diff options
| -rw-r--r-- | Misc/NEWS | 3 | ||||
| -rw-r--r-- | Misc/SpecialBuilds.txt | 11 | ||||
| -rw-r--r-- | Objects/object.c | 3 | ||||
| -rw-r--r-- | Tools/scripts/combinerefs.py | 124 |
4 files changed, 139 insertions, 2 deletions
@@ -162,6 +162,9 @@ Library Tools/Demos ----------- +- New script combinerefs.py helps analyze new PYTHONDUMPREFS output. + See the module docstring for details. + TBD Build diff --git a/Misc/SpecialBuilds.txt b/Misc/SpecialBuilds.txt index a3d3558..fc41767 100644 --- a/Misc/SpecialBuilds.txt +++ b/Misc/SpecialBuilds.txt @@ -62,7 +62,16 @@ sys.getobjects(max[, type]) envar PYTHONDUMPREFS If this envar exists, Py_Finalize() arranges to print a list of - all still-live heap objects. + all still-live heap objects. This is printed twice, in different + formats, before and after Py_Finalize has cleaned up everything it + can clean up. The first output block produces the repr() of each + object so is more informative; however, a lot of stuff destined to + die is still alive then. The second output block is much harder + to work with (repr() can't be invoked anymore -- the interpreter + has been torn down too far), but doesn't list any objects that will + die. The tool script combinerefs.py can be run over this to combine + the info from both output blocks. The second output block, and + combinerefs.py, were new in Python 2.3b1. --------------------------------------------------------------------------- PYMALLOC_DEBUG introduced in 2.3 diff --git a/Objects/object.c b/Objects/object.c index 49b9839..93057c0 100644 --- a/Objects/object.c +++ b/Objects/object.c @@ -2047,7 +2047,8 @@ _Py_PrintReferenceAddresses(FILE *fp) PyObject *op; fprintf(fp, "Remaining object addresses:\n"); for (op = refchain._ob_next; op != &refchain; op = op->_ob_next) - fprintf(fp, "%p [%d]\n", op, op->ob_refcnt); + fprintf(fp, "%p [%d] %s\n", op, op->ob_refcnt, + op->ob_type->tp_name); } PyObject * diff --git a/Tools/scripts/combinerefs.py b/Tools/scripts/combinerefs.py new file mode 100644 index 0000000..3fafd9b --- /dev/null +++ b/Tools/scripts/combinerefs.py @@ -0,0 +1,124 @@ +#! /usr/bin/env python + +""" +combinerefs path + +A helper for analyzing PYTHONDUMPREFS output. + +When the PYTHONDUMPREFS envar is set in a debug build, at Python shutdown +time Py_Finalize() prints the list of all live objects twice: first it +prints the repr() of each object while the interpreter is still fully intact. +After cleaning up everything it can, it prints all remaining live objects +again, but the second time just prints their addresses, refcounts, and type +names. + +Save all this output into a file, then run this script passing the path to +that file. The script finds both output chunks, combines them, then prints +a line of output for each object still alive at the end: + + address refcnt typename repr + +address is the address of the object, in whatever format the platform C +produces for a %p format code. + +refcnt is of the form + + "[" ref "]" + +when the object's refcount is the same in both PYTHONDUMPREFS output blocks, +or + + "[" ref_before "->" ref_after "]" + +if the refcount changed. + +typename is object->ob_type->tp_name, extracted from the second PYTHONDUMPREFS +output block. + +repr is repr(object), extracted from the first PYTHONDUMPREFS output block. + +The objects are listed in allocation order, with most-recently allocated +printed first, and the first object allocated printed last. + + +Simple examples: + + 00857060 [14] str '__len__' + +The str object '__len__' is alive at shutdown time, and both PYTHONDUMPREFS +output blocks said there were 14 references to it. This is probably due to +C modules that intern the string "__len__" and keep a reference to it in a +file static. + + 00857038 [46->5] tuple () + +46-5 = 41 references to the empty tuple were removed by the cleanup actions +between the times PYTHONDUMPREFS produced output. + + 00858028 [1025->1456] str '<dummy key>' + +The string '<dummy key>', which is used in dictobject.c as the name of the +dummy key that overwrites a real key that gets deleted, actually grew +several hundred references during cleanup. It suggests that stuff did get +removed from dicts by cleanup, but that the dicts themselves are staying +alive for some reason. +""" + +import re +import sys + +# Generate lines from fileiter. If whilematch is true, continue reading +# while the regexp object pat matches line. If whilematch is false, lines +# are read so long as pat doesn't match them. In any case, the first line +# that doesn't match pat (when whilematch is true), or that does match pat +# (when whilematch is false), is lost, and fileiter will resume at the line +# following it. +def read(fileiter, pat, whilematch): + result = [] + for line in fileiter: + if bool(pat.match(line)) == whilematch: + result.append(line) + else: + break + return result + +def combine(fname): + f = file(fname) + fi = iter(f) + + for line in read(fi, re.compile(r'^Remaining objects:$'), False): + pass + + crack = re.compile(r'([a-zA-Z\d]+) \[(\d+)\] (.*)') + addr2rc = {} + addr2guts = {} + before = 0 + for line in read(fi, re.compile(r'^Remaining object addresses:$'), False): + m = crack.match(line) + if m: + addr, addr2rc[addr], addr2guts[addr] = m.groups() + before += 1 + else: + print '??? skipped:', line + + after = 0 + for line in read(fi, crack, True): + after += 1 + m = crack.match(line) + assert m + addr, rc, guts = m.groups() # guts is type name here + if addr not in addr2rc: + print '??? new object created while tearing down:', line + continue + print addr, + if rc == addr2rc[addr]: + print '[%s]' % rc, + else: + print '[%s->%s]' % (addr2rc[addr], rc), + print guts, addr2guts[addr] + + f.close() + print "%d objects before, %d after" % (before, after) + +if __name__ == '__main__': + combine(sys.argv[1]) |