Drop support for GDB < 9.2

[disconnect3d]

We do have this in README:

Pwndbg is supported on Ubuntu 20.04, and 22.04 with GDB 9.2 and later.

So we should remove all the old workaround code we used for old GDBs < 9.2. This issue is there to track those efforts: feel free to send a PR fixing just one thing, or a PR fixing many or all of those.

Those are:

• gdb77_get_register and the relevant "if hasattr" code:

  pwndbg/pwndbg/gdblib/regs.py

  Lines 22 to 35 in 1c4307d

  	@pwndbg.gdblib.proc.OnlyWhenRunning
  	def gdb77_get_register(name: str):
  	return gdb.parse_and_eval("$" + name)
  	@pwndbg.gdblib.proc.OnlyWhenRunning
  	def gdb79_get_register(name: str):
  	return gdb.selected_frame().read_register(name)
  	if hasattr(gdb.Frame, "read_register"):
  	get_register = gdb79_get_register
  	else:
  	get_register = gdb77_get_register

• _fs_gs_helper - we don't need to use ptrace anymore iiuc:

  pwndbg/pwndbg/gdblib/regs.py

  Lines 187 to 190 in 1c4307d

  	@pwndbg.lib.cache.cache_until("stop")
  	def _fs_gs_helper(self, regname: str, which):
  	"""Supports fetching based on segmented addressing, a la fs:[0x30].
  	Requires ptrace'ing the child directly for GDB < 8."""

• We probably don't need to try/except this anymore:

  pwndbg/pwndbg/gdblib/events.py

  Lines 120 to 125 in 1c4307d

  	# GDB 7.9 and above only
  	try:
  	registered[gdb.events.memory_changed] = []
  	registered[gdb.events.register_changed] = []
  	except (NameError, AttributeError):
  	pass

• The description for entry command mentions GDB 8.1 - we can probably drop the parentheses here:

  pwndbg/pwndbg/commands/start.py

  Lines 87 to 89 in 1c4307d

  	Note that the entrypoint may not be the first instruction executed
  	by the program. If you want to stop on the first executed instruction,
  	use the GDB's `starti` command (added in GDB 8.1).

• There was some weird GDB 8.2 case in check_repeated, a functionality for "repeating commands execution" - the except can probably be removed, but I recommend investigating the linked issue there:

  pwndbg/pwndbg/commands/__init__.py

  Lines 166 to 171 in 1c4307d

  	try:
  	number = int(number_str)
  	except ValueError:
  	# Workaround for a GDB 8.2 bug when show commands return error value
  	# See issue #523
  	return False

• Handling version:

commands/version.py
29:        return gdb.VERSION  # GDB >= 8.1 (or earlier?)
188:    # 8. showing width

• Some heap cases:

heap/structs.py
40:    # alignof doesn't available in GDB < 8.2 (https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=blob_plain;f=gdb/NEWS;hb=gdb-8.2-release)

heap/ptmalloc.py
974:            # alignof doesn't available in GDB < 8.2 (https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=blob_plain;f=gdb/NEWS;hb=gdb-8.2-release)
977:            # TODO: We can remove this when we drop supports for GDB < 8.2

• Handling symbols:

gdblib/symbol.py
167:            # GDB < 9.x does not have `gdb.lookup_static_symbol`
170:            # TODO/FIXME: Find a way to get the static linkage symbol's address in GDB < 9.x

[dmur1]

i'll take a look at this one

[dmur1]

_fs_gs_helper - we don't need to use ptrace anymore iiuc:

in trying to remove this the tests break - specifically tls command test - doing ptrace when pwndbg.gdblib.arch.current != "x86-64" seems to fix the issue which suggests that ptrace may be required in some cases even if read register is supported @disconnect3d

[disconnect3d]

@dmur1 seems that GDB doesnt allow to read fs/gs on x86 (i386) via read register and that's why we need to use ptrace there.

[disconnect3d]

Also: would be nice to investigate this:

pwndbg/pwndbg/gdblib/events.py

Lines 71 to 75 in 7a528bb

	class EventWrapper:
	"""
	Wrapper for GDB events which may not exist on older GDB versions but we still can
	fire them manually (to invoke them you have to call `invoke_callbacks`).
	"""

We may be able to remove that abstraction nowadays? I am not sure! :P

[disconnect3d]

I'm moving the remaining thing to issue #1887. I believe that we finally dropped support for <9.2.