[lldb] Generic base for testing gdb-remote behavior

[owenpshaw]

Summary

Adds new utilities that make it easier to write test cases for lldb acting as a client over a gdb-remote connection.

• A GDBRemoteTestBase class that starts a mock GDB server and provides an easy way to check client packets
• A MockGDBServer that, via MockGDBServerResponder, can be made to issue server responses that test client behavior.
• Utility functions for handling common data encoding/decoding
• Utility functions for creating dummy targets from YAML files

---

Split from the review at https://reviews.llvm.org/D42145, which was a new feature that necessitated the new testing capabilities.

[clayborg]

Looks fine to me. Wait for Pavel to give it the OK since he did more of the lldb-server testing stuff.

[labath]

My main comment is about making sure the tearDown story is sufficiently robust. I want to be sure we don't introduce flakyness here.

packages/Python/lldbsuite/test/functionalities/gdb_remote_client/TestGDBRemoteClient.py

13

Could you put some basic assertion here? I guess at this point we can expect to see QStartNoAckMode and/or some basic query packets (?)

packages/Python/lldbsuite/test/functionalities/gdb_remote_client/gdbclientutils.py

26

call this yaml2obj? We should be able to use this function to create a Mach-O file as well...

165–168

Maybe at least return valid hex digits here?

247–249

The concurrent access to self._client here makes me very uneasy. If this were C, it would definitely be a race, but I don't think python can do much to avoid a fd-reassignment race anyway. The exception handling code below only reaffirms my suspicion. I think we should make this safer to avoid test flakyness down the line.

I think a client-initiated shutdown should be sufficient. SBProcess.Kill() should be enough to make sure the client socket is closed. After that, you just need to join the server thread (which should exit after it detects a connection drop).

What do you think?

250–251

I don't think calling shutdown on a socket in listen mode is expected (i.e., just delete this comment).

281

Could you try break the _receive function down into simpler chunks. The nested while loops make this quite hard to follow. I suggest a design that would involve functions like:

getFullPacket # checks whether accumulated input contains full packet and returns it
getPayload # verifies checksum, unframes, and unescapes

and making sure all sending happens in _handlePacket only.

432

Add a message that the lines below are the last 10 packets received.

[owenpshaw]

Updated with suggestions

• Using process.Kill() to close client connection
• Refactored server._receive
• renamed elf-specific functions to be generic

[labath]

Thank you for making the changes. The parsing code looks much cleaner now. I just have one more round of tiny remarks.

packages/Python/lldbsuite/test/functionalities/gdb_remote_client/gdbclientutils.py

242

You've deleted closing of the listening socket altogether (I just wanted you to delete the comment). The socket itself needs to be closed, of course, I just don't think calling shutdown on it is needed.

401

There's already a TestBase.process() that should do what you need.

[labath]

Oh, and please add yaml2obj as a dependency of the check-lldb target in cmake.

[owenpshaw]

• Updated to use TestBase.process() instead of a new member
• Added yaml2obj dependency. It it okay to be an unconditional dependency?
• I can put the socket close() back in, but had removed it because python should close it automatically on garbage collection, so it didn't seem to matter either way

[labath]

In D42195#982035, @owenpshaw wrote:

• Added yaml2obj dependency. It it okay to be an unconditional dependency?

Yeah, it probably needs to be conditional (if(TARGET yaml2obj)), because standalone builds will not have that target. OTOH, standalone builds will probably also have an issue with finding yaml2obj. However, I don't think it should be up to you to make standalone builds work. cc'ing Kamil and Michal, as they are the ones who use those builds.

What I am not sure about is whether we need to make any changes to accomodate the XCode build. Jim, Davide: What would it take to make a test like this work from xcode?

• I can put the socket close() back in, but had removed it because python should close it automatically on garbage collection, so it didn't seem to matter either way

Yes, please put it back. There's no telling when GC will happen, so we shouldn't be holding onto relatively scarce os resources waiting for that...

[krytarowski]

yaml2obj looks like a program from LLVM and we build and install it in pkgsrc.

[owenpshaw]

Added back socket close()

It looks like the yaml2obj target hasn't been defined yet when the check-lldb target is defined, so if(TARGET yaml2obj) always returns false. Is there another way to do the conditional dependency? I'm assuming we don't want to reorder the includes in llvm/tools/CMakeLists.txt.

[labath]

In D42195#984003, @owenpshaw wrote:

It looks like the yaml2obj target hasn't been defined yet when the check-lldb target is defined, so if(TARGET yaml2obj) always returns false. Is there another way to do the conditional dependency? I'm assuming we don't want to reorder the includes in llvm/tools/CMakeLists.txt.

Ah, I see. We can then probably gate this on IF(NOT LLDB_BUILT_STANDALONE).

[krytarowski]

In the standalone build, we put yaml2obj into $PREFIX/bin/, so into the default PATH of the toolchain.

[owenpshaw]

• Updated condition for yaml2obj dependency, now based on LLDB_BUILT_STANDALONE
• Find yaml2obj in the same directory as clang
• Move yaml2obj code into lldbtest.py so it's available to all tests

Is the findBuiltClang function the right thing to use here? I can't find anything else that uses it.

[labath]

This is great. Thank you for doing this. I think we're ready to wrap this up.

[clayborg]

Thanks for iterating on this. This will be great.

[owenpshaw]

Thanks for the help! Someone with commit access will need to land it.

[labath]

packages/Python/lldbsuite/test/lldbtest.py

1588 ↗

(On Diff #131310)

I had to modify these to llvm-build/Release+Asserts/x86_64/bin/clang to be able to find the binary. I presume this is a change in xcode project layout. Can anyone confirm that?

[jingham]

The xcode build uses cmake to build llvm & clang. It passes a build flavor, by passing -DCMAKE_BUILD_TYPE=Debug for instance, but other than that it lets the llvm build figure out how to lay itself out.

Moreover, there are a bunch of defines that lldb sets up to find includes and some other bits from the clang build directory. The all assume that the LLVM_BUILD_ARCH is the last part of the path before you get to the actual good stuff. The uses of those defines was last touched on 02/17 (and that actually didn't change substance, it looks like it was just a line-ending change.) So I can't see how you would either get Release+Asserts/x86_64/Release+Asserts as a build layout or how the Xcode project would have dealt successfully with that.

[labath]

Thanks for the clarification. It seems nobody was calling that function, so maybe it never actually worked.

I'll try committing this for Owen with the clang paths modified and check whether the bots like it.

[labath]

lldb/trunk/packages/Python/lldbsuite/test/functionalities/gdb_remote_client/gdbclientutils.py

335–337 ↗

(On Diff #131754)

Which servers were you observing here? I tried lldb-server and gdbserver and they don't send the ack back (It seems like a bad idea anyway, as it could lead to an infinite ack ping-pong).

I have removed this code as it was causing flakyness on the bots. If there is a server that does this, and we care about supporting it, then we can add a new targeted test which emulates this behavior, but then we need to also fix the client to handle this situation better.

[owenpshaw]

lldb/trunk/packages/Python/lldbsuite/test/functionalities/gdb_remote_client/gdbclientutils.py

335–337 ↗

(On Diff #131754)

Just took another look and I misunderstood what was going on. The server does send an opening +, but not in response to the client.

I was observing the packets from openocd, where lldb sends a + and then receives a + from openocd, and assumed one was a response to the other.

(lldb) log enable gdb-remote packets
(lldb) gdb-remote 3333
                 <   1> send packet: +
                 history[1] tid=0x0307 <   1> send packet: +
                 <   1> read packet: +
                 <  19> send packet: $QStartNoAckMode#b0
                 <   1> read packet: +
                 <   6> read packet: $OK#9a
                 <   1> send packet: +

Can't find any docs about these opening acks, but looking at openocd code, it just always sends a + at the start of a connection. Should the test server do the same?

[clayborg]

I am pretty sure that the ack that gets sent should happen in response to getting an ACK. Note that LLDB send an ACK and then waits for an ACK. This helps us see if anything is there. So yes, this code can always just respond to an unsolicited ACK with an ACK.

[labath]

In D42195#995087, @clayborg wrote:

I am pretty sure that the ack that gets sent should happen in response to getting an ACK. Note that LLDB send an ACK and then waits for an ACK. This helps us see if anything is there. So yes, this code can always just respond to an unsolicited ACK with an ACK.

LLDB sends an ACK, and then waits for anything. That "anything" can also be "nothing" (see GDBRemoteCommunicationClient::HandshakeWithServer). lldb-server currently does not respond to the initial ACK, which is good, because the length of the initial client wait (10ms) would mean this reply-ack would often get missed, and the connection would get out of sync (as this code demonstrated).

If we really want to have our servers send an initial ack, then we need to fix the client.

packages/Python/lldbsuite/test/functionalities/gdb_remote_client/TestGDBRemoteClient.py

+import lldb
+from lldbsuite.test.lldbtest import *
+from lldbsuite.test.decorators import *
+from gdbclientutils import *
+
+
+class TestGDBRemoteClient(GDBRemoteTestBase):
+
+    def test_connect(self):
+        """Test connecting to a remote gdb server"""
+        target = self.createTarget("a.yaml")
+        process = self.connect(target)
+        self.assertPacketLogContains(["qProcessInfo", "qfThreadInfo"])

packages/Python/lldbsuite/test/functionalities/gdb_remote_client/a.yaml

+!ELF
+FileHeader:
+  Class:           ELFCLASS32
+  Data:            ELFDATA2LSB
+  Type:            ET_EXEC
+  Machine:         EM_ARM
+Sections:
+  - Name:            .text
+    Type:            SHT_PROGBITS
+    Flags:           [ SHF_ALLOC, SHF_EXECINSTR ]
+    Address:         0x1000
+    AddressAlign:    0x4
+    Content:         "c3c3c3c3"
+  - Name:            .data
+    Type:            SHT_PROGBITS
+    Flags:           [ SHF_ALLOC ]
+    Address:         0x2000
+    AddressAlign:    0x4
+    Content:         "3232"
+ProgramHeaders:
+  - Type: PT_LOAD
+    Flags: [ PF_X, PF_R ]
+    VAddr: 0x1000
+    PAddr: 0x1000
+    Align: 0x4
+    Sections:
+      - Section: .text
+  - Type: PT_LOAD
+    Flags: [ PF_R, PF_W ]
+    VAddr: 0x2000
+    PAddr: 0x1004
+    Align: 0x4
+    Sections:
+      - Section: .data

packages/Python/lldbsuite/test/functionalities/gdb_remote_client/gdbclientutils.py

+import os
+import os.path
+import subprocess
+import threading
+import socket
+import lldb
+from lldbsuite.test.lldbtest import *
+from lldbsuite.test import lldbtest_config
+
+
+def yaml2obj_executable():
+    """
+    Get the path to the yaml2obj executable, which can be used to create test
+    object files from easy to write yaml instructions.
+
+    Throws an Exception if the executable cannot be found.
+    """
+    # Tries to find yaml2obj at the same folder as the lldb
+    path = os.path.join(os.path.dirname(lldbtest_config.lldbExec), "yaml2obj")
+    if os.path.exists(path):
+        return path
+    raise Exception("yaml2obj executable not found")
+
+
+def yaml2obj(yaml_path, obj_path):
+    """
+    Create an object file at the given path from a yaml file at the given path.
+
+    Throws a subprocess.CalledProcessError if the object could not be created.
+    """
+    yaml2obj = yaml2obj_executable()
+    command = [yaml2obj, "-o=%s" % obj_path, yaml_path]
+    system([command])
+
+
+def checksum(message):
+    """
+    Calculate the GDB server protocol checksum of the message.
+
+    The GDB server protocol uses a simple modulo 256 sum.
+    """
+    check = 0
+    for c in message:
+        check += ord(c)
+    return check % 256
+
+
+def frame_packet(message):
+    """
+    Create a framed packet that's ready to send over the GDB connection
+    channel.
+
+    Framing includes surrounding the message between $ and #, and appending
+    a two character hex checksum.
+    """
+    return "$%s#%02x" % (message, checksum(message))
+
+
+def escape_binary(message):
+    """
+    Escape the binary message using the process described in the GDB server
+    protocol documentation.
+
+    Most bytes are sent through as-is, but $, #, and { are escaped by writing
+    a { followed by the original byte mod 0x20.
+    """
+    out = ""
+    for c in message:
+        d = ord(c)
+        if d in (0x23, 0x24, 0x7d):
+            out += chr(0x7d)
+            out += chr(d ^ 0x20)
+        else:
+            out += c
+    return out
+
+
+def hex_encode_bytes(message):
+    """
+    Encode the binary message by converting each byte into a two-character
+    hex string.
+    """
+    out = ""
+    for c in message:
+        out += "%02x" % ord(c)
+    return out
+
+
+def hex_decode_bytes(hex_bytes):
+    """
+    Decode the hex string into a binary message by converting each two-character
+    hex string into a single output byte.
+    """
+    out = ""
+    hex_len = len(hex_bytes)
+    while i < hex_len - 1:
+        out += chr(int(hex_bytes[i:i + 2]), 16)
+        i += 2
+    return out
+
+
+class MockGDBServerResponder:
+    """
+    A base class for handing client packets and issuing server responses for
+    GDB tests.
+
+    This handles many typical situations, while still allowing subclasses to
+    completely customize their responses.
+
+    Most subclasses will be interested in overriding the other() method, which
+    handles any packet not recognized in the common packet handling code.
+    """
+
+    registerCount = 40
+    packetLog = None
+
+    def __init__(self):
+        self.packetLog = []
+
+    def respond(self, packet):
+        """
+        Return the unframed packet data that the server should issue in response
+        to the given packet received from the client.
+        """
+        self.packetLog.append(packet)
+        if packet == "g":
+            return self.readRegisters()
+        if packet[0] == "G":
+            return self.writeRegisters(packet[1:])
+        if packet[0] == "p":
+            return self.readRegister(int(packet[1:], 16))
+        if packet[0] == "P":
+            register, value = packet[1:].split("=")
+            return self.readRegister(int(register, 16), value)
+        if packet[0] == "m":
+            addr, length = [int(x, 16) for x in packet[1:].split(',')]
+            return self.readMemory(addr, length)
+        if packet[0] == "M":
+            location, encoded_data = packet[1:].split(":")
+            addr, length = [int(x, 16) for x in location.split(',')]
+            return self.writeMemory(addr, encoded_data)
+        if packet[0:7] == "qSymbol":
+            return self.qSymbol(packet[8:])
+        if packet[0:10] == "qSupported":
+            return self.qSupported(packet[11:].split(";"))
+        if packet == "qfThreadInfo":
+            return self.qfThreadInfo()
+        if packet == "qC":
+            return self.qC()
+        if packet == "?":
+            return self.haltReason()
+        if packet[0] == "H":
+            return self.selectThread(packet[1], int(packet[2:], 16))
+        if packet[0:6] == "qXfer:":
+            obj, read, annex, location = packet[6:].split(":")
+            offset, length = [int(x, 16) for x in location.split(',')]
+            data, has_more = self.qXferRead(obj, annex, offset, length)
+            if data is not None:
+                return self._qXferResponse(data, has_more)
+            return ""
+        return self.other(packet)
+
+    def readRegisters(self):
+        return "00000000" * self.registerCount
+
+    def readRegister(self, register):
+        return "00000000"
+
+    def writeRegisters(self, registers_hex):
+        return "OK"
+
+    def writeRegister(self, register, value_hex):
+        return "OK"
+
+    def readMemory(self, addr, length):
+        return "00" * length
+
+    def writeMemory(self, addr, data_hex):
+        return "OK"
+
+    def qSymbol(self, symbol_args):
+        return "OK"
+
+    def qSupported(self, client_supported):
+        return "PacketSize=3fff;QStartNoAckMode+"
+
+    def qfThreadInfo(self):
+        return "l"
+
+    def qC(self):
+        return "QC0"
+
+    def haltReason(self):
+        # SIGINT is 2, return type is 2 digit hex string
+        return "S02"
+
+    def qXferRead(self, obj, annex, offset, length):
+        return None, False
+
+    def _qXferResponse(self, data, has_more):
+        return "%s%s" % ("m" if has_more else "l", escape_binary(data))
+
+    def selectThread(self, op, thread_id):
+        return "OK"
+
+    def other(self, packet):
+        # empty string means unsupported
+        return ""
+
+
+class MockGDBServer:
+    """
+    A simple TCP-based GDB server that can test client behavior by receiving
+    commands and issuing custom-tailored responses.
+
+    Responses are generated via the .responder property, which should be an
+    instance of a class based on MockGDBServerResponder.
+    """
+
+    responder = None
+    port = 0
+    _socket = None
+    _client = None
+    _thread = None
+    _receivedData = None
+    _receivedDataOffset = None
+    _shouldSendAck = True
+    _isExpectingAck = False
+
+    def __init__(self, port = 0):
+        self.responder = MockGDBServerResponder()
+        self.port = port
+        self._socket = socket.socket()
+
+    def start(self):
+        # Block until the socket is up, so self.port is available immediately.
+        # Then start a thread that waits for a client connection.
+        addr = ("127.0.0.1", self.port)
+        self._socket.bind(addr)
+        self.port = self._socket.getsockname()[1]
+        self._socket.listen(0)
+        self._thread = threading.Thread(target=self._run)
+        self._thread.start()
+
+    def stop(self):
+        self._thread.join()
+        self._thread = None
+
+    def _run(self):
+        # For testing purposes, we only need to worry about one client
+        # connecting just one time.
+        try:
+            # accept() is stubborn and won't fail even when the socket is
+            # shutdown, so we'll use a timeout
+            self._socket.settimeout(2.0)
+            client, client_addr = self._socket.accept()
+            self._client = client
+            # The connected client inherits its timeout from self._socket,
+            # but we'll use a blocking socket for the client
+            self._client.settimeout(None)
+        except:
+            return
+        self._shouldSendAck = True
+        self._isExpectingAck = False
+        self._receivedData = ""
+        self._receivedDataOffset = 0
+        data = None
+        while True:
+            try:
+                data = self._client.recv(4096)
+                if data is None or len(data) == 0:
+                    break
+            except Exception as e:
+                self._client.close()
+                break
+            self._receive(data)
+
+    def _receive(self, data):
+        """
+        Collects data, parses and responds to as many packets as exist.
+        Any leftover data is kept for parsing the next time around.
+        """
+        self._receivedData += data
+        try:
+            packet = self._parsePacket()
+            while packet is not None:
+                self._handlePacket(packet)
+                packet = self._parsePacket()
+        except self.InvalidPacketException:
+            self._client.close()
+
+    def _parsePacket(self):
+        """
+        Reads bytes from self._receivedData, returning:
+        - a packet's contents if a valid packet is found
+        - the PACKET_ACK unique object if we got an ack
+        - None if we only have a partial packet
+
+        Raises an InvalidPacketException if unexpected data is received
+        or if checksums fail.
+
+        Once a complete packet is found at the front of self._receivedData,
+        its data is removed form self._receivedData.
+        """
+        data = self._receivedData
+        i = self._receivedDataOffset
+        data_len = len(data)
+        if data_len == 0:
+            return None
+        if i == 0:
+            # If we're looking at the start of the received data, that means
+            # we're looking for the start of a new packet, denoted by a $.
+            # It's also possible we'll see an ACK here, denoted by a +
+            if data[0] == '+':
+                self._receivedData = data[1:]
+                return self.PACKET_ACK
+            if data[0] == '$':
+                i += 1
+            else:
+                raise self.InvalidPacketException(
+                        "Unexexpected leading byte: %s" % data[0])
+
+        # If we're looking beyond the start of the received data, then we're
+        # looking for the end of the packet content, denoted by a #.
+        # Note that we pick up searching from where we left off last time
+        while i < data_len and data[i] != '#':
+            i += 1
+
+        # If there isn't enough data left for a checksum, just remember where
+        # we left off so we can pick up there the next time around
+        if i > data_len - 3:
+            self._receivedDataOffset = i
+            return None
+
+        # If we have enough data remaining for the checksum, extract it and
+        # compare to the packet contents
+        packet = data[1:i]
+        i += 1
+        try:
+            check = int(data[i:i + 2], 16)
+        except ValueError:
+            raise self.InvalidPacketException("Checksum is not valid hex")
+        i += 2
+        if check != checksum(packet):
+            raise self.InvalidPacketException(
+                    "Checksum %02x does not match content %02x" %
+                    (check, checksum(packet)))
+        # remove parsed bytes from _receivedData and reset offset so parsing
+        # can start on the next packet the next time around
+        self._receivedData = data[i:]
+        self._receivedDataOffset = 0
+        return packet
+
+    def _handlePacket(self, packet):
+        if packet is self.PACKET_ACK:
+            # If we are expecting an ack, we'll just ignore it because there's
+            # nothing else we're supposed to do.
+            #
+            # However, if we aren't expecting an ack, it's likely the initial
+            # ack that lldb client sends, and observations of real servers
+            # suggest we're supposed to ack back.
+            if not self._isExpectingAck:
+                self._client.sendall('+')
+            return
+        response = ""
+        # We'll handle the ack stuff here since it's not something any of the
+        # tests will be concerned about, and it'll get turned off quicly anyway.
+        if self._shouldSendAck:
+            self._client.sendall('+')
+            self._isExpectingAck = True
+        if packet == "QStartNoAckMode":
+            self._shouldSendAck = False
+            response = "OK"
+        elif self.responder is not None:
+            # Delegate everything else to our responder
+            response = self.responder.respond(packet)
+        # Handle packet framing since we don't want to bother tests with it.
+        framed = frame_packet(response)
+        self._client.sendall(framed)
+
+    PACKET_ACK = object()
+
+    class InvalidPacketException(Exception):
+        pass
+
+
+class GDBRemoteTestBase(TestBase):
+    """
+    Base class for GDB client tests.
+
+    This class will setup and start a mock GDB server for the test to use.
+    It also provides assertPacketLogContains, which simplifies the checking
+    of packets sent by the client.
+    """
+
+    NO_DEBUG_INFO_TESTCASE = True
+    mydir = TestBase.compute_mydir(__file__)
+    server = None
+    temp_files = None
+
+    def setUp(self):
+        TestBase.setUp(self)
+        self.temp_files = []
+        self.server = MockGDBServer()
+        self.server.start()
+
+    def tearDown(self):
+        for temp_file in self.temp_files:
+            self.RemoveTempFile(temp_file)
+        # TestBase.tearDown will kill the process, but we need to kill it early
+        # so its client connection closes and we can stop the server before
+        # finally calling the base tearDown.
+        if self.process() is not None:
+            self.process().Kill()
+        self.server.stop()
+        self.temp_files = []
+        TestBase.tearDown(self)
+
+    def createTarget(self, yaml_path):
+        """
+        Create a target by auto-generating the object based on the given yaml
+        instructions.
+
+        This will track the generated object so it can be automatically removed
+        during tearDown.
+        """
+        yaml_base, ext = os.path.splitext(yaml_path)
+        obj_path = "%s" % yaml_base
+        yaml2obj(yaml_path, obj_path)
+        self.temp_files.append(obj_path)
+        return self.dbg.CreateTarget(obj_path)
+
+    def connect(self, target):
+        """
+        Create a process by connecting to the mock GDB server.
+
+        Includes assertions that the process was successfully created.
+        """
+        listener = self.dbg.GetListener()
+        error = lldb.SBError()
+        url = "connect://localhost:%d" % self.server.port
+        process = target.ConnectRemote(listener, url, "gdb-remote", error)
+        self.assertTrue(error.Success(), error.description)
+        self.assertTrue(process, PROCESS_IS_VALID)
+
+    def assertPacketLogContains(self, packets):
+        """
+        Assert that the mock server's packet log contains the given packets.
+
+        The packet log includes all packets sent by the client and received
+        by the server.  This fuction makes it easy to verify that the client
+        sent the expected packets to the server.
+
+        The check does not require that the packets be consecutive, but does
+        require that they are ordered in the log as they ordered in the arg.
+        """
+        i = 0
+        j = 0
+        log = self.server.responder.packetLog
+        while i < len(packets) and j < len(log):
+            if log[j] == packets[i]:
+                i += 1
+            j += 1
+        if i < len(packets):
+            self.fail("Did not receive: %s\nLast 10 packets:\n\t%s" %
+                    (packets[i], '\n\t'.join(log[-10:])))

test/CMakeLists.txt

 function(add_python_test_target name test_script args comment)
   set(PYTHON_TEST_COMMAND
     ${PYTHON_EXECUTABLE}
     ${test_script}
     ${args}
     )
 
   add_custom_target(${name}
     COMMAND ${PYTHON_TEST_COMMAND} ${ARG_DEFAULT_ARGS}
     COMMENT "${comment}"
     DEPENDS ${LLDB_TEST_DEPS}
     USES_TERMINAL
     )
 endfunction()
 
-set(LLDB_TEST_DEPS lldb)
+set(LLDB_TEST_DEPS lldb yaml2obj)
 
 # darwin-debug is an hard dependency for the testsuite.
 if (CMAKE_SYSTEM_NAME MATCHES "Darwin")
   list(APPEND LLDB_TEST_DEPS darwin-debug)
 endif()
 
 if(TARGET lldb-server)
   list(APPEND LLDB_TEST_DEPS lldb-server)