* Switch PyUnicode_InternInPlace to _PyUnicode_InternMortal, clarify docs
* Document immortality in some functions that take `const char *`
This is PyUnicode_InternFromString;
PyDict_SetItemString, PyObject_SetAttrString;
PyObject_DelAttrString; PyUnicode_InternFromString;
and the PyModule_Add convenience functions.
Always point out a non-immortalizing alternative.
* Don't immortalize user-provided attr names in _ctypes
We should maintain the invariant that a zero `ob_tid` implies the
refcount fields are merged.
* Move the assignment in `_Py_MergeZeroLocalRefcount` to immediately
before the refcount merge.
* Update `_PyTrash_thread_destroy_chain` to set `ob_ref_shared` to
`_Py_REF_MERGED` when setting `ob_tid` to zero.
Also check this invariant with assertions in the GC in debug builds.
That uncovered a bug when running out of memory during GC.
They are alternate constructors which only accept numbers
(including objects with special methods __float__, __complex__
and __index__), but not strings.
It is our general practice to make new optional parameters keyword-only,
even if the existing parameters are all positional-or-keyword. Passing
this parameter as positional would look confusing and could be error-prone
if additional parameters are added in the future.
Performance improvement to `float.fromhex`: use a lookup table
for computing the hexadecimal value of a character, in place of the
previous switch-case construct. Patch by Bruno Lima.
On POSIX systems, excluding macOS framework installs, the lib directory
for the free-threaded build now includes a "t" suffix to avoid conflicts
with a co-located default build installation.
When builtin static types are initialized for a subinterpreter, various "tp" slots have already been inherited (for the main interpreter). This was interfering with the logic in add_operators() (in Objects/typeobject.c), causing a wrapper to get created when it shouldn't. This change fixes that by preserving the original data from the static type struct and checking that.
The `_PySeqLock_EndRead` function needs an acquire fence to ensure that
the load of the sequence happens after any loads within the read side
critical section. The missing fence can trigger bugs on macOS arm64.
Additionally, we need a release fence in `_PySeqLock_LockWrite` to
ensure that the sequence update is visible before any modifications to
the cache entry.
Make error message for index() methods consistent
Remove the repr of the searched value (which can be arbitrary large)
from ValueError messages for list.index(), range.index(), deque.index(),
deque.remove() and ShareableList.index(). Make the error messages
consistent with error messages for other index() and remove()
methods.
This reduces the system call count of a simple program[0] that reads all
the `.rst` files in Doc by over 10% (5706 -> 4734 system calls on my
linux system, 5813 -> 4875 on my macOS)
This reduces the number of `fstat()` calls always and seek calls most
the time. Stat was always called twice, once at open (to error early on
directories), and a second time to get the size of the file to be able
to read the whole file in one read. Now the size is cached with the
first call.
The code keeps an optimization that if the user had previously read a
lot of data, the current position is subtracted from the number of bytes
to read. That is somewhat expensive so only do it on larger files,
otherwise just try and read the extra bytes and resize the PyBytes as
needeed.
I built a little test program to validate the behavior + assumptions
around relative costs and then ran it under `strace` to get a log of the
system calls. Full samples below[1].
After the changes, this is everything in one `filename.read_text()`:
```python3
openat(AT_FDCWD, "cpython/Doc/howto/clinic.rst", O_RDONLY|O_CLOEXEC) = 3`
fstat(3, {st_mode=S_IFREG|0644, st_size=343, ...}) = 0`
ioctl(3, TCGETS, 0x7ffdfac04b40) = -1 ENOTTY (Inappropriate ioctl for device)
lseek(3, 0, SEEK_CUR) = 0
read(3, ":orphan:\n\n.. This page is retain"..., 344) = 343
read(3, "", 1) = 0
close(3) = 0
```
This does make some tradeoffs
1. If the file size changes between open() and readall(), this will
still get all the data but might have more read calls.
2. I experimented with avoiding the stat + cached result for small files
in general, but on my dev workstation at least that tended to reduce
performance compared to using the fstat().
[0]
```python3
from pathlib import Path
nlines = []
for filename in Path("cpython/Doc").glob("**/*.rst"):
nlines.append(len(filename.read_text()))
```
[1]
Before small file:
```
openat(AT_FDCWD, "cpython/Doc/howto/clinic.rst", O_RDONLY|O_CLOEXEC) = 3
fstat(3, {st_mode=S_IFREG|0644, st_size=343, ...}) = 0
ioctl(3, TCGETS, 0x7ffe52525930) = -1 ENOTTY (Inappropriate ioctl for device)
lseek(3, 0, SEEK_CUR) = 0
lseek(3, 0, SEEK_CUR) = 0
fstat(3, {st_mode=S_IFREG|0644, st_size=343, ...}) = 0
read(3, ":orphan:\n\n.. This page is retain"..., 344) = 343
read(3, "", 1) = 0
close(3) = 0
```
After small file:
```
openat(AT_FDCWD, "cpython/Doc/howto/clinic.rst", O_RDONLY|O_CLOEXEC) = 3
fstat(3, {st_mode=S_IFREG|0644, st_size=343, ...}) = 0
ioctl(3, TCGETS, 0x7ffdfac04b40) = -1 ENOTTY (Inappropriate ioctl for device)
lseek(3, 0, SEEK_CUR) = 0
read(3, ":orphan:\n\n.. This page is retain"..., 344) = 343
read(3, "", 1) = 0
close(3) = 0
```
Before large file:
```
openat(AT_FDCWD, "cpython/Doc/c-api/typeobj.rst", O_RDONLY|O_CLOEXEC) = 3
fstat(3, {st_mode=S_IFREG|0644, st_size=133104, ...}) = 0
ioctl(3, TCGETS, 0x7ffe52525930) = -1 ENOTTY (Inappropriate ioctl for device)
lseek(3, 0, SEEK_CUR) = 0
lseek(3, 0, SEEK_CUR) = 0
fstat(3, {st_mode=S_IFREG|0644, st_size=133104, ...}) = 0
read(3, ".. highlight:: c\n\n.. _type-struc"..., 133105) = 133104
read(3, "", 1) = 0
close(3) = 0
```
After large file:
```
openat(AT_FDCWD, "cpython/Doc/c-api/typeobj.rst", O_RDONLY|O_CLOEXEC) = 3
fstat(3, {st_mode=S_IFREG|0644, st_size=133104, ...}) = 0
ioctl(3, TCGETS, 0x7ffdfac04b40) = -1 ENOTTY (Inappropriate ioctl for device)
lseek(3, 0, SEEK_CUR) = 0
lseek(3, 0, SEEK_CUR) = 0
read(3, ".. highlight:: c\n\n.. _type-struc"..., 133105) = 133104
read(3, "", 1) = 0
close(3) = 0
```
Co-authored-by: Shantanu <12621235+hauntsaninja@users.noreply.github.com>
Co-authored-by: Erlend E. Aasland <erlend.aasland@protonmail.com>
Co-authored-by: Victor Stinner <vstinner@python.org>
As noted in gh-117983, the import importlib.util can be triggered at
interpreter startup under some circumstances, so adding threading makes
it a potentially obligatory load.
Lazy loading is not used in the stdlib, so this removes an unnecessary
load for the majority of users and slightly increases the cost of the
first lazily loaded module.
An obligatory threading load breaks gevent, which monkeypatches the
stdlib. Although unsupported, there doesn't seem to be an offsetting
benefit to breaking their use case.
For reference, here are benchmarks for the current main branch:
```
❯ hyperfine -w 8 './python -c "import importlib.util"'
Benchmark 1: ./python -c "import importlib.util"
Time (mean ± σ): 9.7 ms ± 0.7 ms [User: 7.7 ms, System: 1.8 ms]
Range (min … max): 8.4 ms … 13.1 ms 313 runs
```
And with this patch:
```
❯ hyperfine -w 8 './python -c "import importlib.util"'
Benchmark 1: ./python -c "import importlib.util"
Time (mean ± σ): 8.4 ms ± 0.7 ms [User: 6.8 ms, System: 1.4 ms]
Range (min … max): 7.2 ms … 11.7 ms 352 runs
```
Compare to:
```
❯ hyperfine -w 8 './python -c pass'
Benchmark 1: ./python -c pass
Time (mean ± σ): 7.6 ms ± 0.6 ms [User: 5.9 ms, System: 1.6 ms]
Range (min … max): 6.7 ms … 11.3 ms 390 runs
```
This roughly halves the import time of importlib.util.
