Home | History | Annotate | Download | only in scripts 
      1 #! /usr/bin/env python
      2 
      3 """
      4 combinerefs path
      5 
      6 A helper for analyzing PYTHONDUMPREFS output.
      7 
      8 When the PYTHONDUMPREFS envar is set in a debug build, at Python shutdown
      9 time Py_Finalize() prints the list of all live objects twice:  first it
     10 prints the repr() of each object while the interpreter is still fully intact.
     11 After cleaning up everything it can, it prints all remaining live objects
     12 again, but the second time just prints their addresses, refcounts, and type
     13 names (because the interpreter has been torn down, calling repr methods at
     14 this point can get into infinite loops or blow up).
     15 
     16 Save all this output into a file, then run this script passing the path to
     17 that file.  The script finds both output chunks, combines them, then prints
     18 a line of output for each object still alive at the end:
     19 
     20     address refcnt typename repr
     21 
     22 address is the address of the object, in whatever format the platform C
     23 produces for a %p format code.
     24 
     25 refcnt is of the form
     26 
     27     "[" ref "]"
     28 
     29 when the object's refcount is the same in both PYTHONDUMPREFS output blocks,
     30 or
     31 
     32     "[" ref_before "->" ref_after "]"
     33 
     34 if the refcount changed.
     35 
     36 typename is object->ob_type->tp_name, extracted from the second PYTHONDUMPREFS
     37 output block.
     38 
     39 repr is repr(object), extracted from the first PYTHONDUMPREFS output block.
     40 CAUTION:  If object is a container type, it may not actually contain all the
     41 objects shown in the repr:  the repr was captured from the first output block,
     42 and some of the containees may have been released since then.  For example,
     43 it's common for the line showing the dict of interned strings to display
     44 strings that no longer exist at the end of Py_Finalize; this can be recognized
     45 (albeit painfully) because such containees don't have a line of their own.
     46 
     47 The objects are listed in allocation order, with most-recently allocated
     48 printed first, and the first object allocated printed last.
     49 
     50 
     51 Simple examples:
     52 
     53     00857060 [14] str '__len__'
     54 
     55 The str object '__len__' is alive at shutdown time, and both PYTHONDUMPREFS
     56 output blocks said there were 14 references to it.  This is probably due to
     57 C modules that intern the string "__len__" and keep a reference to it in a
     58 file static.
     59 
     60     00857038 [46->5] tuple ()
     61 
     62 46-5 = 41 references to the empty tuple were removed by the cleanup actions
     63 between the times PYTHONDUMPREFS produced output.
     64 
     65     00858028 [1025->1456] str '<dummy key>'
     66 
     67 The string '<dummy key>', which is used in dictobject.c to overwrite a real
     68 key that gets deleted, grew several hundred references during cleanup.  It
     69 suggests that stuff did get removed from dicts by cleanup, but that the dicts
     70 themselves are staying alive for some reason. """
     71 
     72 import re
     73 import sys
     74 
     75 # Generate lines from fileiter.  If whilematch is true, continue reading
     76 # while the regexp object pat matches line.  If whilematch is false, lines
     77 # are read so long as pat doesn't match them.  In any case, the first line
     78 # that doesn't match pat (when whilematch is true), or that does match pat
     79 # (when whilematch is false), is lost, and fileiter will resume at the line
     80 # following it.
     81 def read(fileiter, pat, whilematch):
     82     for line in fileiter:
     83         if bool(pat.match(line)) == whilematch:
     84             yield line
     85         else:
     86             break
     87 
     88 def combine(fname):
     89     f = file(fname)
     90     fi = iter(f)
     91 
     92     for line in read(fi, re.compile(r'^Remaining objects:$'), False):
     93         pass
     94 
     95     crack = re.compile(r'([a-zA-Z\d]+) \[(\d+)\] (.*)')
     96     addr2rc = {}
     97     addr2guts = {}
     98     before = 0
     99     for line in read(fi, re.compile(r'^Remaining object addresses:$'), False):
    100         m = crack.match(line)
    101         if m:
    102             addr, addr2rc[addr], addr2guts[addr] = m.groups()
    103             before += 1
    104         else:
    105             print '??? skipped:', line
    106 
    107     after = 0
    108     for line in read(fi, crack, True):
    109         after += 1
    110         m = crack.match(line)
    111         assert m
    112         addr, rc, guts = m.groups() # guts is type name here
    113         if addr not in addr2rc:
    114             print '??? new object created while tearing down:', line.rstrip()
    115             continue
    116         print addr,
    117         if rc == addr2rc[addr]:
    118             print '[%s]' % rc,
    119         else:
    120             print '[%s->%s]' % (addr2rc[addr], rc),
    121         print guts, addr2guts[addr]
    122 
    123     f.close()
    124     print "%d objects before, %d after" % (before, after)
    125 
    126 if __name__ == '__main__':
    127     combine(sys.argv[1])
    128