[ESS] Running gdb under emacs 23.1 and attaching to an ess-based R process

[Douglas Bates]

On Mon, Jan 17, 2011 at 3:29 AM, Martin Maechler
<maechler at stat.math.ethz.ch> wrote:
>>>>>> Martin Maechler <maechler at stat.math.ethz.ch>
>>>>>>     on Mon, 17 Jan 2011 10:15:17 +0100 writes:
>
>>>>>> Douglas Bates <bates at stat.wisc.edu>
>>>>>>     on Wed, 12 Jan 2011 15:24:52 -0600 writes:
>
>    >> In the past I had a wonderfully effective way of getting
>    >> at compiled code in an R package with gdb.  I would start
>    >> R with ESS (M-x R) and load the package.  Then I would
>    >> start gdb on the binary (M-x gdb <RET>
>    >> /usr/lib/R/bin/exec/R), find the process number of the R
>    >> process (M-!  pgrep R) and attach gdb to that
>    >> already-running process.  I would interrupt R, set
>    >> breakpoints and then send the sig 0 signal.
>
>    >> Now I am running emacs 23.1 (which, for some reason, is
>    >> still called emacs-snapshot in Ubuntu) with enhanced
>    >> support for gdb.  M-x gdb ends up opening multiple windows
>    >> and I can't get things straightened out to have the input
>    >> and output from the attached process associated with the
>    >> *R* buffer.
>
>    >> Has anyone been able to run the gdb/R combination with
>    >> emacs support?  My problem is that I am so accustomed to
>    >> the emacs/ess key sequences that I feel crippled when
>    >> trying to run R in any other environment :-)
>
>    > As a workaround :
>
>    > I've been using the much less sophisticated scheme,
>    > with the advantage that it has always worked on all versions of
>    > Emacs and R ..
>
>
>    > 1)  C-u M-x R  [Enter] -d gdb [Enter]    # starting R with gdb in a *R* buffer
>    > 2)  C-g {to interrupt Emacs waiting for R's prompt .. ouch!}
>    > 3)  C-x b *R ... to manually switch the *R..* buffer in which
>    > you now have a prompt '(gdb) '
>    > 4) r                 # start running R
>    > 5) require(...) ...  ## load all R packages and R code I need
>    > 6) C-c C-c         (interrupt): break out of R back into gdb
>    > 7) br ...             ## setting breakpoints in gdb for C code
>
>    > 8) r   # back to running R
>
> oops! this should have been
>
>      8) c   # [c]ontinue the running R ..
>
>
>
>    > 9) in R, call the code that will get into calling the break
>    > pointed C code
>    > 10) work in the *R* buffer entering gdb commands
>    > 11) "goto 8)"
>
>
>    > This allows to use ESS (Emacs R interface), almost normally,
>    > but of course the disadvantage is that this does not make any
>    > use of the Emacs gdb interface.
>
>    > We (ESS developers) could make steps 1) -- 3) more easy, notably
>    > allowing ESS to also detect "(gdb) " as an R prompt which should
>    > basically eliminate '2)' and '3)'.
>
>    > I agree that the real solution should make Doug's original way
>    > workable again, but we need someone (not me) tackle that...

I found a couple of ways around the problem.  One is to start gdb with

M-x gud-gdb

which starts the old interface.

If you want to use the new graphical interface you can set first set

(setq gdb-use-separate-io-buffer nil)

then start M-x gdb on the R executable.  The graphical interface still
gets a bit confused and, after "attach <R-PID>" you need to type "sig
0" or "b <position>" without a prompt.  Just hitting <Enter> at that
time bombs you out of gdb because it tries to re-execute the previous
line, which was the attach.

Also, there are peculiarities with trying to get rid of the *gud*
buffer.   You need to kill the buffer rather than quitting from gdb.
If you quit from gdb the buffer seems to become a zombie in that you
can't kill it.