefined behaviour.
Fortunately, multiplication works just as well, is defined behaviour,
and gets compiled to the same machine code as before by optimizing
compilers.
Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com>
These were intentionally skipped when operator was updated to use the argument clinic:
https://github.com/python/cpython/issues/64385#issuecomment-1093641466
However, by not using the argument clinic, they missed out on getting signatures.
This is a narrow PR to update the docstrings so that `__text_signature__` can be
extracted from them. Updating to use the argument clinic is beyond scope.
`methodcaller` uses `*args, **kwargs` to match variadic names used elsewhere,
including in `operator.call`.
The macOS 13 SDK includes support for the `mkfifoat` and `mknodat` system calls.
Using the `dir_fd` option with either `os.mkfifo` or `os.mknod` could result in a
segfault if cpython is built with the macOS 13 SDK but run on an earlier
version of macOS. Prevent this by adding runtime support for detection of
these system calls ("weaklinking") as is done for other newer syscalls on
macOS.
Fix the Python path configuration used to initialized sys.path at
Python startup. Paths are no longer encoded to UTF-8/strict to avoid
encoding errors if it contains surrogate characters (bytes paths are
decoded with the surrogateescape error handler).
getpath_basename() and getpath_dirname() functions no longer encode
the path to UTF-8/strict, but work directly on Unicode strings. These
functions now use PyUnicode_FindChar() and PyUnicode_Substring() on
the Unicode path, rather than strrchr() on the encoded bytes string.
This PR fixes undefined behaviour in the struct module unpacking support functions `bu_longlong`, `lu_longlong`, `bu_int` and `lu_int`; thanks to @kumaraditya303 for finding these.
The fix is to accumulate the bytes in an unsigned integer type instead of a signed integer type, then to convert to the appropriate signed type. In cases where the width matches, that conversion will typically be compiled away to a no-op.
(Evidence from Godbolt: https://godbolt.org/z/5zvxodj64 .)
To make the conversions efficient, I've specialised the relevant functions for their output size: for `bu_longlong` and `lu_longlong`, this only entails checking that the output size is indeed `8`. But `bu_int` and `lu_int` were used for format sizes `2` and `4` - I've split those into two separate functions each.
No tests, because all of the affected cases are already exercised by the test suite.
Fix the faulthandler implementation of faulthandler.register(signal,
chain=True) if the sigaction() function is not available: don't call
the previous signal handler if it's NULL.
⚠️⚠️ Note for reviewers, hackers and fellow systems/low-level/compiler engineers ⚠️⚠️
If you have a lot of experience with this kind of shenanigans and want to improve the **first** version, **please make a PR against my branch** or **reach out by email** or **suggest code changes directly on GitHub**.
If you have any **refinements or optimizations** please, wait until the first version is merged before starting hacking or proposing those so we can keep this PR productive.
datetime.isoformat generates the tzoffset with colons, but there
was no format code to make strftime output the same format.
for simplicity and consistency the %:z formatting behaves mostly
as %z, with the exception of adding colons. this includes the
dynamic behaviour of adding seconds and microseconds only when
needed (when not 0).
this fixes the still open "generate" part of this issue:
https://github.com/python/cpython/issues/69142
Co-authored-by: Kumar Aditya <59607654+kumaraditya303@users.noreply.github.com>
- Limited API needs to be enabled per source file
- Some builds don't support Limited API, so Limited API tests must be skipped on those builds
(currently this is `Py_TRACE_REFS`, but that may change.)
- `Py_LIMITED_API` must be defined before `<Python.h>` is included.
This puts the hoop-jumping in `testcapi/parts.h`, so individual
test files can be relatively simple. (Currently that's only
`vectorcall_limited.c`, imagine more.)
- On WASI `ENOTCAPABLE` is now mapped to `PermissionError`.
- The `errno` modules exposes the new error number.
- `getpath.py` now ignores `PermissionError` when it cannot open landmark
files `pybuilddir.txt` and `pyenv.cfg`.
* Make sure that tp_dictoffset is correct with Py_TPFLAGS_MANAGED_DICT is set.
* Avoid traversing managed dict twice when subclassing class with Py_TPFLAGS_MANAGED_DICT set.
We only statically initialize for core code and builtin modules. Extension modules still create
the tuple at runtime. We'll solve that part of interpreter isolation separately.
This change includes generated code. The non-generated changes are in:
* Tools/clinic/clinic.py
* Python/getargs.c
* Include/cpython/modsupport.h
* Makefile.pre.in (re-generate global strings after running clinic)
* very minor tweaks to Modules/_codecsmodule.c and Python/Python-tokenize.c
All other changes are generated code (clinic, global strings).
- Move PyUnicode tests to a separate file
- Add some more tests for PyUnicode_FromFormat
Co-authored-by: philg314 <110174000+philg314@users.noreply.github.com>
* Add test for inheriting explicit __dict__ and weakref.
* Restore 3.10 behavior for multiple inheritance of C extension classes that store their dictionary at the end of the struct.
- check for ``dup()`` libc function
- handle missing ``F_DUPFD`` in ``dup2()`` replacement function
- add workaround for WASI libc bug in MSG_TRUNC
- ESHUTDOWN is missing, use EPIPE instead
- POLLPRI is missing, define as 0 (no-op)
* syslog_get_argv() swallows exceptions, but not in all cases.
* if ident is non UTF-8 encodable, syslog.openlog() fails after setting the
global reference to ident. Now the C string saved internally in the previous
call to openlog() points to the freed memory.
* PySys_Audit() can crash if ident is NULL.
* There may be a race condition with syslog.syslog(), because the global
reference to ident is decrefed before setting the new value.
* Possible use of freed memory if syslog.openlog() is called while
the GIL is released in syslog.syslog().
This PR partially reverts gh-24421 (PR) and fixes the remaining concerns
given in gh-93044 (issue):
- keyword arguments are passed as positional arguments to factory()
- if an argument is not passed to sqlite3.connect(), its default value
is passed to factory()
Co-authored-by: Serhiy Storchaka <storchaka@gmail.com>
The wrapper macros are more readable and match the form recommended in
the OpenSSL documentation. They also slightly less error-prone, as the
mapping of arguments to SSL_CTX_ctrl is not always clear. (Though in
this case it's straightforward.)
https://www.openssl.org/docs/man1.1.1/man3/SSL_CTX_get_max_proto_version.html
When binding a unix socket to an empty address on Linux, the socket is
automatically bound to an available address in the abstract namespace.
>>> s = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
>>> s.bind("")
>>> s.getsockname()
b'\x0075499'
Since python 3.9, the socket is bound to the one address:
>>> s.getsockname()
b'\x00'
And trying to bind multiple sockets will fail with:
Traceback (most recent call last):
File "/home/nsoffer/src/cpython/Lib/test/test_socket.py", line 5553, in testAutobind
s2.bind("")
OSError: [Errno 98] Address already in use
Added 2 tests:
- Auto binding empty address on Linux
- Failing to bind an empty address on other platforms
Fixes f6b3a07b7d (bpo-44493: Add missing terminated NUL in sockaddr_un's length (GH-26866)
The `_testcapimodule.c` file is getting too large to work with effectively.
This PR lays out a general structure of how tests can be split up, with more splitting to come later if the structure is OK.
Vectorcall tests aren't the biggest issue -- it's just an area I want to work on next, so I'm starting here.
An issue specific to vectorcall tests is that it wasn't clear that e.g. `MethodDescriptor2` is related to testing vectorcall: the `/* Test PEP 590 */` section had an ambiguous end. Separate file should make things like this much clearer.
OTOH, for some pieces it might not be clear where they should be -- I left `meth_fastcall` with tests of the other calling conventions. IMO, even with the ambiguity it's still worth it to split the huge file up.
I'm not sure about the buildsystem changes, hopefully CI will tell me what's wrong.
@vstinner, @markshannon: Do you think this is a good idea?
Automerge-Triggered-By: GH:encukou
Nikita Sobolev
Gregory P. Smith
Kumar Aditya
Shantanu
Serhiy Storchaka
Steve Dower
Jelle Zijlstra
Eric Snow
Charlie Zhao
David Buchanan
Shaun Walbridge
Wenzel Jakob
Benjamin Peterson
Raymond Hettinger
Furkan Onder
Carl Meyer
Noam Cohen
Victor Stinner
Ruben Vorderman
Matthias Görgens
Michael Droettboom
Pablo Galindo Salgado
Hagai Helman Tov
Stanley
Guido van Rossum
Dong-hee Na
Erik Welch
Ned Deily
Mark Shannon
MonadChains
Ofey Chan
Michael Curran
chgnrdv
Mark Dickinson
adphrost
Dennis Sweeney
TW
Ansab Gillani
Carl Friedrich Bolz-Tereick
Christian Heimes
Irit Katriel
Petr Viktorin
Yury Selivanov
Erlend E. Aasland
Tim Gates
Aivars Kalvāns
Ken Jin
Łukasz Langa
Oleg Iarygin
Raghunandan Bhat
Erlend Egeberg Aasland
David Benjamin
Brandt Bucher
Nir Soffer