[Rd] meaning of browser(skipCalls=) [and multiple mouse buttons]

Martin Maechler m@ech|er @end|ng |rom @t@t@m@th@ethz@ch
Thu Dec 16 10:02:13 CET 2021


>>>>> Frederick Eaton 
>>>>>     on Wed, 15 Dec 2021 20:09:46 -0800 writes:

    > Just following up to check if anyone has had time to look over these patches.
    > Frederick

I strongly guess that nobody has.

Let me give you my perception of what you have tried to
propose/use,  and why I hadn't thought I should put in time for it:

You had started the thread by proposing "to override stopifnot()",
something which I (even though principal author of the function)
don't find a good idea at all:

stopifnot() is just one important utility function that will
call stop() under some circumstances.
If you want to tweak  error handling / debugging / browser, ..
you need to work on the level of error conditions, their
handlers, etc. 

Secondly, you've mixed this up with mouse button
action/interrupt/.. handling  which may be a cool and nice idea,
but then your  `xbindkey`-etc code is, I think, only/entirely
for X11-based R interfaces, and I think this would only be a
Linux console, possibly one from using ESS (Emacs Speaks Statistics),
but most probably (but I'm guessing here) not even relevant when
using Rstudio on Linux, and even less relevant for any of the
other ways R is used interactively on non-Linux platforms. Maybe
it would also apply to *some* uses of R on the Mac, but not even
the default R-Mac GUI..

Sorry that this not as much encouraging as it probably should
be, but I though you'd rather want *some* feedback than none...

Best,
Martin



    > On Wed, Dec 08, 2021 at 12:24:47AM -0800, Frederick Eaton wrote:
    >> Dear R Core Team,
    >> 
    >> I'm attaching a proposed patch to hopefully address my confusions regarding the documentation of browser(). I'm not sure if all the material I added is correct, but I made experiments to confirm that the behavior is at least roughly as described.
    >> 
    >> patch ./src/library/base/man/browser.Rd < browser.patch
    >> 
    >> Also, here is a patch to support multiple mouse buttons in getGraphicsEvent(). This must be edited before it can be applied, I decided to keep the old code in an 'if(0)' to help make it clearer that my code is essentially doing the same thing.
    >> 
    >> https://github.com/navarum/tweaks/blob/master/r/patches/0001-Add-support-for-multiple-mouse-buttons.patch
    >> wget -O - https://raw.githubusercontent.com/navarum/tweaks/master/r/patches/0001-Add-support-for-multiple-mouse-buttons.patch | patch -p1
    >> 
    >> It would be useful to have support in R for more than three mouse buttons because this enables the use of the mouse wheel (buttons 4 and 5), which can provide a more convenient interface when adjusting numbers and graphics and so on. I also have shift+wheel bound to buttons 6 and 7 via xbindkeys and xte, which I use for horizontal scrolling, via a trick from the web somewhere:
    >> 
    >> $ cat .xbindkeysrc.scm | grep xte
    >> (xbindkey '(shift "b:4") "xte 'mouseclick 6'")
    >> (xbindkey '(shift "b:5") "xte 'mouseclick 7'")
    >> 
    >> I hope that these contributions can be found acceptable.
    >> 
    >> Thank you in advance,
    >> 
    >> Frederick
    >> 
    >> 
    >> 
    >> On Mon, Nov 22, 2021 at 09:13:58AM -0800, Frederick Eaton wrote:
    >>> Dear R Devel,
    >>> 
    >>> I have been advised to use "options(error=recover)" to enable
    >>> debugging on errors. But sometimes it would seem more convenient to
    >>> override "stopifnot", for example:
    >>> 
    >>> stopifnot = function(b) { if(!b) { browser(skipCalls=1); } }
    >>> 
    >>> However, this doesn't do what I expected. On looking closer I find
    >>> that the "skipCalls" argument seems to be ignored except when printing
    >>> the "Called from: " message; it does not affect the evaluation context
    >>> or the output of 'where':
    >>> 
    >>> > var=2; f=function(){var=1; browser(skipCalls=0)}; f()
    >>> Called from: f()
    >>> Browse[1]> var
    >>> [1] 1
    >>> Browse[1]> where
    >>> where 1: f()
    >>> 
    >>> Browse[1]> Q
    >>> > var=2; f=function(){var=1; browser(skipCalls=1)}; f()
    >>> Called from: top level     Browse[1]> var
    >>> [1] 1
    >>> Browse[1]> where
    >>> where 1: f()
    >>> 
    >>> Browse[1]> Q
    >>> > var=2; f=function(){var=1; browser(skipCalls=2)}; f()
    >>> Called from: top level     Browse[1]> var
    >>> [1] 1
    >>> Browse[1]> where
    >>> where 1: f()
    >>> 
    >>> Browse[1]> Q
    >>> 
    >>> So it appears that the "browser()" API does not actually make it
    >>> possible to call this built-in function from within another R function
    >>> and thereby emulate the same behavior as calling browser() directly.
    >>> 
    >>> If this is the case, it might be good to have it fixed or documented.
    >>> I am aware of "browser(expr=)", but this requires editing the
    >>> particular call that failed. The documentation for "browser()" led me
    >>> to hope that my use case would be supported, if only because it admits
    >>> that users might want to build other debugging functions with
    >>> browser(): "The 'skipCalls' argument should be used when the
    >>> 'browser()' call is nested within another debugging function". An
    >>> example where this 'skipCalls' parameter is used to build a useful
    >>> debugging function would help to clarify its English description in
    >>> the manual.
    >>> 
    >>> Also, from the browser() command line I could not find a way to step
    >>> *out* of the current function. This would have been a way to recover
    >>> from skipCalls not working as expected. Am I missing something? For
    >>> example is there some command other than "n", where the below
    >>> interaction could pause before "hi" and "bye"?
    >>> 
    >>> > f=function(){browser(); message("in f"); message("out f")}; f(); message("hi"); message("bye")
    >>> Called from: f()
    >>> Browse[1]> n
    >>> debug at #1: message("in f")
    >>> Browse[2]> n
    >>> in f
    >>> debug at #1: message("out f")
    >>> Browse[2]> n
    >>> out f
    >>> hi
    >>> bye
    >>> 
    >>> If it is not possible for the R debugger to step out of a function, it
    >>> would be good to document that too, maybe after the list of browser
    >>> prompt commands in "?browser". Being confined within a single function
    >>> is not an obvious disability for a debugger to have.
    >>> 
    >>> I feel that R is an excellent tool, but sometimes I think that if the
    >>> shortcomings of the system were better documented, then this would
    >>> save users a lot of time in certain cases.
    >>> 
    >>> Thank you,
    >>> 
    >>> Frederick
    >>> 

    >> --- browser-orig.Rd	2021-12-07 22:35:51.991222137 -0800
    >> +++ browser-new.Rd	2021-12-07 23:53:36.414106296 -0800
    >> @@ -45,8 +45,16 @@
    >> will be simpler.
    >> 
    >> The \code{skipCalls} argument should be used when the \code{browser()}
    >> -  call is nested within another debugging function:  it will look further
    >> -  up the call stack to report its location.
    >> +  call is nested within another debugging function: it will look further
    >> +  up the call stack to report its location. However, currently this
    >> +  argument only affects the \code{Called from:} message printed before
    >> +  the first \code{Browse>} prompt, and does not change the evaluation
    >> +  environment of the expressions entered there, nor the output of the
    >> +  \code{where} command. It is not clear when this argument would be
    >> +  useful. It is for example currently impossible to change occurrences
    >> +  of \code{stopifnot(x)} to act like \code{browser(expr=!x)} by
    >> +  redefining \code{stopifnot()}; this must instead be done through a
    >> +  textual search-and-replace or \code{options(error=recover)}.
    >> 
    >> At the browser prompt the user can enter commands or \R expressions,
    >> followed by a newline.  The commands are
    >> @@ -58,11 +66,9 @@
    >> \item{\code{f}}{finish execution of the current loop or function}
    >> \item{\code{help}}{print this list of commands}
    >> \item{\code{n}}{evaluate the next statement, stepping over
    >> -      function calls.  For byte compiled functions interrupted by
    >> -      \code{browser} calls, \code{n} is equivalent to \code{c}.}
    >> +      function calls.}
    >> \item{\code{s}}{evaluate the next statement, stepping into
    >> -      function calls.  Again, byte compiled functions make
    >> -      \code{s} equivalent to \code{c}.}
    >> +      function calls.}
    >> \item{\code{where}}{print a stack trace of all active function calls.}
    >> \item{\code{r}}{invoke a \code{"resume"} restart if one is
    >> available; interpreted as an \R expression otherwise. Typically
    >> @@ -71,6 +77,15 @@
    >> \item{\code{Q}}{exit the browser and the current evaluation and
    >> return to the top-level prompt.}
    >> }
    >> +
    >> +  Both \code{n} and \code{s} also step out of function calls. However,
    >> +  it is currently not possible with these commands to step out of the
    >> +  function which called \code{browser()}. At the end of the function
    >> +  which called \code{browser}, \code{n} and \code{s} are both equivalent
    >> +  to \code{c}. These commands (\code{n} and \code{s}) are also
    >> +  equivalent to \code{c} when trying to use them inside byte compiled
    >> +  functions interrupted by \code{browser} calls.
    >> +
    >> Leading and trailing whitespace is ignored, except for an empty line.
    >> Handling of empty lines depends on the \code{"browserNLdisabled"}
    >> \link[=options]{option}; if it is \code{TRUE}, empty lines are ignored.



More information about the R-devel mailing list