https://github.com/python/cpython/commit/f300a1fa4c121f7807cbda4fc8bb26240c6... commit: f300a1fa4c121f7807cbda4fc8bb26240c69ea74 branch: main author: Eric Snow <ericsnowcurrently@gmail.com> committer: ericsnowcurrently <ericsnowcurrently@gmail.com> date: 2023-02-28T13:14:40-07:00 summary: gh-100227: Move the dtoa State to PyInterpreterState (gh-102331) https://github.com/python/cpython/issues/100227 files: M Include/internal/pycore_dtoa.h M Include/internal/pycore_interp.h M Include/internal/pycore_runtime.h M Include/internal/pycore_runtime_init.h M Python/dtoa.c M Python/pystate.c diff --git a/Include/internal/pycore_dtoa.h b/Include/internal/pycore_dtoa.h index 67189cf0ade6..fb524770efed 100644 --- a/Include/internal/pycore_dtoa.h +++ b/Include/internal/pycore_dtoa.h @@ -24,10 +24,11 @@ Bigint { #ifdef Py_USING_MEMORY_DEBUGGER -struct _dtoa_runtime_state { +struct _dtoa_state { int _not_used; }; -#define _dtoa_runtime_state_INIT {0} +#define _dtoa_interp_state_INIT(INTERP) \ + {0} #else // !Py_USING_MEMORY_DEBUGGER @@ -40,7 +41,7 @@ struct _dtoa_runtime_state { #define Bigint_PREALLOC_SIZE \ ((PRIVATE_MEM+sizeof(double)-1)/sizeof(double)) -struct _dtoa_runtime_state { +struct _dtoa_state { /* p5s is a linked list of powers of 5 of the form 5**(2**i), i >= 2 */ // XXX This should be freed during runtime fini. struct Bigint *p5s; @@ -48,9 +49,9 @@ struct _dtoa_runtime_state { double preallocated[Bigint_PREALLOC_SIZE]; double *preallocated_next; }; -#define _dtoa_runtime_state_INIT(runtime) \ +#define _dtoa_state_INIT(INTERP) \ { \ - .preallocated_next = runtime.dtoa.preallocated, \ + .preallocated_next = (INTERP)->dtoa.preallocated, \ } #endif // !Py_USING_MEMORY_DEBUGGER diff --git a/Include/internal/pycore_interp.h b/Include/internal/pycore_interp.h index 60de31b336f6..7ef9c40153e4 100644 --- a/Include/internal/pycore_interp.h +++ b/Include/internal/pycore_interp.h @@ -16,6 +16,7 @@ extern "C" { #include "pycore_code.h" // struct callable_cache #include "pycore_context.h" // struct _Py_context_state #include "pycore_dict_state.h" // struct _Py_dict_state +#include "pycore_dtoa.h" // struct _dtoa_state #include "pycore_exceptions.h" // struct _Py_exc_state #include "pycore_floatobject.h" // struct _Py_float_state #include "pycore_function.h" // FUNC_MAX_WATCHERS @@ -139,6 +140,7 @@ struct _is { struct _Py_unicode_state unicode; struct _Py_float_state float_state; struct _Py_long_state long_state; + struct _dtoa_state dtoa; /* Using a cache is very effective since typically only a single slice is created and then deleted again. */ PySliceObject *slice_cache; diff --git a/Include/internal/pycore_runtime.h b/Include/internal/pycore_runtime.h index 9ef270791576..2350eaab5976 100644 --- a/Include/internal/pycore_runtime.h +++ b/Include/internal/pycore_runtime.h @@ -11,7 +11,6 @@ extern "C" { #include "pycore_atomic.h" /* _Py_atomic_address */ #include "pycore_ceval_state.h" // struct _ceval_runtime_state #include "pycore_dict_state.h" // struct _Py_dict_runtime_state -#include "pycore_dtoa.h" // struct _dtoa_runtime_state #include "pycore_floatobject.h" // struct _Py_float_runtime_state #include "pycore_faulthandler.h" // struct _faulthandler_runtime_state #include "pycore_function.h" // struct _func_runtime_state @@ -141,7 +140,6 @@ typedef struct pyruntimestate { struct _ceval_runtime_state ceval; struct _gilstate_runtime_state gilstate; struct _getargs_runtime_state getargs; - struct _dtoa_runtime_state dtoa; struct _fileutils_state fileutils; struct _faulthandler_runtime_state faulthandler; struct _tracemalloc_runtime_state tracemalloc; diff --git a/Include/internal/pycore_runtime_init.h b/Include/internal/pycore_runtime_init.h index a8d5953ff98b..b54adf04761d 100644 --- a/Include/internal/pycore_runtime_init.h +++ b/Include/internal/pycore_runtime_init.h @@ -53,7 +53,6 @@ extern "C" { .gilstate = { \ .check_enabled = 1, \ }, \ - .dtoa = _dtoa_runtime_state_INIT(runtime), \ .fileutils = { \ .force_ascii = -1, \ }, \ @@ -94,10 +93,10 @@ extern "C" { }, \ }, \ }, \ - ._main_interpreter = _PyInterpreterState_INIT, \ + ._main_interpreter = _PyInterpreterState_INIT(runtime._main_interpreter), \ } -#define _PyInterpreterState_INIT \ +#define _PyInterpreterState_INIT(INTERP) \ { \ .id_refcount = -1, \ .imports = IMPORTS_INIT, \ @@ -113,6 +112,7 @@ extern "C" { { .threshold = 10, }, \ }, \ }, \ + .dtoa = _dtoa_state_INIT(&(INTERP)), \ .static_objects = { \ .singletons = { \ ._not_used = 1, \ diff --git a/Python/dtoa.c b/Python/dtoa.c index cff5f1b0658e..6ea60ac9946e 100644 --- a/Python/dtoa.c +++ b/Python/dtoa.c @@ -119,7 +119,7 @@ #include "Python.h" #include "pycore_dtoa.h" // _PY_SHORT_FLOAT_REPR -#include "pycore_runtime.h" // _PyRuntime +#include "pycore_pystate.h" // _PyInterpreterState_GET() #include <stdlib.h> // exit() /* if _PY_SHORT_FLOAT_REPR == 0, then don't even try to compile @@ -339,9 +339,9 @@ typedef struct Bigint Bigint; Bfree to PyMem_Free. Investigate whether this has any significant performance on impact. */ -#define freelist _PyRuntime.dtoa.freelist -#define private_mem _PyRuntime.dtoa.preallocated -#define pmem_next _PyRuntime.dtoa.preallocated_next +#define freelist interp->dtoa.freelist +#define private_mem interp->dtoa.preallocated +#define pmem_next interp->dtoa.preallocated_next /* Allocate space for a Bigint with up to 1<<k digits */ @@ -351,6 +351,7 @@ Balloc(int k) int x; Bigint *rv; unsigned int len; + PyInterpreterState *interp = _PyInterpreterState_GET(); if (k <= Bigint_Kmax && (rv = freelist[k])) freelist[k] = rv->next; @@ -385,6 +386,7 @@ Bfree(Bigint *v) if (v->k > Bigint_Kmax) FREE((void*)v); else { + PyInterpreterState *interp = _PyInterpreterState_GET(); v->next = freelist[v->k]; freelist[v->k] = v; } @@ -692,7 +694,8 @@ pow5mult(Bigint *b, int k) if (!(k >>= 2)) return b; - p5 = _PyRuntime.dtoa.p5s; + PyInterpreterState *interp = _PyInterpreterState_GET(); + p5 = interp->dtoa.p5s; if (!p5) { /* first time */ p5 = i2b(625); @@ -700,7 +703,7 @@ pow5mult(Bigint *b, int k) Bfree(b); return NULL; } - _PyRuntime.dtoa.p5s = p5; + interp->dtoa.p5s = p5; p5->next = 0; } for(;;) { diff --git a/Python/pystate.c b/Python/pystate.c index 3c655bf38958..28606e4f32f7 100644 --- a/Python/pystate.c +++ b/Python/pystate.c @@ -3,7 +3,8 @@ #include "Python.h" #include "pycore_ceval.h" -#include "pycore_code.h" // stats +#include "pycore_code.h" // stats +#include "pycore_dtoa.h" // _dtoa_state_INIT() #include "pycore_frame.h" #include "pycore_initconfig.h" #include "pycore_object.h" // _PyType_InitCache() @@ -618,6 +619,18 @@ free_interpreter(PyInterpreterState *interp) e.g. by PyMem_RawCalloc() or memset(), or otherwise pre-initialized. The runtime state is not manipulated. Instead it is assumed that the interpreter is getting added to the runtime. + + Note that the main interpreter was statically initialized as part + of the runtime and most state is already set properly. That leaves + a small number of fields to initialize dynamically, as well as some + that are initialized lazily. + + For subinterpreters we memcpy() the main interpreter in + PyInterpreterState_New(), leaving it in the same mostly-initialized + state. The only difference is that the interpreter has some + self-referential state that is statically initializexd to the + main interpreter. We fix those fields here, in addition + to the other dynamically initialized fields. */ static void @@ -645,6 +658,11 @@ init_interpreter(PyInterpreterState *interp, PyConfig_InitPythonConfig(&interp->config); _PyType_InitCache(interp); + if (interp != &runtime->_main_interpreter) { + /* Fix the self-referential, statically initialized fields. */ + interp->dtoa = (struct _dtoa_state)_dtoa_state_INIT(interp); + } + interp->_initialized = 1; }