1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
|
# Copyright 2002-2017 Free Software Foundation, Inc.
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
if { [skip_cplus_tests] } { continue }
standard_testfile hang1.cc hang2.cc hang3.cc
if {[prepare_for_testing "failed to prepare" $testfile \
[list $srcfile $srcfile2 $srcfile3] {debug c++}]} {
return -1
}
# As of May 1, 2002, GDB hangs trying to read the debug info for the
# `hang2.o' compilation unit from the executable `hang', when compiled
# by g++ 2.96 with STABS debugging info. Here's what's going on, as
# best as I can tell.
#
# The definition of `struct A' in `hang.H' refers to `struct B' as an
# incomplete type. The stabs declare type number (1,3) to be a cross-
# reference type, `xsB:'.
#
# The definition of `struct C' contains a nested definition for
# `struct B' --- or more properly, `struct C::B'. However, the stabs
# fail to qualify the structure tag: it just looks like a definition
# for `struct B'. I think this is a compiler bug, but perhaps GCC
# doesn't emit qualified names for a reason.
#
# `hang.H' gets #included by both `hang1.C' and `hang2.C'. So the
# stabs for `struct A', the incomplete `struct B', and `struct C'
# appear in both hang1.o's and hang2.o's stabs.
#
# When those two files are linked together, since hang2.o appears
# later in the command line, its #inclusion of `hang.H' gets replaced
# with an N_EXCL stab, referring back to hang1.o's stabs for the
# header file.
#
# When GDB builds psymtabs for the executable hang, it notes that
# hang2.o's stabs contain an N_EXCL referring to a header that appears
# in full in hang1.o's stabs. So hang2.o's psymtab lists a dependency
# on hang1.o's psymtab.
#
# When the user types the command `print var_in_b', GDB scans the
# psymtabs for a symbol by that name, and decides to read full symbols
# for `hang2.o'.
#
# Since `hang2.o''s psymtab lists `hang1.o' as a dependency, GDB first
# reads `hang1.o''s symbols. When GDB sees `(1,3)=xsB:', it creates a
# type object for `struct B', sets its TYPE_STUB flag, and records it
# as type number `(1,3)'.
#
# When GDB finds the definition of `struct C::B', since the stabs
# don't indicate that the type is nested within C, it treats it as
# a definition of `struct B'.
#
# When GDB is finished reading `hang1.o''s symbols, it calls
# `cleanup_undefined_types'. This function mistakes the definition of
# `struct C::B' for a definition for `struct B', and overwrites the
# incomplete type object for the real `struct B', using `memcpy'. Now
# stabs type number `(1,3)' refers to this (incorrect) complete type.
# Furthermore, the `memcpy' simply copies the original's `cv_type'
# field to the target, giving the target a corrupt `cv_type' ring: the
# chain does not point back to the target type.
#
# Having satisfied `hang2.o''s psymtab's dependencies, GDB begins to
# read `hang2.o''s symbols. These contain the true definition for
# `struct B', which refers to type number `(1,3)' as the type it's
# defining. GDB looks up type `(1,3)', and finds the (incorrect)
# complete type established by the call to `cleanup_undefined_types'
# above. However, it doesn't notice that the type is already defined,
# and passes it to `read_struct_type', which then writes the new
# definition's size, field list, etc. into the type object which
# already has those fields initialized. Adding insult to injury,
# `read_struct_type' then calls `finish_cv_type'; since the `memcpy'
# in `cleanup_undefined_types' corrupted the target type's `cv_type'
# ring, `finish_cv_type' enters an infinite loop.
# This checks that GDB recognizes when a structure is about to be
# overwritten, and refuses, with a complaint.
gdb_test "print var_in_b" " = 1729" "doesn't overwrite struct type"
# This checks that cleanup_undefined_types doesn't create corrupt
# cv_type chains. Note that var_in_hang3 does need to be declared in
# a separate compilation unit, whose psymtab depends on hang1.o's
# psymtab. Otherwise, GDB won't call cleanup_undefined_types (as it
# finishes hang1.o's symbols) before it calls make_cv_type (while
# reading hang3.o's symbols).
#
# The bug only happens when you compile with -gstabs+; Otherwise, GCC
# won't include the `const' qualifier on `const_B_ptr' in `hang3.o''s
# STABS, so GDB won't try to create a const variant of the smashed
# struct type, and get caught by the corrupted cv_type chain.
gdb_test "print var_in_hang3" " = 42" "doesn't corrupt cv_type chain"
|
GitWeb