From a81c8564365d4485bcf1d413b92fb275c091831d Mon Sep 17 00:00:00 2001 From: "Gregory P. Smith" Date: Sun, 3 Jun 2012 14:30:44 -0700 Subject: Fixes Issue #14992: os.makedirs(path, exist_ok=True) would raise an OSError when the path existed and had the S_ISGID mode bit set when it was not explicitly asked for. This is no longer an exception as mkdir cannot control if the OS sets that bit for it or not. --- Lib/os.py | 16 ++++++++++++++-- Lib/test/test_os.py | 40 ++++++++++++++++++++++++++++++++++------ Misc/NEWS | 5 +++++ 3 files changed, 53 insertions(+), 8 deletions(-) diff --git a/Lib/os.py b/Lib/os.py index 5862383..864edfd 100644 --- a/Lib/os.py +++ b/Lib/os.py @@ -152,8 +152,20 @@ def makedirs(name, mode=0o777, exist_ok=False): mkdir(name, mode) except OSError as e: import stat as st - if not (e.errno == errno.EEXIST and exist_ok and path.isdir(name) and - st.S_IMODE(lstat(name).st_mode) == _get_masked_mode(mode)): + dir_exists = path.isdir(name) + expected_mode = _get_masked_mode(mode) + if dir_exists: + # S_ISGID is automatically copied by the OS from parent to child + # directories on mkdir. Don't consider it being set to be a mode + # mismatch as mkdir does not unset it when not specified in mode. + actual_mode = st.S_IMODE(lstat(name).st_mode) & ~st.S_ISGID + else: + actual_mode = -1 + if not (e.errno == errno.EEXIST and exist_ok and dir_exists and + actual_mode == expected_mode): + if dir_exists and actual_mode != expected_mode: + e.strerror += ' (mode %o != expected mode %o)' % ( + actual_mode, expected_mode) raise def removedirs(name): diff --git a/Lib/test/test_os.py b/Lib/test/test_os.py index 8bc8ba9..605d2a4 100644 --- a/Lib/test/test_os.py +++ b/Lib/test/test_os.py @@ -15,6 +15,7 @@ from test import support import contextlib import mmap import uuid +import stat from test.script_helper import assert_python_ok # Detect whether we're on a Linux system that uses the (now outdated @@ -574,12 +575,39 @@ class MakedirTests(unittest.TestCase): path = os.path.join(support.TESTFN, 'dir1') mode = 0o777 old_mask = os.umask(0o022) - os.makedirs(path, mode) - self.assertRaises(OSError, os.makedirs, path, mode) - self.assertRaises(OSError, os.makedirs, path, mode, exist_ok=False) - self.assertRaises(OSError, os.makedirs, path, 0o776, exist_ok=True) - os.makedirs(path, mode=mode, exist_ok=True) - os.umask(old_mask) + try: + os.makedirs(path, mode) + self.assertRaises(OSError, os.makedirs, path, mode) + self.assertRaises(OSError, os.makedirs, path, mode, exist_ok=False) + self.assertRaises(OSError, os.makedirs, path, 0o776, exist_ok=True) + os.makedirs(path, mode=mode, exist_ok=True) + finally: + os.umask(old_mask) + + def test_exist_ok_s_isgid_directory(self): + path = os.path.join(support.TESTFN, 'dir1') + S_ISGID = stat.S_ISGID + mode = 0o777 + old_mask = os.umask(0o022) + try: + existing_testfn_mode = stat.S_IMODE( + os.lstat(support.TESTFN).st_mode) + os.chmod(support.TESTFN, existing_testfn_mode | S_ISGID) + if (os.lstat(support.TESTFN).st_mode & S_ISGID != S_ISGID): + raise unittest.SkipTest('No support for S_ISGID dir mode.') + # The os should apply S_ISGID from the parent dir for us, but + # this test need not depend on that behavior. Be explicit. + os.makedirs(path, mode | S_ISGID) + # http://bugs.python.org/issue14992 + # Should not fail when the bit is already set. + os.makedirs(path, mode, exist_ok=True) + # remove the bit. + os.chmod(path, stat.S_IMODE(os.lstat(path).st_mode) & ~S_ISGID) + with self.assertRaises(OSError): + # Should fail when the bit is not already set when demanded. + os.makedirs(path, mode | S_ISGID, exist_ok=True) + finally: + os.umask(old_mask) def test_exist_ok_existing_regular_file(self): base = support.TESTFN diff --git a/Misc/NEWS b/Misc/NEWS index 3f99773..62c1312 100644 --- a/Misc/NEWS +++ b/Misc/NEWS @@ -10,6 +10,11 @@ What's New in Python 3.2.4 Core and Builtins ----------------- +- Issue #14992: os.makedirs(path, exist_ok=True) would raise an OSError + when the path existed and had the S_ISGID mode bit set when it was + not explicitly asked for. This is no longer an exception as mkdir + cannot control if the OS sets that bit for it or not. + - Issue #14775: Fix a potential quadratic dict build-up due to the garbage collector repeatedly trying to untrack dicts. -- cgit v0.12 From 61ed804cd747d680606efb59dc2c81559a5b0c33 Mon Sep 17 00:00:00 2001 From: "Gregory P. Smith" Date: Sun, 3 Jun 2012 14:36:01 -0700 Subject: Move the 14992 note to the correct section. --- Misc/NEWS | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/Misc/NEWS b/Misc/NEWS index 62c1312..1fd7901 100644 --- a/Misc/NEWS +++ b/Misc/NEWS @@ -10,11 +10,6 @@ What's New in Python 3.2.4 Core and Builtins ----------------- -- Issue #14992: os.makedirs(path, exist_ok=True) would raise an OSError - when the path existed and had the S_ISGID mode bit set when it was - not explicitly asked for. This is no longer an exception as mkdir - cannot control if the OS sets that bit for it or not. - - Issue #14775: Fix a potential quadratic dict build-up due to the garbage collector repeatedly trying to untrack dicts. @@ -75,6 +70,11 @@ Core and Builtins Library ------- +- Issue #14992: os.makedirs(path, exist_ok=True) would raise an OSError + when the path existed and had the S_ISGID mode bit set when it was + not explicitly asked for. This is no longer an exception as mkdir + cannot control if the OS sets that bit for it or not. + - Issue #14962: Update text coloring in IDLE shell window after changing options. Patch by Roger Serwy. -- cgit v0.12 From 2d7d56abf84b8fa0ed425c01b1cae62918fd850a Mon Sep 17 00:00:00 2001 From: "Gregory P. Smith" Date: Sun, 3 Jun 2012 14:39:26 -0700 Subject: Revert the modification of e.strerror in 3.2 as that kind of change could break someone's over specified test that depends on the exact error message. --- Lib/os.py | 3 --- 1 file changed, 3 deletions(-) diff --git a/Lib/os.py b/Lib/os.py index 864edfd..d1101a2 100644 --- a/Lib/os.py +++ b/Lib/os.py @@ -163,9 +163,6 @@ def makedirs(name, mode=0o777, exist_ok=False): actual_mode = -1 if not (e.errno == errno.EEXIST and exist_ok and dir_exists and actual_mode == expected_mode): - if dir_exists and actual_mode != expected_mode: - e.strerror += ' (mode %o != expected mode %o)' % ( - actual_mode, expected_mode) raise def removedirs(name): -- cgit v0.12 From ca75b000695f4d98cc4ee8264734e17c9b05cd63 Mon Sep 17 00:00:00 2001 From: Benjamin Peterson Date: Sun, 3 Jun 2012 18:15:15 -0700 Subject: __GNUC__ does not imply gcc version is present, so just check for version (closes #14994) --- Include/pyerrors.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/Include/pyerrors.h b/Include/pyerrors.h index fb6281c..cfae922 100644 --- a/Include/pyerrors.h +++ b/Include/pyerrors.h @@ -87,7 +87,7 @@ PyAPI_FUNC(void) PyErr_GetExcInfo(PyObject **, PyObject **, PyObject **); PyAPI_FUNC(void) PyErr_SetExcInfo(PyObject *, PyObject *, PyObject *); #if defined(__clang__) || \ - (defined(__GNUC__) && \ + (defined(__GNUC_MAJOR__) && \ ((__GNUC_MAJOR__ >= 3) || \ (__GNUC_MAJOR__ == 2) && (__GNUC_MINOR__ >= 5))) #define _Py_NO_RETURN __attribute__((__noreturn__)) -- cgit v0.12 From 0c9050c25db74108d1fa4051bca4d16d2394f826 Mon Sep 17 00:00:00 2001 From: Raymond Hettinger Date: Mon, 4 Jun 2012 00:21:14 -0700 Subject: Separate key creation logic from the sequence class that memoizes its hash value. --- Lib/functools.py | 43 ++++++++++++++++++++++++------------------- 1 file changed, 24 insertions(+), 19 deletions(-) diff --git a/Lib/functools.py b/Lib/functools.py index 9f024f1..226a46e 100644 --- a/Lib/functools.py +++ b/Lib/functools.py @@ -142,30 +142,35 @@ except ImportError: _CacheInfo = namedtuple("CacheInfo", ["hits", "misses", "maxsize", "currsize"]) -class _CacheKey(list): - 'Make a cache key from optionally typed positional and keyword arguments' - +class _HashedSeq(list): __slots__ = 'hashvalue' - def __init__(self, args, kwds, typed, - kwd_mark = (object(),), - sorted=sorted, tuple=tuple, type=type, hash=hash): - key = args - if kwds: - sorted_items = sorted(kwds.items()) - key += kwd_mark - for item in sorted_items: - key += item - if typed: - key += tuple(type(v) for v in args) - if kwds: - key += tuple(type(v) for k, v in sorted_items) - self[:] = key - self.hashvalue = hash(key) # so we only have to hash just once + def __init__(self, tup, hash=hash): + self[:] = tup + self.hashvalue = hash(tup) def __hash__(self): return self.hashvalue +def _make_key(args, kwds, typed, + kwd_mark = (object(),), + fasttypes = {int, str, frozenset, type(None)}, + sorted=sorted, tuple=tuple, type=type, len=len): + 'Make a cache key from optionally typed positional and keyword arguments' + key = args + if kwds: + sorted_items = sorted(kwds.items()) + key += kwd_mark + for item in sorted_items: + key += item + if typed: + key += tuple(type(v) for v in args) + if kwds: + key += tuple(type(v) for k, v in sorted_items) + elif len(key) == 1 and type(key[0]) in fasttypes: + return key[0] + return _HashedSeq(key) + def lru_cache(maxsize=128, typed=False): """Least-recently-used cache decorator. @@ -193,7 +198,7 @@ def lru_cache(maxsize=128, typed=False): # Constants shared by all lru cache instances: sentinel = object() # unique object used to signal cache misses - make_key = _CacheKey # build a key from the function arguments + make_key = _make_key # build a key from the function arguments PREV, NEXT, KEY, RESULT = 0, 1, 2, 3 # names for the link fields def decorating_function(user_function): -- cgit v0.12 From 7d74effc67281b505756e040437241e5b1436a7c Mon Sep 17 00:00:00 2001 From: Raymond Hettinger Date: Mon, 4 Jun 2012 00:32:15 -0700 Subject: Add usage note. --- Doc/library/functools.rst | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/Doc/library/functools.rst b/Doc/library/functools.rst index 1b0d82a..f5c6608 100644 --- a/Doc/library/functools.rst +++ b/Doc/library/functools.rst @@ -49,8 +49,9 @@ The :mod:`functools` module defines the following functions: Since a dictionary is used to cache results, the positional and keyword arguments to the function must be hashable. - If *maxsize* is set to None, the LRU feature is disabled and the cache - can grow without bound. + If *maxsize* is set to None, the LRU feature is disabled and the cache can + grow without bound. The LRU feature performs best when *maxsize* is a + power-of-two. If *typed* is set to True, function arguments of different types will be cached separately. For example, ``f(3)`` and ``f(3.0)`` will be treated -- cgit v0.12 From bcd304480f16f3f936a7bdbff4dfab41f8e30292 Mon Sep 17 00:00:00 2001 From: Hynek Schlawack Date: Mon, 4 Jun 2012 14:19:39 +0200 Subject: #14814: Use correct comparison for IP addresses ipaddress._BaseV4.is_unspecified() compared IP addresses using "in" which fails. --- Lib/ipaddress.py | 2 +- Lib/test/test_ipaddress.py | 1 + 2 files changed, 2 insertions(+), 1 deletion(-) diff --git a/Lib/ipaddress.py b/Lib/ipaddress.py index a8edcd1..25bcccd 100644 --- a/Lib/ipaddress.py +++ b/Lib/ipaddress.py @@ -1130,7 +1130,7 @@ class _BaseV4: """ unspecified_address = IPv4Address('0.0.0.0') if isinstance(self, _BaseAddress): - return self in unspecified_address + return self == unspecified_address return (self.network_address == self.broadcast_address == unspecified_address) diff --git a/Lib/test/test_ipaddress.py b/Lib/test/test_ipaddress.py index aff4ff9..5b7d013 100644 --- a/Lib/test/test_ipaddress.py +++ b/Lib/test/test_ipaddress.py @@ -837,6 +837,7 @@ class IpaddrUnitTest(unittest.TestCase): self.assertEqual(False, ipaddress.ip_network('128.0.0.0').is_loopback) # test addresses + self.assertEqual(True, ipaddress.ip_address('0.0.0.0').is_unspecified) self.assertEqual(True, ipaddress.ip_address('224.1.1.1').is_multicast) self.assertEqual(False, ipaddress.ip_address('240.0.0.0').is_multicast) -- cgit v0.12 312' href='#n312'>312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 
#ifdef _Py_JIT

#include "Python.h"

#include "pycore_abstract.h"
#include "pycore_call.h"
#include "pycore_ceval.h"
#include "pycore_dict.h"
#include "pycore_intrinsics.h"
#include "pycore_long.h"
#include "pycore_opcode_metadata.h"
#include "pycore_opcode_utils.h"
#include "pycore_optimizer.h"
#include "pycore_pyerrors.h"
#include "pycore_setobject.h"
#include "pycore_sliceobject.h"
#include "pycore_jit.h"

#include "jit_stencils.h"

// Memory management stuff: ////////////////////////////////////////////////////

#ifndef MS_WINDOWS
    #include <sys/mman.h>
#endif

static size_t
get_page_size(void)
{
#ifdef MS_WINDOWS
    SYSTEM_INFO si;
    GetSystemInfo(&si);
    return si.dwPageSize;
#else
    return sysconf(_SC_PAGESIZE);
#endif
}

static void
jit_error(const char *message)
{
#ifdef MS_WINDOWS
    int hint = GetLastError();
#else
    int hint = errno;
#endif
    PyErr_Format(PyExc_RuntimeWarning, "JIT %s (%d)", message, hint);
}

static char *
jit_alloc(size_t size)
{
    assert(size);
    assert(size % get_page_size() == 0);
#ifdef MS_WINDOWS
    int flags = MEM_COMMIT | MEM_RESERVE;
    char *memory = VirtualAlloc(NULL, size, flags, PAGE_READWRITE);
    int failed = memory == NULL;
#else
    int flags = MAP_ANONYMOUS | MAP_PRIVATE;
    char *memory = mmap(NULL, size, PROT_READ | PROT_WRITE, flags, -1, 0);
    int failed = memory == MAP_FAILED;
#endif
    if (failed) {
        jit_error("unable to allocate memory");
        return NULL;
    }
    return memory;
}

static int
jit_free(char *memory, size_t size)
{
    assert(size);
    assert(size % get_page_size() == 0);
#ifdef MS_WINDOWS
    int failed = !VirtualFree(memory, 0, MEM_RELEASE);
#else
    int failed = munmap(memory, size);
#endif
    if (failed) {
        jit_error("unable to free memory");
        return -1;
    }
    return 0;
}

static int
mark_executable(char *memory, size_t size)
{
    if (size == 0) {
        return 0;
    }
    assert(size % get_page_size() == 0);
    // Do NOT ever leave the memory writable! Also, don't forget to flush the
    // i-cache (I cannot begin to tell you how horrible that is to debug):
#ifdef MS_WINDOWS
    if (!FlushInstructionCache(GetCurrentProcess(), memory, size)) {
        jit_error("unable to flush instruction cache");
        return -1;
    }
    int old;
    int failed = !VirtualProtect(memory, size, PAGE_EXECUTE_READ, &old);
#else
    __builtin___clear_cache((char *)memory, (char *)memory + size);
    int failed = mprotect(memory, size, PROT_EXEC | PROT_READ);
#endif
    if (failed) {
        jit_error("unable to protect executable memory");
        return -1;
    }
    return 0;
}

static int
mark_readable(char *memory, size_t size)
{
    if (size == 0) {
        return 0;
    }
    assert(size % get_page_size() == 0);
#ifdef MS_WINDOWS
    DWORD old;
    int failed = !VirtualProtect(memory, size, PAGE_READONLY, &old);
#else
    int failed = mprotect(memory, size, PROT_READ);
#endif
    if (failed) {
        jit_error("unable to protect readable memory");
        return -1;
    }
    return 0;
}

// JIT compiler stuff: /////////////////////////////////////////////////////////

// Warning! AArch64 requires you to get your hands dirty. These are your gloves:

// value[value_start : value_start + len]
static uint32_t
get_bits(uint64_t value, uint8_t value_start, uint8_t width)
{
    assert(width <= 32);
    return (value >> value_start) & ((1ULL << width) - 1);
}

// *loc[loc_start : loc_start + width] = value[value_start : value_start + width]
static void
set_bits(uint32_t *loc, uint8_t loc_start, uint64_t value, uint8_t value_start,
         uint8_t width)
{
    assert(loc_start + width <= 32);
    // Clear the bits we're about to patch:
    *loc &= ~(((1ULL << width) - 1) << loc_start);
    assert(get_bits(*loc, loc_start, width) == 0);
    // Patch the bits:
    *loc |= get_bits(value, value_start, width) << loc_start;
    assert(get_bits(*loc, loc_start, width) == get_bits(value, value_start, width));
}

// See https://developer.arm.com/documentation/ddi0602/2023-09/Base-Instructions
// for instruction encodings:
#define IS_AARCH64_ADD_OR_SUB(I) (((I) & 0x11C00000) == 0x11000000)
#define IS_AARCH64_ADRP(I)       (((I) & 0x9F000000) == 0x90000000)
#define IS_AARCH64_BRANCH(I)     (((I) & 0x7C000000) == 0x14000000)
#define IS_AARCH64_LDR_OR_STR(I) (((I) & 0x3B000000) == 0x39000000)
#define IS_AARCH64_MOV(I)        (((I) & 0x9F800000) == 0x92800000)

// Fill all of stencil's holes in the memory pointed to by base, using the
// values in patches.
static void
patch(char *base, const Stencil *stencil, uint64_t *patches)
{
    for (uint64_t i = 0; i < stencil->holes_size; i++) {
        const Hole *hole = &stencil->holes[i];
        void *location = base + hole->offset;
        uint64_t value = patches[hole->value] + (uint64_t)hole->symbol + hole->addend;
        uint32_t *loc32 = (uint32_t *)location;
        uint64_t *loc64 = (uint64_t *)location;
        // LLD is a great reference for performing relocations... just keep in
        // mind that Tools/jit/build.py does filtering and preprocessing for us!
        // Here's a good place to start for each platform:
        // - aarch64-apple-darwin:
        //   - https://github.com/llvm/llvm-project/blob/main/lld/MachO/Arch/ARM64Common.cpp
        //   - https://github.com/llvm/llvm-project/blob/main/lld/MachO/Arch/ARM64Common.h
        // - aarch64-unknown-linux-gnu:
        //   - https://github.com/llvm/llvm-project/blob/main/lld/ELF/Arch/AArch64.cpp
        // - i686-pc-windows-msvc:
        //   - https://github.com/llvm/llvm-project/blob/main/lld/COFF/Chunks.cpp
        // - x86_64-apple-darwin:
        //   - https://github.com/llvm/llvm-project/blob/main/lld/MachO/Arch/X86_64.cpp
        // - x86_64-pc-windows-msvc:
        //   - https://github.com/llvm/llvm-project/blob/main/lld/COFF/Chunks.cpp
        // - x86_64-unknown-linux-gnu:
        //   - https://github.com/llvm/llvm-project/blob/main/lld/ELF/Arch/X86_64.cpp
        switch (hole->kind) {
            case HoleKind_IMAGE_REL_I386_DIR32:
                // 32-bit absolute address.
                // Check that we're not out of range of 32 unsigned bits:
                assert(value < (1ULL << 32));
                *loc32 = (uint32_t)value;
                continue;
            case HoleKind_ARM64_RELOC_UNSIGNED:
            case HoleKind_IMAGE_REL_AMD64_ADDR64:
            case HoleKind_R_AARCH64_ABS64:
            case HoleKind_X86_64_RELOC_UNSIGNED:
            case HoleKind_R_X86_64_64:
                // 64-bit absolute address.
                *loc64 = value;
                continue;
            case HoleKind_R_AARCH64_CALL26:
            case HoleKind_R_AARCH64_JUMP26:
                // 28-bit relative branch.
                assert(IS_AARCH64_BRANCH(*loc32));
                value -= (uint64_t)location;
                // Check that we're not out of range of 28 signed bits:
                assert((int64_t)value >= -(1 << 27));
                assert((int64_t)value < (1 << 27));
                // Since instructions are 4-byte aligned, only use 26 bits:
                assert(get_bits(value, 0, 2) == 0);
                set_bits(loc32, 0, value, 2, 26);
                continue;
            case HoleKind_R_AARCH64_MOVW_UABS_G0_NC:
                // 16-bit low part of an absolute address.
                assert(IS_AARCH64_MOV(*loc32));
                // Check the implicit shift (this is "part 0 of 3"):
                assert(get_bits(*loc32, 21, 2) == 0);
                set_bits(loc32, 5, value, 0, 16);
                continue;
            case HoleKind_R_AARCH64_MOVW_UABS_G1_NC:
                // 16-bit middle-low part of an absolute address.
                assert(IS_AARCH64_MOV(*loc32));
                // Check the implicit shift (this is "part 1 of 3"):
                assert(get_bits(*loc32, 21, 2) == 1);
                set_bits(loc32, 5, value, 16, 16);
                continue;
            case HoleKind_R_AARCH64_MOVW_UABS_G2_NC:
                // 16-bit middle-high part of an absolute address.
                assert(IS_AARCH64_MOV(*loc32));
                // Check the implicit shift (this is "part 2 of 3"):
                assert(get_bits(*loc32, 21, 2) == 2);
                set_bits(loc32, 5, value, 32, 16);
                continue;
            case HoleKind_R_AARCH64_MOVW_UABS_G3:
                // 16-bit high part of an absolute address.
                assert(IS_AARCH64_MOV(*loc32));
                // Check the implicit shift (this is "part 3 of 3"):
                assert(get_bits(*loc32, 21, 2) == 3);
                set_bits(loc32, 5, value, 48, 16);
                continue;
            case HoleKind_ARM64_RELOC_GOT_LOAD_PAGE21:
                // 21-bit count of pages between this page and an absolute address's
                // page... I know, I know, it's weird. Pairs nicely with
                // ARM64_RELOC_GOT_LOAD_PAGEOFF12 (below).
                assert(IS_AARCH64_ADRP(*loc32));
                // Number of pages between this page and the value's page:
                value = (value >> 12) - ((uint64_t)location >> 12);
                // Check that we're not out of range of 21 signed bits:
                assert((int64_t)value >= -(1 << 20));
                assert((int64_t)value < (1 << 20));
                // value[0:2] goes in loc[29:31]:
                set_bits(loc32, 29, value, 0, 2);
                // value[2:21] goes in loc[5:26]:
                set_bits(loc32, 5, value, 2, 19);
                continue;
            case HoleKind_ARM64_RELOC_GOT_LOAD_PAGEOFF12:
                // 12-bit low part of an absolute address. Pairs nicely with
                // ARM64_RELOC_GOT_LOAD_PAGE21 (above).
                assert(IS_AARCH64_LDR_OR_STR(*loc32) || IS_AARCH64_ADD_OR_SUB(*loc32));
                // There might be an implicit shift encoded in the instruction:
                uint8_t shift = 0;
                if (IS_AARCH64_LDR_OR_STR(*loc32)) {
                    shift = (uint8_t)get_bits(*loc32, 30, 2);
                    // If both of these are set, the shift is supposed to be 4.
                    // That's pretty weird, and it's never actually been observed...
                    assert(get_bits(*loc32, 23, 1) == 0 || get_bits(*loc32, 26, 1) == 0);
                }
                value = get_bits(value, 0, 12);
                assert(get_bits(value, 0, shift) == 0);
                set_bits(loc32, 10, value, shift, 12);
                continue;
        }
        Py_UNREACHABLE();
    }
}

static void
copy_and_patch(char *base, const Stencil *stencil, uint64_t *patches)
{
    memcpy(base, stencil->body, stencil->body_size);
    patch(base, stencil, patches);
}

static void
emit(const StencilGroup *group, uint64_t patches[])
{
    copy_and_patch((char *)patches[HoleValue_CODE], &group->code, patches);
    copy_and_patch((char *)patches[HoleValue_DATA], &group->data, patches);
}

// Compiles executor in-place. Don't forget to call _PyJIT_Free later!
int
_PyJIT_Compile(_PyExecutorObject *executor, const _PyUOpInstruction *trace, size_t length)
{
    // Loop once to find the total compiled size:
    size_t code_size = 0;
    size_t data_size = 0;
    for (size_t i = 0; i < length; i++) {
        _PyUOpInstruction *instruction = (_PyUOpInstruction *)&trace[i];
        const StencilGroup *group = &stencil_groups[instruction->opcode];
        code_size += group->code.body_size;
        data_size += group->data.body_size;
    }
    // Round up to the nearest page (code and data need separate pages):
    size_t page_size = get_page_size();
    assert((page_size & (page_size - 1)) == 0);
    code_size += page_size - (code_size & (page_size - 1));
    data_size += page_size - (data_size & (page_size - 1));
    char *memory = jit_alloc(code_size + data_size);
    if (memory == NULL) {
        return -1;
    }
    // Loop again to emit the code:
    char *code = memory;
    char *data = memory + code_size;
    char *top = code;
    if (trace[0].opcode == _START_EXECUTOR) {
        // Don't want to execute this more than once:
        top += stencil_groups[_START_EXECUTOR].code.body_size;
    }
    for (size_t i = 0; i < length; i++) {
        _PyUOpInstruction *instruction = (_PyUOpInstruction *)&trace[i];
        const StencilGroup *group = &stencil_groups[instruction->opcode];
        // Think of patches as a dictionary mapping HoleValue to uint64_t:
        uint64_t patches[] = GET_PATCHES();
        patches[HoleValue_CODE] = (uint64_t)code;
        patches[HoleValue_CONTINUE] = (uint64_t)code + group->code.body_size;
        patches[HoleValue_DATA] = (uint64_t)data;
        patches[HoleValue_EXECUTOR] = (uint64_t)executor;
        patches[HoleValue_OPARG] = instruction->oparg;
        patches[HoleValue_OPERAND] = instruction->operand;
        patches[HoleValue_TARGET] = instruction->target;
        patches[HoleValue_TOP] = (uint64_t)top;
        patches[HoleValue_ZERO] = 0;
        emit(group, patches);
        code += group->code.body_size;
        data += group->data.body_size;
    }
    if (mark_executable(memory, code_size) ||
        mark_readable(memory + code_size, data_size))
    {
        jit_free(memory, code_size + data_size);
        return -1;
    }
    executor->jit_code = memory;
    executor->jit_size = code_size + data_size;
    return 0;
}

void
_PyJIT_Free(_PyExecutorObject *executor)
{
    char *memory = (char *)executor->jit_code;
    size_t size = executor->jit_size;
    if (memory) {
        executor->jit_code = NULL;
        executor->jit_size = 0;
        if (jit_free(memory, size)) {
            PyErr_WriteUnraisable(NULL);
        }
    }
}

#endif  // _Py_JIT
generated by cgit v0.12 at 2025-05-22 19:17:26 (GMT)