[lldb-dev] lldb access in Emacs via realgud

Rocky Bernstein via lldb-dev lldb-dev at lists.llvm.org
Thu Sep 12 16:45:30 PDT 2019

On Thu, Sep 12, 2019 at 5:51 PM Adrian Prantl <aprantl at apple.com> wrote:

> > On Sep 12, 2019, at 2:24 PM, Rocky Bernstein <rb at dustyfeet.com> wrote:
> >
> >
> >
> > On Thu, Sep 12, 2019 at 3:37 PM Adrian Prantl <aprantl at apple.com> wrote:
> >>
> >> > On Sep 11, 2019, at 1:08 AM, Rocky Bernstein via lldb-dev <
> lldb-dev at lists.llvm.org> wrote:
> >> >
> >> >
> >> > Hi - I just wanted to mention that if there are emacs users there is
> an interface to lldb via realgud. See
> https://github.com/realgud/realgud-lldb
> >> >
> >> > A MELPA package and ELPA packageElpa package are available too
> >>
> >> Nice. It's always exciting to see wider adoption through better editor
> integration. Out of curiosity, how does this compare to regular gud.el? (
> https://opensource.apple.com/source/lldb/lldb-69/utils/emacs/gud.el.auto.html
> )
> >
> > "Regular" gud? The most recent copyright on that link is 2008. I see a
> gud.el in 26.2 and in the GNU savannah git sources, but neither mentions
> lldb. Assuming that file is really from 2008, has lldb changed since then?
> (This is a rhetorical question).  But the broader question is really who is
> maintaining that file you link, clearly it is not the GNU Emacs community.
> And how easy is it to do so? I see an "arch" tag on the file, so I guess
> this in version control somewhere. But if there is a bug in this file, what
> does one do? (This is not a rhetorical question; if you know the answer, I
> am interested.)
> I think this file is effectively abandoned as it is neither part of the
> LLDB repository nor is it shipping with emacs in macOS any longer.
> > Adapted from https://github.com/realgud/realgud/blob/master/realgud.el
> >
> >> Here we make use of more modern programming practices, more numerous
> and smaller files, unit tests, and better use of Emacs primitives, e.g.
> buffer marks, buffer-local variables, structures, rings, hash tables.
> Although there is still much to be desired, this code is more scalable and
> suitable as a common base for an Emacs front-end to modern debuggers.
> >> Oh, and because global variables are largely banned, we can support
> several simultaneous debug sessions.
> > gdb-mi has a nicer multi-frame display,  but you were linking to gud.el
> which doesn't have that as far as I know. realgud-lldb at this point
> probably knows more about lldb. But you guys can probably verify that, and
> if realgud-lldb is lacking, I'd be interested to learn what should be
> added.
> >
> > gud has always been a bit too monolithic - it contains every debugger
> (including some obsolete ones - does anyone really still use dbx, and if so
> inside Emacs?). All of this is in that one several-thousand-line file. I
> find this ironic because the principal author is wrote something about
> "Cathedral versus Bazaar" and this is clearly Cathedral style.
> >
> > It took me quite a while to be able to break realgud into several
> distinct files. In fact I had to write my own nodejs-like "require" package
> to be able to do internal or relative module linking. And then after that,
> more work was done to split off the debuggers from the core debugger module
> into separate github projects.
> Thanks for explaining the differences!
> Is realgud scraping lldb's console output like gud was or is it using the
> LLDB scripting API? If it isn't already you might want to consider using
> the scripting API since LLDB's console output is not necessarily stable,
> but the scripting API is.

Yes, it scrapes console output and that is a problem. A big problem.

I looked at the LLDB API and that seems pretty extensive and seems to cover
a lot more of what interaction from Emacs would like and is currently

However as is in its current state, this isn't going to cut it because
Emacs uses its own Lisp, not Python.

I have toyed with the idea of hooking into something more standard, and as
I said, there are so many to choose from. For example, the V8 debugger
protcol works over a websocket,  and speaking over a websocket is something
you can expect IDEs to grok.  Python LLDB's API to say Microsoft's Debug
Adaptor Protocol <https://microsoft.github.io/debug-adapter-protocol/> makes
sense, and https://github.com/vadimcn/vscode-lldb/blob/v1.3.0/MANUAL.md seems
to get pretty close to that.

> >> >
> >> > A question: what ever became of the effort to port the Emacs gdb-mi
> to lldb?
> >>
> >> We recently removed lldb-mi from the LLDB repository because nobody in
> the community was willing to maintain it. In particular the tests were so
> unreliable that most bots disabled them wholesale because they were so
> noisy. We had a GSoC student a year ago who was able to rewrite many of the
> tests in a more reliable fashion, but there were still a lot of issues
> outstanding after the project was completed. If you are interested in
> picking this up, it may be worthwhile to think about implementing lldb-mi
> 2.0 as thin python layer using the python SBAPI. Python may be a better
> choice for the kind of text-heavy glue-code that lldb-mi is. Alternatively
> it also shouldn't be hard at all to revive the existing C++ code. It's
> written in a different style than most of LLDB or LLVM (and IMO it should
> have never been accepted upstream in this form), but it shouldn't be hard
> to get building since it (thanks to the GSoC project!) is using only the
> stable public SBAPI.
> >
> >
> > The great thing about standards is that there are so many to choose
> from!
> It sounds like you are very much invested in realgud, but if anyone else
> is reading this and interested in taking up maintainership for lldb-mi, I
> think we'd be happy to welcome it back in tree as long as it is 100%
> reliably(!) tested and maintained.

Or since we are soliciting help, I'd rather see something that goes to the
Microsoft Debug Adaptor Protocol if vscode-lldb doesn't fit that already.
It might even just be repackaging parts of that code so that it can present
itself that way. In the long term I don't see either gdb-mi and lldb-mi as
viable solution that is going to reduce effort across different IDE's like
the Microsoft Debug Adaptor Protocol could/does.

> -- adrian
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-dev/attachments/20190912/c5986133/attachment.html>

More information about the lldb-dev mailing list