• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1Projects
2========
3
4The following is a mostly unordered set of the ideas for improvements to the
5LLDB debugger. Some are fairly deep, some would require less effort.
6
7.. contents::
8   :local:
9
10Speed up type realization in lldb
11---------------------------------
12
13The type of problem I'm addressing here is the situation where you are
14debugging a large program (lldb built with debug clang/swift will do) and you
15go to print a simple expression, and lldb goes away for 30 seconds. When you
16sample it, it is always busily churning through all the CU's in the world
17looking for something.  The problem isn't that looking for something in
18particular is slow, but rather that we somehow turned an bounded search (maybe
19a subtype of "std::string" into an unbounded search (all things with the name
20of that subtype.)  Or didn't stop when we got a reasonable answer proximate to
21the context of the search, but let the search leak out globally. And quite
22likely there are other issues that I haven't guessed yet. But if you end up
23churning though 3 or 4 Gig of debug info, that's going to be slow no matter how
24well written your debug reader is...
25
26My guess is the work will be more in the general symbol lookup than in the
27DWARF parser in particular, but it may be a combination of both.
28
29As a user debugging a largish program, this is the most obvious lameness of
30lldb.
31
32Symbol name completion in the expression parser
33-----------------------------------------------
34
35This is the other obvious lameness of lldb.  You can do:
36
37::
38
39   (lldb) frame var foo.b
40
41and we will tell you it is "foo.bar". But you can't do that in the expression
42parser. This will require collaboration with the clang/swift folks to get the
43right extension points in the compiler. And whatever they are, lldb will need
44use them to tell the compiler about what names are available. It will be
45important to avoid the pitfalls of #1 where we wander into the entire DWARF
46world.
47
48Make a high speed asynchronous communication channel
49----------------------------------------------------
50
51All lldb debugging nowadays is done by talking to a debug agent. We used the
52gdb-remote protocol because that is universal, and good enough, and you have to
53support it anyway since so many little devices & JTAG's and VM's etc support
54it. But it is really old, not terribly high performance, and can't really
55handle sending or receiving messages while the process is supposedly running.
56It should have compression built in, remove the hand-built checksums and rely
57on the robust communication protocols we always have nowadays, allow for
58out-of-order requests/replies, allow for reconnecting to a temporarily
59disconnected debug session, regularize all of the packet formatting into JSON
60or BSON or whatever while including a way to do large binary transfers. It must
61be possible to come up with something faster, and better tunable for the many
62communications pathways we end up supporting.
63
64Fix local variable lookup in the lldb expression parser
65-------------------------------------------------------
66
67The injection of local variables into the clang expression parser is
68currently done incorrectly - it happens too late in the lookup. This results
69in namespace variables & functions, same named types and ivars shadowing
70locals when it should be the other way around. An attempt was made to fix
71this by manually inserting all the visible local variables into wrapper
72function in the expression text. This mostly gets the job done but that
73method means you have to realize all the types and locations of all local
74variables for even the simplest of expressions, and when run on large
75programs (e.g. lldb) it would cause unacceptable delays. And it was very
76fragile since an error in realizing any of the locals would cause all
77expressions run in that context to fail. We need to fix this by adjusting
78the points where name lookup calls out to lldb in clang.
79
80Support calling SB & commands everywhere and support non-stop debugging
81-----------------------------------------------------------------------
82
83There is a fairly ad-hoc system to handle when it is safe to run SB API's and
84command line commands. This is actually a bit of a tricky problem, since we
85allow access to the command line and SB API from some funky places in lldb. The
86Operating System plugins are the most obvious instance, since they get run
87right after lldb is told by debugserver that the process has stopped, but
88before it has finished collating the information from the stop for presentation
89to the higher levels. But breakpoint callbacks have some of the same problems,
90and other things like the scripted stepping operations and any fancier
91extension points we want to add to the debugger are going to be hard to
92implement robustly till we work on a finer-grained and more explicit control
93over who gets to control the process state.
94
95We also won't have any chance of supporting non-stop debugging - which is a
96useful mode for programs that have a lot of high-priority or real-time worker
97threads - until we get this sorted out.
98
99Finish the language abstraction and remove all the unnecessary API's
100--------------------------------------------------------------------
101
102An important part of making lldb a more useful "debugger toolkit" as opposed to
103a C/C++/ObjC/Swift debugger is to have a clean abstraction for language
104support. We did most, but not all, of the physical separation.  We need to
105finish that. And then by force of necessity the API's really look like the
106interface to a C++ type system with a few swift bits added on.  How you would
107go about adding a new language is unclear and much more trouble than it is
108worth at present. But if we made this nice, we could add a lot of value to
109other language projects.
110
111Add some syntax to generate data formatters from type definitions
112-----------------------------------------------------------------
113
114Uses of the data formatters fall into two types. There are data formatters for
115types where the structure elements pretty much tell you how to present the
116data, you just need a little expression language to express how to turn them
117into what the user expects to see. Then there are the ones (like pretty much
118all our Foundation/AppKit/UIKit formatters) that use deep magic to figure out
119how the type is actually laid out. The latter are pretty much always going to
120have to be done by hand.
121
122But for the ones where the information is expressed in the fields, it would be
123great to have a way to express the instructions to produce summaries and
124children in some form you could embed next to the types and have the compiler
125produce a byte code form of the instructions and then make that available to
126lldb along with the library. This isn't as simple as having clang run over the
127headers and produce something from the types directly. After all, clang has no
128way of knowing that the interesting thing about a std::vector is the elements
129that you get by calling size (for the summary) and [] for the elements. But it
130shouldn't be hard to come up with a generic markup to express this.
131
132Allow the expression parser to access dynamic type/data formatter information
133-----------------------------------------------------------------------------
134
135This seems like a smaller one. The symptom is your object is Foo child of
136Bar, and in the Locals view you see all the fields of Foo, but because the
137static type of the object is Bar, you can't see any of the fields of Foo.
138But if you could get this working, you could hijack the mechanism to make
139the results of the value object summaries/synthetic children available to
140expressions. And if you can do that, you could add other properties to an
141object externally (through Python or some other extension point) and then
142have these also available in the expression parser. You could use this to
143express invariants for data structures, or other more advanced uses of types
144in the debugger.
145
146Another version of this is to allow access to synthetic children in the
147expression parser. Otherwise you end up in situations like:
148
149::
150
151  (lldb) print return_a_foo()
152  (SomeVectorLikeType) $1 = {
153    [0] = 0
154    [1] = 1
155    [2] = 2
156    [3] = 3
157    [4] = 4
158  }
159
160That's good but:
161
162::
163
164  (lldb) print return_a_foo()[2]
165
166fails because the expression parser doesn't know anything about the
167array-like nature of SomeVectorLikeType that it gets from the synthetic
168children.
169
170Recover thread information lazily
171---------------------------------
172
173LLDB stores all the user intentions for a thread in the ThreadPlans stored in
174the Thread class. That allows us to reliably implement a very natural model for
175users moving through a debug session. For example, if step-over stops at a
176breakpoint in an function in a younger region of the stack, continue will
177complete the step-over rather than having to manually step out. But that means
178that it is important that the Thread objects live as long as the Threads they
179represent. For programs with many threads, but only one that you are debugging,
180that makes stepping less efficient, since now you have to fetch the thread list
181on every step or stepping doesn't work correctly. This is especially an issue
182when the threads are provided by an Operating System plugin, where it may take
183non-trivial work to reconstruct the thread list. It would be better to fetch
184threads lazily but keep "unseen" threads in a holding area, and only retire
185them when we know we've fetched the whole thread list and ensured they are no
186longer alive.
187
188Make Python-backed commands first class citizens
189------------------------------------------------
190
191As it stands, Python commands have no way to advertise their options. They are
192required to parse their arguments by hand. That leads to inconsistency, and
193more importantly means they can't take advantage of auto-generated help and
194command completion. This leaves python-backed commands feeling worse than
195built-in ones.
196
197As part of this job, it would also be great to hook automatically hook the
198"type" of an option value or argument (e.g. eArgTypeShlibName) to sensible
199default completers. You need to be able to over-ride this in more complicated
200scenarios (like in "break set" where the presence of a "-s" option limits the
201search for completion of a "-n" option.) But in common cases it is unnecessary
202busy-work to have to supply the completer AND the type. If this worked, then it
203would be easier for Python commands to also get correct completers.
204
205Reimplement the command interpreter commands using the SB API
206-------------------------------------------------------------
207
208Currently, all the CommandObject::DoExecute methods are implemented using the
209lldb_private API's. That generally means that there's code that gets duplicated
210between the CommandObject and the SB API that does roughly the same thing. We
211would reduce this code duplication, present a single coherent face to the users
212of lldb, and keep ourselves more honest about what we need in the SB API's if
213we implemented the CommandObjects::DoExecute methods using the SB API's.
214
215BTW, it is only the way it was much easier to develop lldb if it had a
216functioning command-line early on. So we did that first, and developed the SB
217API's when lldb was more mature. There's no good technical reason to have the
218commands use the lldb_private API's.
219
220Documentation and better examples
221---------------------------------
222
223We need to put the lldb syntax docs in the tutorial somewhere that is more
224easily accessible. On suggestion is to add non-command based help to the help
225system, and then have a "help lldb" or "help syntax" type command with this
226info. Be nice if the non-command based help could be hierarchical so you could
227make topics.
228
229There's a fair bit of docs about the SB API's, but it is spotty. Some classes
230are well documented in the Python "help (lldb.SBWhatever)" and some are not.
231
232We need more conceptual docs. And we need more examples. And we could provide a
233clean pluggable example for using LLDB standalone from Python. The
234process_events.py is a start of this, but it just handles process events, and
235it is really a quick sketch not a polished expandable proto-tool.
236
237Make a more accessible plugin architecture for lldb
238---------------------------------------------------
239
240Right now, you can only use the Python or SB API's to extend an extant lldb.
241You can't implement any of the actual lldb Plugins as plugins. That means
242anybody that wants to add new Object file/Process/Language etc support has to
243build and distribute their own lldb. This is tricky because the API's the
244plugins use are currently not stable (and recently have been changing quite a
245lot.) We would have to define a subset of lldb_private that you could use, and
246some way of telling whether the plugins were compatible with the lldb. But
247long-term, making this sort of extension possible will make lldb more appealing
248for research and 3rd party uses.
249
250Use instruction emulation to reduce the overhead for breakpoints
251----------------------------------------------------------------
252
253At present, breakpoints are implemented by inserting a trap instruction, then
254when the trap is hit, replace the trap with the actual instruction and single
255step. Then swap back and continue. This causes problems for read only text, and
256also means that no-stop debugging ust either stop all threads briefly to handle
257this two-step or risk missing some breakpoint hits. If you emulated the
258instruction and wrote back the results, you wouldn't have these problems, and
259it would also save a stop per breakpoint hit. Since we use breakpoints to
260implement stepping, this savings could be significant on slow connections.
261
262Use the JIT to speed up conditional breakpoint evaluation
263---------------------------------------------------------
264
265We already JIT and cache the conditional expressions for breakpoints for the C
266family of languages, so we aren't re-compiling every time you hit the
267breakpoint. And if we couldn't IR interpret the expression, we leave the JIT'ed
268code in place for reuse. But it would be even better if we could also insert
269the "stop or not" decision into the code at the breakpoint, so you would only
270actually stop the process when the condition was true. Greg's idea was that if
271you had a conditional breakpoint set when you started the debug session, Xcode
272could rebuild and insert enough no-ops that we could instrument the breakpoint
273site and call the conditional expression, and only trap if the conditional was
274true.
275
276Broaden the idea in "target stop-hook" to cover more events in the debugger
277---------------------------------------------------------------------------
278
279Shared library loads, command execution, User directed memory/register reads
280and writes are all places where you would reasonably want to hook into the
281debugger.
282
283Mock classes for testing
284------------------------
285
286We need "ProcessMock" and "ObjectFileMock" and the like. These would be real
287plugin implementations for their underlying lldb classes, with the addition
288that you can prime them from some sort of text based input files. For classes
289that manage changes over time (like process) you would need to program the
290state at StopPoint 0, StopPoint 1, etc. These could then be used for testing
291reactions to complex threading problems & the like, and also for simulating
292hard-to-test environments (like bare board debugging).
293
294Expression parser needs syntax for "{symbol,type} A in CU B.cpp"
295----------------------------------------------------------------
296
297Sometimes you need to specify non-visible or ambiguous types to the expression
298parser. We were planning to do $b_dot_cpp$A or something like. You might want
299to specify a static in a function, in a source file, or in a shared library. So
300the syntax should support all these.
301
302Add a "testButDontAbort" style test to the UnitTest framework
303-------------------------------------------------------------
304
305The way we use unittest now (maybe this is the only way it can work, I don't
306know) you can't report a real failure and continue with the test. That is
307appropriate in some cases: if I'm supposed to hit breakpoint A before I
308evaluate an expression, and don't hit breakpoint A, the test should fail. But
309it means that if I want to test five different expressions, I can either do it
310in one test, which is good because it means I only have to fire up one process,
311attach to it, and get it to a certain point. But it also means if the first
312test fails, the other four don't even get run. So though at first we wrote a
313bunch of test like this, as time went on we switched more to writing "one at a
314time" tests because they were more robust against a single failure. That makes
315the test suite run much more slowly. It would be great to add a
316"test_but_dont_abort" variant of the tests, then we could gang tests that all
317drive to the same place and do similar things. As an added benefit, it would
318allow us to be more thorough in writing tests, since each test would have lower
319costs.
320
321Convert the dotest style tests to use lldbutil.run_to_source_breakpoint
322-----------------------------------------------------------------------
323
324run_to_source_breakpoint & run_to_name_breakpoint provide a compact API that
325does in one line what the first 10 or 20 lines of most of the old tests now do
326by hand. Using these functions makes tests much more readable, and by
327centralizing common functionality will make maintaining the testsuites easier
328in the future. This is more of a finger exercise, and perhaps best implemented
329by a rule like: "If you touch a test case, and it isn't using
330run_to_source_breakpoint, please make it do so".
331
332Unify Watchpoint's & Breakpoints
333--------------------------------
334
335Option handling isn't shared, and more importantly the PerformAction's have a
336lot of duplicated common code, most of which works less well on the Watchpoint
337side.
338
339Reverse debugging
340-----------------
341
342This is kind of a holy grail, it's hard to support for complex apps (many
343threads, shared memory, etc.) But it would be SO nice to have...
344
345Non-stop debugging
346------------------
347
348By this I mean allowing some threads in the target program to run while
349stopping other threads. This is supported in name in lldb at present, but lldb
350makes the assumption "If I get a stop, I won't get another stop unless I
351actually run the program." in a bunch of places so getting it to work reliably
352will be some a good bit of work. And figuring out how to present this in the UI
353will also be tricky.
354
355Fix and continue
356----------------
357
358We did this in gdb without a real JIT. The implementation shouldn't be that
359hard, especially if you can build the executable for fix and continue. The
360tricky part is how to verify that the user can only do the kinds of fixes that
361are safe to do. No changing object sizes is easy to detect, but there were many
362more subtle changes (function you are fixing is on the stack...) that take more
363work to prevent. And then you have to explain these conditions the user in some
364helpful way.
365
366Unified IR interpreter
367----------------------
368
369Currently IRInterpreter implements a portion of the LLVM IR, but it doesn't
370handle vector data types and there are plenty of instructions it also doesn't
371support. Conversely, lli supports most of LLVM's IR but it doesn't handle
372remote memory and its function calling support is very rudimentary. It would be
373useful to unify these and make the IR interpreter -- both for LLVM and LLDB --
374better. An alternate strategy would be simply to JIT into the current process
375but have callbacks for non-stack memory access.
376
377Teach lldb to predict exception propagation at the throw site
378-------------------------------------------------------------
379
380There are a bunch of places in lldb where we need to know at the point where an
381exception is thrown, what frame will catch the exception.
382
383For instance, if an expression throws an exception, we need to know whether the
384exception will be caught in the course of the expression evaluation.  If so it
385would be safe to let the expression continue.  But since we would destroy the
386state of the thread if we let the exception escape the expression, we currently
387stop the expression evaluation if we see a throw.  If we knew where it would be
388caught we could distinguish these two cases.
389
390Similarly, when you step over a call that throws, you want to stop at the throw
391point if you know the exception will unwind past the frame you were stepping in,
392but it would annoying to have the step abort every time an exception was thrown.
393If we could predict the catching frame, we could do this right.
394
395And of course, this would be a useful piece of information to display when stopped
396at a throw point.
397
398Add predicates to the nodes of settings
399---------------------------------------
400
401It would be very useful to be able to give values to settings that are dependent
402on the triple, or executable name, for targets, or on whether a process is local
403or remote, or on the name of a thread, etc.  The original intent (and there is
404a sketch of this in the settings parsing code) was to be able to say:
405
406::
407
408  (lldb) settings set target{arch=x86_64}.process.thread{name=foo}...
409
410The exact details are still to be worked out, however.
411
412Resurrect Type Validators
413-------------------------
414
415This half-implemented feature was removed in
416https://reviews.llvm.org/D71310 but the general idea might still be
417useful: Type Validators look at a ValueObject, and make sure that
418there is nothing semantically wrong with the object's contents to
419easily catch corrupted data.
420