Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 1 | This file describes some special Python build types enabled via |
| 2 | compile-time preprocessor defines. |
| 3 | |
| 4 | --------------------------------------------------------------------------- |
| 5 | Py_REF_DEBUG |
| 6 | |
| 7 | Turn on aggregate reference counting. This arranges that extern |
| 8 | _Py_RefTotal hold a count of all references, the sum of ob_refcnt across |
| 9 | all objects. In a debug-mode build, this is where the "8288" comes from |
| 10 | in |
| 11 | |
| 12 | >>> 23 |
| 13 | 23 |
| 14 | [8288 refs] |
| 15 | >>> |
| 16 | |
| 17 | Note that if this count increases when you're not storing away new objects, |
| 18 | there's probably a leak. Remember, though, that in interactive mode the |
| 19 | special name "_" holds a reference to the last result displayed! |
| 20 | |
| 21 | Py_REF_DEBUG also checks after every decref to verify that the refcount |
| 22 | hasn't gone negative, and causes an immediate fatal error if it has. |
| 23 | |
| 24 | Special gimmicks: |
| 25 | |
| 26 | sys.gettotalrefcount() |
| 27 | Return current total of all refcounts. |
| 28 | Available under Py_REF_DEBUG in Python 2.3. |
| 29 | Before 2.3, Py_TRACE_REFS was required to enable this function. |
| 30 | --------------------------------------------------------------------------- |
| 31 | Py_TRACE_REFS |
| 32 | |
| 33 | Turn on heavy reference debugging. This is major surgery. Every PyObject |
| 34 | grows two more pointers, to maintain a doubly-linked list of all live |
| 35 | heap-allocated objects (note that, e.g., most builtin type objects are not |
| 36 | in this list, as they're statically allocated). Note that because the |
| 37 | fundamental PyObject layout changes, Python modules compiled with |
| 38 | Py_TRACE_REFS are incompatible with modules compiled without it. |
| 39 | |
| 40 | Py_TRACE_REFS implies Py_REF_DEBUG. |
| 41 | |
| 42 | Special gimmicks: |
| 43 | |
| 44 | sys.getobjects(max[, type]) |
Tim Peters | a788f5e | 2002-07-10 18:47:03 +0000 | [diff] [blame] | 45 | Return list of the (no more than) max most-recently allocated objects, |
| 46 | most recently allocated first in the list, least-recently allocated |
| 47 | last in the list. max=0 means no limit on list length. |
| 48 | If an optional type object is passed, the list is also restricted to |
| 49 | objects of that type. |
| 50 | The return list itself, and some temp objects created just to call |
| 51 | sys.getobjects(), are excluded from the return list. Note that the |
| 52 | list returned is just another object, though, so may appear in the |
| 53 | return list the next time you call getobjects(); note that every |
| 54 | object in the list is kept alive too, simply by virtue of being in |
| 55 | the list. |
Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 56 | |
| 57 | envar PYTHONDUMPREFS |
| 58 | If this envar exists, Py_Finalize() arranges to print a list of |
| 59 | all still-live heap objects. |
| 60 | --------------------------------------------------------------------------- |
Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 61 | PYMALLOC_DEBUG |
| 62 | |
Tim Peters | 889f61d | 2002-07-10 19:29:49 +0000 | [diff] [blame] | 63 | When pymalloc is enabled (WITH_PYMALLOC is defined), calls to the PyObject_ |
| 64 | memory routines are handled by Python's own small-object allocator, while |
| 65 | calls to the PyMem_ memory routines are directed to the system malloc/ |
| 66 | realloc/free. If PYMALLOC_DEBUG is also defined, calls to both PyObject_ |
| 67 | and PyMem_ memory routines are directed to a special debugging mode of |
| 68 | Python's small-object allocator. |
| 69 | |
| 70 | This mode fills dynamically allocated memory blocks with special, |
| 71 | recognizable bit patterns, and adds debugging info on each end of |
| 72 | dynamically allocated memory blocks. The special bit patterns are: |
| 73 | |
| 74 | #define CLEANBYTE 0xCB /* clean (newly allocated) memory */ |
| 75 | #define DEADBYTE 0xDB /* dead (newly freed) memory */ |
| 76 | #define FORBIDDENBYTE 0xFB /* fordidden -- untouchable bytes */ |
| 77 | |
| 78 | Strings of these bytes are unlikely to be valid addresses, floats, or 7-bit |
| 79 | ASCII strings. |
| 80 | |
| 81 | 8 bytes are added at each end of each block of N bytes requested. The |
| 82 | memory layout is like so, where p represents the address returned by a |
Tim Peters | 20c8a04 | 2002-07-11 00:02:52 +0000 | [diff] [blame^] | 83 | malloc-like or realloc-like function (p[i:j] means the slice of bytes |
| 84 | from *(p+i) inclusive up to *(p+j) exclusive; note that the treatment |
| 85 | of negative indices differs from a Python slice): |
Tim Peters | 889f61d | 2002-07-10 19:29:49 +0000 | [diff] [blame] | 86 | |
| 87 | p[-8:-4] |
| 88 | Number of bytes originally asked for. 4-byte unsigned integer, |
| 89 | big-endian (easier to read in a memory dump). |
| 90 | p[-4:0] |
| 91 | Copies of FORBIDDENBYTE. Used to catch under- writes and reads. |
| 92 | p[0:N] |
| 93 | The requested memory, filled with copies of CLEANBYTE. |
| 94 | Used to catch reference to uninitialized memory. |
| 95 | When a realloc-like function is called requesting a larger memory |
| 96 | block, the new excess bytes are also filled with CLEANBYTE. |
| 97 | When a free-like function is called, these are overwritten with |
| 98 | DEADBYTE, to catch reference to free()ed memory. When a realloc- |
| 99 | like function is called requesting a smaller memory block, the excess |
| 100 | old bytes are also filled with DEADBYTE. |
| 101 | p[N:N+4] |
| 102 | Copies of FORBIDDENBYTE. Used to catch over- writes and reads. |
| 103 | p[N+4:N+8] |
| 104 | A serial number, incremented by 1 on each call to a malloc-like or |
| 105 | realloc-like function. |
| 106 | 4-byte unsigned integer, big-endian. |
| 107 | If "bad memory" is detected later, the serial number gives an |
| 108 | excellent way to set a breakpoint on the next run, to capture the |
Tim Peters | 20c8a04 | 2002-07-11 00:02:52 +0000 | [diff] [blame^] | 109 | instant at which this block was passed out. The static function |
| 110 | bumpserialno() in obmalloc.c is the only place the serial number |
| 111 | is incremented, and exists so you can set such a breakpoint easily. |
Tim Peters | 889f61d | 2002-07-10 19:29:49 +0000 | [diff] [blame] | 112 | |
| 113 | A malloc-like or free-like function first checks that the FORBIDDENBYTEs |
| 114 | at each end are intact. If they've been altered, diagnostic output is |
| 115 | written to stderr, and the program is aborted by Py_FatalError(). |
| 116 | |
| 117 | Note that PYMALLOC_DEBUG requires WITH_PYMALLOC. |
| 118 | |
Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 119 | Special gimmicks: |
| 120 | |
| 121 | envar PYTHONMALLOCSTATS |
| 122 | If this envar exists, a report of pymalloc summary statistics is |
| 123 | printed to stderr whenever a new arena is allocated, and also |
| 124 | by Py_Finalize(). |
| 125 | --------------------------------------------------------------------------- |
| 126 | Py_DEBUG |
| 127 | |
| 128 | This is what is generally meant by "a debug build" of Python. |
| 129 | |
| 130 | Py_DEBUG implies Py_REF_DEBUG, Py_TRACE_REFS, and PYMALLOC_DEBUG (if |
Tim Peters | 20c8a04 | 2002-07-11 00:02:52 +0000 | [diff] [blame^] | 131 | WITH_PYMALLOC is enabled). In addition, C assert()s are enabled (via |
| 132 | the C way: by not defining NDEBUG), and some routines do additional |
| 133 | sanity checks inside "#ifdef Py_DEBUG" blocks. |
Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 134 | --------------------------------------------------------------------------- |
Tim Peters | 48ba649 | 2002-07-09 19:24:54 +0000 | [diff] [blame] | 135 | COUNT_ALLOCS |
| 136 | |
| 137 | Each type object grows three new members: |
| 138 | |
| 139 | /* Number of times an object of this type was allocated. */ |
| 140 | int tp_allocs; |
| 141 | |
| 142 | /* Number of times an object of this type was deallocated. */ |
| 143 | int tp_frees; |
| 144 | |
| 145 | /* Highwater mark: the maximum value of tp_allocs - tp_frees so |
| 146 | * far; or, IOW, the largest number of objects of this type alive at |
| 147 | * the same time. |
| 148 | */ |
| 149 | int tp_maxalloc; |
| 150 | |
| 151 | Allocation and deallocation code keeps these counts up to date. |
| 152 | Py_Finalize() displays a summary of the info returned by sys.getcounts() |
| 153 | (see below), along with assorted other special allocation counts (like |
| 154 | the number of tuple allocations satisfied by a tuple free-list, the number |
| 155 | of 1-character strings allocated, etc). |
| 156 | |
| 157 | Before Python 2.2, type objects were immortal, and the COUNT_ALLOCS |
| 158 | implementation relies on that. As of Python 2.2, heap-allocated type/ |
| 159 | class objects can go away. COUNT_ALLOCS can blow up in 2.2 and 2.2.1 |
| 160 | because of this; this was fixed in 2.2.2. Use of COUNT_ALLOCS makes |
| 161 | all heap-allocated type objects immortal, except for those for which no |
| 162 | object of that type is ever allocated. |
| 163 | |
| 164 | Special gimmicks: |
| 165 | |
| 166 | sys.getcounts() |
| 167 | Return a list of 4-tuples, one entry for each type object for which |
| 168 | at least one object of that type was allocated. Each tuple is of |
| 169 | the form: |
| 170 | |
| 171 | (tp_name, tp_allocs, tp_frees, tp_maxalloc) |
| 172 | |
Tim Peters | 44c1a7b | 2002-07-09 19:27:20 +0000 | [diff] [blame] | 173 | Each distinct type object gets a distinct entry in this list, even |
Tim Peters | 48ba649 | 2002-07-09 19:24:54 +0000 | [diff] [blame] | 174 | if two or more type objects have the same tp_name (in which case |
| 175 | there's no way to distinguish them by looking at this list). The |
| 176 | list is ordered by time of first object allocation: the type object |
| 177 | for which the first allocation of an object of that type occurred |
| 178 | most recently is at the front of the list. |
| 179 | --------------------------------------------------------------------------- |