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 | |
Antoine Pitrou | 3cbb24d | 2010-12-15 15:47:34 +0000 | [diff] [blame^] | 4 | IMPORTANT: if you want to build a debug-enabled Python, it is recommended |
| 5 | that you use ``./configure --with-pydebug``, rather than the options listed |
| 6 | here. |
| 7 | |
| 8 | However, if you wish to define some of these options individually, it is best |
| 9 | to define them in the EXTRA_CFLAGS make variable; |
Thomas Wouters | 49fd7fa | 2006-04-21 10:40:58 +0000 | [diff] [blame] | 10 | ``make EXTRA_CFLAGS="-DPy_REF_DEBUG"``. |
Brett Cannon | a267563 | 2005-04-19 20:28:09 +0000 | [diff] [blame] | 11 | |
Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 12 | --------------------------------------------------------------------------- |
Tim Peters | 62fc52e | 2002-07-11 00:23:58 +0000 | [diff] [blame] | 13 | Py_REF_DEBUG introduced in 1.4 |
| 14 | named REF_DEBUG before 1.4 |
Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 15 | |
| 16 | Turn on aggregate reference counting. This arranges that extern |
| 17 | _Py_RefTotal hold a count of all references, the sum of ob_refcnt across |
| 18 | all objects. In a debug-mode build, this is where the "8288" comes from |
| 19 | in |
| 20 | |
| 21 | >>> 23 |
| 22 | 23 |
| 23 | [8288 refs] |
| 24 | >>> |
| 25 | |
| 26 | Note that if this count increases when you're not storing away new objects, |
| 27 | there's probably a leak. Remember, though, that in interactive mode the |
| 28 | special name "_" holds a reference to the last result displayed! |
| 29 | |
| 30 | Py_REF_DEBUG also checks after every decref to verify that the refcount |
| 31 | hasn't gone negative, and causes an immediate fatal error if it has. |
| 32 | |
| 33 | Special gimmicks: |
| 34 | |
| 35 | sys.gettotalrefcount() |
| 36 | Return current total of all refcounts. |
| 37 | Available under Py_REF_DEBUG in Python 2.3. |
| 38 | Before 2.3, Py_TRACE_REFS was required to enable this function. |
| 39 | --------------------------------------------------------------------------- |
Tim Peters | 62fc52e | 2002-07-11 00:23:58 +0000 | [diff] [blame] | 40 | Py_TRACE_REFS introduced in 1.4 |
| 41 | named TRACE_REFS before 1.4 |
Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 42 | |
| 43 | Turn on heavy reference debugging. This is major surgery. Every PyObject |
| 44 | grows two more pointers, to maintain a doubly-linked list of all live |
Georg Brandl | 93dc9eb | 2010-03-14 10:56:14 +0000 | [diff] [blame] | 45 | heap-allocated objects. Most built-in type objects are not in this list, |
Tim Peters | 78be799 | 2003-03-23 02:51:01 +0000 | [diff] [blame] | 46 | as they're statically allocated. Starting in Python 2.3, if COUNT_ALLOCS |
| 47 | (see below) is also defined, a static type object T does appear in this |
| 48 | list if at least one object of type T has been created. |
| 49 | |
| 50 | Note that because the fundamental PyObject layout changes, Python modules |
| 51 | compiled with Py_TRACE_REFS are incompatible with modules compiled without |
| 52 | it. |
Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 53 | |
| 54 | Py_TRACE_REFS implies Py_REF_DEBUG. |
| 55 | |
| 56 | Special gimmicks: |
| 57 | |
| 58 | sys.getobjects(max[, type]) |
Tim Peters | a788f5e | 2002-07-10 18:47:03 +0000 | [diff] [blame] | 59 | Return list of the (no more than) max most-recently allocated objects, |
| 60 | most recently allocated first in the list, least-recently allocated |
| 61 | last in the list. max=0 means no limit on list length. |
| 62 | If an optional type object is passed, the list is also restricted to |
| 63 | objects of that type. |
| 64 | The return list itself, and some temp objects created just to call |
| 65 | sys.getobjects(), are excluded from the return list. Note that the |
| 66 | list returned is just another object, though, so may appear in the |
| 67 | return list the next time you call getobjects(); note that every |
| 68 | object in the list is kept alive too, simply by virtue of being in |
| 69 | the list. |
Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 70 | |
| 71 | envar PYTHONDUMPREFS |
| 72 | If this envar exists, Py_Finalize() arranges to print a list of |
Tim Peters | 21d7d4d | 2003-04-18 00:45:59 +0000 | [diff] [blame] | 73 | all still-live heap objects. This is printed twice, in different |
| 74 | formats, before and after Py_Finalize has cleaned up everything it |
| 75 | can clean up. The first output block produces the repr() of each |
| 76 | object so is more informative; however, a lot of stuff destined to |
| 77 | die is still alive then. The second output block is much harder |
| 78 | to work with (repr() can't be invoked anymore -- the interpreter |
| 79 | has been torn down too far), but doesn't list any objects that will |
| 80 | die. The tool script combinerefs.py can be run over this to combine |
| 81 | the info from both output blocks. The second output block, and |
| 82 | combinerefs.py, were new in Python 2.3b1. |
Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 83 | --------------------------------------------------------------------------- |
Tim Peters | 62fc52e | 2002-07-11 00:23:58 +0000 | [diff] [blame] | 84 | PYMALLOC_DEBUG introduced in 2.3 |
Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 85 | |
Tim Peters | 889f61d | 2002-07-10 19:29:49 +0000 | [diff] [blame] | 86 | When pymalloc is enabled (WITH_PYMALLOC is defined), calls to the PyObject_ |
| 87 | memory routines are handled by Python's own small-object allocator, while |
| 88 | calls to the PyMem_ memory routines are directed to the system malloc/ |
| 89 | realloc/free. If PYMALLOC_DEBUG is also defined, calls to both PyObject_ |
| 90 | and PyMem_ memory routines are directed to a special debugging mode of |
| 91 | Python's small-object allocator. |
| 92 | |
| 93 | This mode fills dynamically allocated memory blocks with special, |
| 94 | recognizable bit patterns, and adds debugging info on each end of |
| 95 | dynamically allocated memory blocks. The special bit patterns are: |
| 96 | |
| 97 | #define CLEANBYTE 0xCB /* clean (newly allocated) memory */ |
| 98 | #define DEADBYTE 0xDB /* dead (newly freed) memory */ |
Thomas Wouters | 89f507f | 2006-12-13 04:49:30 +0000 | [diff] [blame] | 99 | #define FORBIDDENBYTE 0xFB /* forbidden -- untouchable bytes */ |
Tim Peters | 889f61d | 2002-07-10 19:29:49 +0000 | [diff] [blame] | 100 | |
| 101 | Strings of these bytes are unlikely to be valid addresses, floats, or 7-bit |
| 102 | ASCII strings. |
| 103 | |
Thomas Wouters | 73e5a5b | 2006-06-08 15:35:45 +0000 | [diff] [blame] | 104 | Let S = sizeof(size_t). 2*S bytes are added at each end of each block of N |
| 105 | bytes requested. The memory layout is like so, where p represents the |
| 106 | address returned by a malloc-like or realloc-like function (p[i:j] means |
| 107 | the slice of bytes from *(p+i) inclusive up to *(p+j) exclusive; note that |
| 108 | the treatment of negative indices differs from a Python slice): |
Tim Peters | 889f61d | 2002-07-10 19:29:49 +0000 | [diff] [blame] | 109 | |
Thomas Wouters | 73e5a5b | 2006-06-08 15:35:45 +0000 | [diff] [blame] | 110 | p[-2*S:-S] |
| 111 | Number of bytes originally asked for. This is a size_t, big-endian |
| 112 | (easier to read in a memory dump). |
| 113 | p[-S:0] |
Tim Peters | 889f61d | 2002-07-10 19:29:49 +0000 | [diff] [blame] | 114 | Copies of FORBIDDENBYTE. Used to catch under- writes and reads. |
| 115 | p[0:N] |
Tim Peters | 62fc52e | 2002-07-11 00:23:58 +0000 | [diff] [blame] | 116 | The requested memory, filled with copies of CLEANBYTE, used to catch |
| 117 | reference to uninitialized memory. |
Tim Peters | 889f61d | 2002-07-10 19:29:49 +0000 | [diff] [blame] | 118 | When a realloc-like function is called requesting a larger memory |
| 119 | block, the new excess bytes are also filled with CLEANBYTE. |
| 120 | When a free-like function is called, these are overwritten with |
Tim Peters | 62fc52e | 2002-07-11 00:23:58 +0000 | [diff] [blame] | 121 | DEADBYTE, to catch reference to freed memory. When a realloc- |
Tim Peters | 889f61d | 2002-07-10 19:29:49 +0000 | [diff] [blame] | 122 | like function is called requesting a smaller memory block, the excess |
| 123 | old bytes are also filled with DEADBYTE. |
Thomas Wouters | 73e5a5b | 2006-06-08 15:35:45 +0000 | [diff] [blame] | 124 | p[N:N+S] |
Tim Peters | 889f61d | 2002-07-10 19:29:49 +0000 | [diff] [blame] | 125 | Copies of FORBIDDENBYTE. Used to catch over- writes and reads. |
Thomas Wouters | 73e5a5b | 2006-06-08 15:35:45 +0000 | [diff] [blame] | 126 | p[N+S:N+2*S] |
Tim Peters | 889f61d | 2002-07-10 19:29:49 +0000 | [diff] [blame] | 127 | A serial number, incremented by 1 on each call to a malloc-like or |
| 128 | realloc-like function. |
Thomas Wouters | 73e5a5b | 2006-06-08 15:35:45 +0000 | [diff] [blame] | 129 | Big-endian size_t. |
Tim Peters | 889f61d | 2002-07-10 19:29:49 +0000 | [diff] [blame] | 130 | If "bad memory" is detected later, the serial number gives an |
| 131 | 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] | 132 | instant at which this block was passed out. The static function |
| 133 | bumpserialno() in obmalloc.c is the only place the serial number |
| 134 | is incremented, and exists so you can set such a breakpoint easily. |
Tim Peters | 889f61d | 2002-07-10 19:29:49 +0000 | [diff] [blame] | 135 | |
Tim Peters | 62fc52e | 2002-07-11 00:23:58 +0000 | [diff] [blame] | 136 | A realloc-like or free-like function first checks that the FORBIDDENBYTEs |
Tim Peters | 889f61d | 2002-07-10 19:29:49 +0000 | [diff] [blame] | 137 | at each end are intact. If they've been altered, diagnostic output is |
Tim Peters | 62fc52e | 2002-07-11 00:23:58 +0000 | [diff] [blame] | 138 | written to stderr, and the program is aborted via Py_FatalError(). The |
| 139 | other main failure mode is provoking a memory error when a program |
| 140 | reads up one of the special bit patterns and tries to use it as an address. |
| 141 | If you get in a debugger then and look at the object, you're likely |
| 142 | to see that it's entirely filled with 0xDB (meaning freed memory is |
| 143 | getting used) or 0xCB (meaning uninitialized memory is getting used). |
Tim Peters | 889f61d | 2002-07-10 19:29:49 +0000 | [diff] [blame] | 144 | |
| 145 | Note that PYMALLOC_DEBUG requires WITH_PYMALLOC. |
| 146 | |
Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 147 | Special gimmicks: |
| 148 | |
| 149 | envar PYTHONMALLOCSTATS |
| 150 | If this envar exists, a report of pymalloc summary statistics is |
| 151 | printed to stderr whenever a new arena is allocated, and also |
| 152 | by Py_Finalize(). |
Thomas Wouters | 73e5a5b | 2006-06-08 15:35:45 +0000 | [diff] [blame] | 153 | |
| 154 | Changed in 2.5: The number of extra bytes allocated is 4*sizeof(size_t). |
| 155 | Before it was 16 on all boxes, reflecting that Python couldn't make use of |
| 156 | allocations >= 2**32 bytes even on 64-bit boxes before 2.5. |
Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 157 | --------------------------------------------------------------------------- |
Tim Peters | 62fc52e | 2002-07-11 00:23:58 +0000 | [diff] [blame] | 158 | Py_DEBUG introduced in 1.5 |
| 159 | named DEBUG before 1.5 |
Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 160 | |
| 161 | This is what is generally meant by "a debug build" of Python. |
| 162 | |
Michael W. Hudson | a625523 | 2002-07-30 09:49:29 +0000 | [diff] [blame] | 163 | Py_DEBUG implies LLTRACE, Py_REF_DEBUG, Py_TRACE_REFS, and |
| 164 | PYMALLOC_DEBUG (if WITH_PYMALLOC is enabled). In addition, C |
| 165 | assert()s are enabled (via the C way: by not defining NDEBUG), and |
| 166 | some routines do additional sanity checks inside "#ifdef Py_DEBUG" |
| 167 | blocks. |
Tim Peters | 6045d48 | 2002-07-09 18:35:34 +0000 | [diff] [blame] | 168 | --------------------------------------------------------------------------- |
Michael W. Hudson | 202a4b6 | 2002-07-30 15:25:57 +0000 | [diff] [blame] | 169 | COUNT_ALLOCS introduced in 0.9.9 |
| 170 | partly broken in 2.2 and 2.2.1 |
Tim Peters | 48ba649 | 2002-07-09 19:24:54 +0000 | [diff] [blame] | 171 | |
| 172 | Each type object grows three new members: |
| 173 | |
| 174 | /* Number of times an object of this type was allocated. */ |
Guido van Rossum | 0c08864 | 2002-07-11 01:04:32 +0000 | [diff] [blame] | 175 | int tp_allocs; |
Tim Peters | 48ba649 | 2002-07-09 19:24:54 +0000 | [diff] [blame] | 176 | |
| 177 | /* Number of times an object of this type was deallocated. */ |
Guido van Rossum | 0c08864 | 2002-07-11 01:04:32 +0000 | [diff] [blame] | 178 | int tp_frees; |
Tim Peters | 48ba649 | 2002-07-09 19:24:54 +0000 | [diff] [blame] | 179 | |
Guido van Rossum | 0c08864 | 2002-07-11 01:04:32 +0000 | [diff] [blame] | 180 | /* Highwater mark: the maximum value of tp_allocs - tp_frees so |
| 181 | * far; or, IOW, the largest number of objects of this type alive at |
| 182 | * the same time. |
| 183 | */ |
| 184 | int tp_maxalloc; |
Tim Peters | 48ba649 | 2002-07-09 19:24:54 +0000 | [diff] [blame] | 185 | |
| 186 | Allocation and deallocation code keeps these counts up to date. |
| 187 | Py_Finalize() displays a summary of the info returned by sys.getcounts() |
| 188 | (see below), along with assorted other special allocation counts (like |
| 189 | the number of tuple allocations satisfied by a tuple free-list, the number |
| 190 | of 1-character strings allocated, etc). |
| 191 | |
| 192 | Before Python 2.2, type objects were immortal, and the COUNT_ALLOCS |
| 193 | implementation relies on that. As of Python 2.2, heap-allocated type/ |
| 194 | class objects can go away. COUNT_ALLOCS can blow up in 2.2 and 2.2.1 |
| 195 | because of this; this was fixed in 2.2.2. Use of COUNT_ALLOCS makes |
| 196 | all heap-allocated type objects immortal, except for those for which no |
| 197 | object of that type is ever allocated. |
| 198 | |
Tim Peters | 78be799 | 2003-03-23 02:51:01 +0000 | [diff] [blame] | 199 | Starting with Python 2.3, If Py_TRACE_REFS is also defined, COUNT_ALLOCS |
| 200 | arranges to ensure that the type object for each allocated object |
| 201 | appears in the doubly-linked list of all objects maintained by |
| 202 | Py_TRACE_REFS. |
| 203 | |
Tim Peters | 48ba649 | 2002-07-09 19:24:54 +0000 | [diff] [blame] | 204 | Special gimmicks: |
| 205 | |
| 206 | sys.getcounts() |
| 207 | Return a list of 4-tuples, one entry for each type object for which |
| 208 | at least one object of that type was allocated. Each tuple is of |
| 209 | the form: |
| 210 | |
| 211 | (tp_name, tp_allocs, tp_frees, tp_maxalloc) |
| 212 | |
Tim Peters | 44c1a7b | 2002-07-09 19:27:20 +0000 | [diff] [blame] | 213 | Each distinct type object gets a distinct entry in this list, even |
Tim Peters | 48ba649 | 2002-07-09 19:24:54 +0000 | [diff] [blame] | 214 | if two or more type objects have the same tp_name (in which case |
| 215 | there's no way to distinguish them by looking at this list). The |
| 216 | list is ordered by time of first object allocation: the type object |
| 217 | for which the first allocation of an object of that type occurred |
| 218 | most recently is at the front of the list. |
| 219 | --------------------------------------------------------------------------- |
Michael W. Hudson | 202a4b6 | 2002-07-30 15:25:57 +0000 | [diff] [blame] | 220 | LLTRACE introduced well before 1.0 |
Michael W. Hudson | a625523 | 2002-07-30 09:49:29 +0000 | [diff] [blame] | 221 | |
Michael W. Hudson | 46e6d92 | 2005-01-18 15:53:59 +0000 | [diff] [blame] | 222 | Compile in support for Low Level TRACE-ing of the main interpreter loop. |
Michael W. Hudson | a625523 | 2002-07-30 09:49:29 +0000 | [diff] [blame] | 223 | |
Michael W. Hudson | 46e6d92 | 2005-01-18 15:53:59 +0000 | [diff] [blame] | 224 | When this preprocessor symbol is defined, before PyEval_EvalFrame |
| 225 | (eval_frame in 2.3 and 2.2, eval_code2 before that) executes a frame's code |
| 226 | it checks the frame's global namespace for a variable "__lltrace__". If |
| 227 | such a variable is found, mounds of information about what the interpreter |
| 228 | is doing are sprayed to stdout, such as every opcode and opcode argument |
| 229 | and values pushed onto and popped off the value stack. |
Michael W. Hudson | a625523 | 2002-07-30 09:49:29 +0000 | [diff] [blame] | 230 | |
| 231 | Not useful very often, but very useful when needed. |
Jeremy Hylton | 985eba5 | 2003-02-05 23:13:00 +0000 | [diff] [blame] | 232 | |
| 233 | --------------------------------------------------------------------------- |
| 234 | CALL_PROFILE introduced for Python 2.3 |
| 235 | |
| 236 | Count the number of function calls executed. |
| 237 | |
| 238 | When this symbol is defined, the ceval mainloop and helper functions |
| 239 | count the number of function calls made. It keeps detailed statistics |
| 240 | about what kind of object was called and whether the call hit any of |
| 241 | the special fast paths in the code. |
Michael W. Hudson | 800ba23 | 2004-08-12 18:19:17 +0000 | [diff] [blame] | 242 | |
| 243 | --------------------------------------------------------------------------- |
| 244 | WITH_TSC introduced for Python 2.4 |
| 245 | |
| 246 | Super-lowlevel profiling of the interpreter. When enabled, the sys |
| 247 | module grows a new function: |
| 248 | |
| 249 | settscdump(bool) |
| 250 | If true, tell the Python interpreter to dump VM measurements to |
| 251 | stderr. If false, turn off dump. The measurements are based on the |
| 252 | processor's time-stamp counter. |
| 253 | |
| 254 | This build option requires a small amount of platform specific code. |
| 255 | Currently this code is present for linux/x86 and any PowerPC platform |
| 256 | that uses GCC (i.e. OS X and linux/ppc). |
| 257 | |
| 258 | On the PowerPC the rate at which the time base register is incremented |
| 259 | is not defined by the architecture specification, so you'll need to |
Michael W. Hudson | 46e6d92 | 2005-01-18 15:53:59 +0000 | [diff] [blame] | 260 | find the manual for your specific processor. For the 750CX, 750CXe |
| 261 | and 750FX (all sold as the G3) we find: |
Michael W. Hudson | 800ba23 | 2004-08-12 18:19:17 +0000 | [diff] [blame] | 262 | |
Thomas Wouters | 73e5a5b | 2006-06-08 15:35:45 +0000 | [diff] [blame] | 263 | The time base counter is clocked at a frequency that is |
Michael W. Hudson | 800ba23 | 2004-08-12 18:19:17 +0000 | [diff] [blame] | 264 | one-fourth that of the bus clock. |
| 265 | |
| 266 | This build is enabled by the --with-tsc flag to configure. |