Respected GDB community members,

Hi,

This is a patch requesting to modify the gdb.base/examine-backwards testcase so that AIX can also pass this test. Currently we fail by saying

Type "apropos word" to search for commands related to "word"...

Reading symbols from /home/aditya/gdb_tests/examine-backward...

(gdb) b main

Breakpoint 1 at 0x100004ac: file /home/aditya/gdb_tests/examine-backward.c, line 96.

(gdb) r

Starting program: /home/aditya/gdb_tests/examine-backward

Breakpoint 1, main () at /home/aditya/gdb_tests/examine-backward.c:96

96 int dummy = Barrier[0] + TestStrings[0] + TestStringsH[0] + TestStringsW[0];

(gdb) x/6s TestStrings

0xffffffff: <error: Cannot access memory at address 0xffffffff>

(gdb)

This is because linker has not assigned any address to the variable TestStrings and still treats it as unused.

In the dwarf dump in AIX for the gdb.base/examine-backward.c file’s binary we see:-

<1>< 252> DW_TAG_variable

DW_AT_name TestStrings

DW_AT_decl_file 1

DW_AT_decl_line 43

DW_AT_decl_column 21

DW_AT_type <247>

DW_AT_external yes

DW_AT_location DW_OP_addr 0xffffffff

<1>< 278> DW_TAG_array_type

DW_AT_type <312>

DW_AT_sibling <294>

<2>< 287> DW_TAG_subrange_type

DW_AT_type <166>

DW_AT_upper_bound 72

<1>< 294> DW_TAG_const_type

DW_AT_type <278>

<1>< 299> DW_TAG_base_type

DW_AT_byte_size 2

DW_AT_encoding DW_ATE_signed

DW_AT_name short int

<1>< 312> DW_TAG_const_type

DW_AT_type <299>

<1>< 317> DW_TAG_variable

DW_AT_name TestStringsH

DW_AT_decl_file 1

This is a problem. After we modify the test case as done by this patch say using {+ char *dummy_string = TestStrings;} we get,

(gdb) b main

Breakpoint 1 at 0x100004ac: file /home/aditya/gdb_tests/examine-backward.c, line 96.

(gdb) r

Starting program: /home/aditya/gdb_tests/examine-backward

Breakpoint 1, main () at /home/aditya/gdb_tests/examine-backward.c:96

96 int dummy = Barrier[0] + TestStrings[0] + TestStringsH[0] + TestStringsW[0];

(gdb) x/6s TestStrings

0x10000c74 <TestStrings>: "ABCDEFGHIJKLMNOPQRSTUVWXYZ"

0x10000c8f <TestStrings+27>: ""

0x10000c90 <TestStrings+28>: ""

0x10000c91 <TestStrings+29>: "�\201��\201\222�\201��\201\222"

0x10000c9e <TestStrings+42>: "012345678901234567890123456789"

0x10000cbd <TestStrings+73>: "!!!!!!"

(gdb)

Dwarf dump after modification of the test case:-
<1>< 242> DW_TAG_variable

DW_AT_name TestStrings

DW_AT_decl_file 1

DW_AT_decl_line 43

DW_AT_decl_column 21

DW_AT_type <237>

DW_AT_external yes

DW_AT_location DW_OP_addr 0x10000c74

Test case numbers after patch applied:-

=== gdb Summary ===

# of expected passes 70

# of unsupported tests 1

Before we applied this patch, we had around 28 failures. 1 unsupported is info proc mappings which is not there in AIX as of now.

Kindly let me know what you think. I know that we have some dummy variables that is used to negate a similar problem using clang with lto. But if we can produce something that works for both AIX GCC and other targets Clang with LTO it will be great, and we can create a patch.

Have a nice day ahead.

Thanks and regards,
Aditya Kamath.