[Bug gdb/30008] New: gdb fails to get line numbers with gold/mold with --gdb-index

["rui314 at gmail dot com" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=30008

            Bug ID: 30008
           Summary: gdb fails to get line numbers with gold/mold with
                    --gdb-index
           Product: gdb
           Version: HEAD
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: gdb
          Assignee: unassigned at sourceware dot org
          Reporter: rui314 at gmail dot com
  Target Milestone: ---

I found this gdb's odd behavior when I was debugging a mold-generated
executable, but I could reproduce the same issue with GNU gold, so I'll report
it as a gdb issue. If it is actually a linker's issue, please let me know so
that I can fix it on the linker side. Since the GDB index is a gdb-specific
feature and not very documented well, it is very hard for me to debug this
issue.

So, here is the issue. I built the mold linker (https://github.com/rui314/mold)
with GNU ld and ran it under gdb. Almost all functions in mold are actually C++
templates that takes a target type as a type parameter. Therefore, for examle,
we have `parse_nonpositional_args<mold::elf::X86_64>`,
`parse_nonpositional_args<mold::elf::ARM32>`, etc in the same executable.

When I set a breakpoint at just `parse_nonpositional_args`, gdb fails to obtain
its line number, as shown below.


ruiu@blue:~/mold/build-gold$ gdb ./mold -nx -batch -ex 'b
parse_nonpositional_args' -ex 'run --no-fork --no-threads' -ex bt -ex c -ex bt
-ex quit

Breakpoint 1 at 0x1194e4 (16 locations)
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".

Breakpoint 1, 0x000055555566d4e4 in
std::vector<std::__cxx11::basic_string<char, std::char_traits<char>,
std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char,
std::char_traits<char>, std::allocator<char> > > >
mold::elf::parse_nonpositional_args<mold::elf::X86_64>(mold::elf::Context<mold::elf::X86_64>&)
()
#0  0x000055555566d4e4 in std::vector<std::__cxx11::basic_string<char,
std::char_traits<char>, std::allocator<char> >,
std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>,
std::allocator<char> > > >
mold::elf::parse_nonpositional_args<mold::elf::X86_64>(mold::elf::Context<mold::elf::X86_64>&)
()
#1  0x0000555555886568 in int mold::elf::elf_main<mold::elf::X86_64>(int,
char**) ()
#2  0x000055555588212b in mold::elf::main(int, char**) ()
#3  0x0000555556dba344 in main (argc=3, argv=0x7fffffffe178) at
/home/ruiu/mold/common/main.cc:155
mold: fatal: -m option is missing
[Inferior 1 (process 3523615) exited with code 01]
No stack.



However, if I specify the exact name of the function (in this case
`parse_nonpositional_args<mold::elf::X86_64>`), gdb can obtain its line number
without problems.


ruiu@blue:~/mold/build-gold$ gdb ./mold -nx -batch -ex 'b
parse_nonpositional_args<mold::elf::X86_64>' -ex 'run --no-fork --no-threads'
-ex bt -ex c -ex bt -ex quit

Breakpoint 1 at 0x119504: file /home/ruiu/mold/elf/cmdline.cc, line 395.
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".

Breakpoint 1, mold::elf::parse_nonpositional_args<mold::elf::X86_64> (ctx=...)
at /home/ruiu/mold/elf/cmdline.cc:395
395       std::span<std::string_view> args = ctx.cmdline_args;
#0  mold::elf::parse_nonpositional_args<mold::elf::X86_64> (ctx=...) at
/home/ruiu/mold/elf/cmdline.cc:395
#1  0x0000555555886568 in mold::elf::elf_main<mold::elf::X86_64> (argc=3,
argv=0x7fffffffe178) at /home/ruiu/mold/elf/elf-main.cc:387
#2  0x000055555588212b in mold::elf::main (argc=3, argv=0x7fffffffe178) at
/home/ruiu/mold/elf/elf-main.cc:758
#3  0x0000555556dba344 in main (argc=3, argv=0x7fffffffe178) at
/home/ruiu/mold/common/main.cc:155
mold: fatal: -m option is missing
[Inferior 1 (process 3524620) exited with code 01]
No stack.


Why does gdb behave differently for the above two test cases?

Uploaded a gold-linked mold executable here:
https://drive.google.com/file/d/1YuLUCJFJV-cJ9zQ7NRxvSSqYgoShCF_N/view?usp=sharing

-- 
You are receiving this mail because:
You are on the CC list for the bug.