bypassing the R.app help browser?

-----Original Message-----
From: r-sig-mac-bounces at r-project.org [mailto:r-sig-mac-bounces at r-
project.org] On Behalf Of John Fox
Sent: Wednesday, August 13, 2014 12:57 PM
To: 'Prof Brian Ripley'; 'peter dalgaard'
Cc: 'R-SIG-Mac'
Subject: Re: [R-SIG-Mac] bypassing the R.app help browser?

Dear Brian and Peter,

Thanks for picking up this issue.

The behaviour that Brian reports is exactly what I observed, and the
Tcl/Tk
doc that he quotes is what I consulted. It's not surprising to me that
the R
process waits until the Tk window calling tkwait.window() is destroyed.
I
suppose that because the internal help browser runs under the R
process, it
too waits, while an external browser -- as is spawned by help.start() -
-
runs in an independent process.

As I mentioned, I've removed the call to tkwait.window() in the Rcmdr
sources (it's in a "macro" called by every Rcmdr modal dialog) and will
test
whether there are negative consequences. I've observed none so far.

BTW, the same issue arises when the Rcmdr is run inside of RStudio,
which
directs help to its own browser.

Best,
 John

-----Original Message-----
From: Prof Brian Ripley [mailto:ripley at stats.ox.ac.uk]
Sent: Wednesday, August 13, 2014 11:08 AM
To: peter dalgaard; John Fox
Cc: R-SIG-Mac
Subject: Re: [R-SIG-Mac] bypassing the R.app help browser?

On 13/08/2014 15:11, peter dalgaard wrote:

This isn't unique to tcltk. Anything that blocks the keyboard loop

blocks the help browser too. Try e.g. opening the help for ls, type
Sys.sleep(15) and watch the beach ball in the help browser as you try
to scroll in it.

But Sys.sleep should not be blocking an event loop: from its help

      The intention is that this function suspends execution of R
      expressions but wakes the process up often enough to respond to
      GUI events, typically every 0.5 seconds.

The mechanisms to mesh event loops which are in place for Sys.sleep
are
R_CheckUserInterrupt (which calls R_ProcessEvents) and R_runHandlers.

Note that the help for tkwait says (on my box)

        While  the  tkwait command is waiting it processes events in
the
normal
        fashion, so the application will continue to respond to  user
interac-
        tions.   If  an  event handler invokes tkwait again, the

nested

call to
        tkwait must complete before the outer call can complete.

but as this is X11 Tk, it means X11/Unix events.  You can demonstrate
that, as e.g the http server still works (use help.start() first).

-pd


On 13 Aug 2014, at 15:14 , John Fox <jfox at mcmaster.ca> wrote:

Dear Simon,

Here's a simple script that will demonstrate the problem:

----- snip -----

library(tcltk)

top <- tktoplevel()
button <- ttkbutton(top, text="help", command=function()

print(help(lm)))

tkgrid(button)
tkwait.window(top)

----- snip -----

The problem is produced by tkwait.window(), and this call is in

all

Rcmdr modal dialogs. As I read the Tcl/Tk docs, it shouldn't cause
problems, but obviously it's causing this problem.  I'm also not
certain whether calling tkwait.windows() is necessary and will look
into the consequences of removing it -- I believe that it's been

there

for many years, from the earliest versions of the Rcmdr.

With respect to changing using preferences, this is done only

until

the Commander() exits. If getting rid of the call to tkwait.window()
proves problematic, I can ask the user for permission in a pop-up
dialog.

Thanks for your help,
John

On Wed, 13 Aug 2014 00:25:30 -0400
Simon Urbanek <simon.urbanek at r-project.org> wrote:

John,

can't you address the underlying issue instead and don't block

the

event loop? A lot of things don't work if the event loop is blocked

and

I would argue that changing user's preferences behind the scenes is a
violation of the CRAN policies.

I'm happy to help if you send me a bit more details.

Cheers,
Simon


On Aug 12, 2014, at 6:15 PM, John Fox <jfox at mcmaster.ca> wrote:

Hi Marc,

-----Original Message-----
From: Marc Schwartz [mailto:marc_schwartz at me.com]
Sent: Tuesday, August 12, 2014 5:10 PM
To: John Fox
Cc: r-sig-mac at r-project.org
Subject: Re: [R-SIG-Mac] bypassing the R.app help browser?

Hi John,

Happy to help. I recalled seeing something previously on this,

so

a

search using rseek.org was fruitful.

The potential gotcha, of course, is if for some reason the GUI

exits in

a manner possibly not under your control. The setting would not

be

returned to the default and the therefore, as you note,

retained

for a

subsequent session where the behavior may not be desired.

Yes, there is that possibility.

If this is for Rcmdr, perhaps this is something that could be

added to

a menu, where the user can alter the behavior in either

direction

as

desired, if that makes sense.

As you guessed, this is for the Rcmdr, where the built-in R.app

browser

doesn't play well with dialog help buttons -- the browser is

unresponsive

until the dialog that called it closes -- while an external

html-

help

browser works fine.

I've now successfully implemented the approach I outlined, in

which the

previous setting is restored when the Commander GUI closes. As

you

point

out, this isn't bullet-proof, but I think it is the best I can

do

for now.

Allowing the user to apply the change would be safer, but users

are unlikely

to discover the option.

Simon would need to comment on the potential for alternatives.

Yes, that would be welcome. I still think that a setting via

options() would

be better.

Thanks again for your help,
John

Best regards,

Marc


On Aug 12, 2014, at 3:46 PM, John Fox <jfox at mcmaster.ca> wrote:

Hi Marc,

Thanks for this. It does work, and I wasn't aware of it --

you've

obviously

done a better job than I did of searching for a solution!

Although this approach works, it has the disadvantage of

permanently

changing the help browser in R.app, beyond the current

session,

at

least

until the change is explicitly undone. I think that I can work

around

that

by something like

	current <- system("defaults read org.R-project.R",

intern=TRUE)

to discover whether the use.external.help preference exists,

and

if

so, what

its value is; to then set the preference to YES if it's NO or

unset;

and

finally to remove the preference on exit.

Again, thanks -- I think that I can work with this, though it

would

in my

opinion be better if the help browser were settable for the

current

session

directly via options() in R, or if one could specify the

browser

in a

call

to help().

Best (and thanks again),
John

-----Original Message-----
From: Marc Schwartz [mailto:marc_schwartz at me.com]
Sent: Tuesday, August 12, 2014 4:04 PM
To: John Fox
Cc: r-sig-mac at r-project.org
Subject: Re: [R-SIG-Mac] bypassing the R.app help browser?

On Aug 12, 2014, at 2:33 PM, John Fox <jfox at mcmaster.ca>

wrote:

Dear list members,

Is there a way to bypass the R.app help browser, and to use

instead

an alternative browser, such as the one pointed to by
getOption("browser")?

I've tried a number of strategies, including setting

.Platform$GUI

<-

"unknown". The only approach I tried that works is to mask

the

help()

function with a modified version, but this produces other

problems,

such as referencing unexported objects from utils and tools.

It would be nice if the help() function had a browser

argument,

similar to that in browseURL(), and defaulting to the current
behaviour.

Any suggestions would be appreciated.

John

John,

I found this post from Simon that seems to work:

https://stat.ethz.ch/pipermail/r-sig-mac/2009-

December/006908.html

I tried it on my Mac in the latest version of R.app, which I

normally

do not use and the help system does now popup a browser.


Regards,

Marc Schwartz

--
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Emeritus Professor of Applied Statistics, University of Oxford
1 South Parks Road, Oxford OX1 3TG, UK