[R-SIG-Mac] tcltk causes help() to hang in R 4.2.0 on macOS

Fri Jun 10 03:12:41 CEST 2022

Dear Simon,

On 2022-06-09 8:19 p.m., Simon Urbanek wrote:
> John,
> 
> as Tomas pointed out the problem was there, it was just not as easily triggered, because the help before 4.2.0 was not relying on a set of additional files (here KaTeX). Rcmdr does use tcltk, obviously, so it will always trigger the loop meshing issues.

Thanks for the additional explanation.

For what it's worth, I've tried the Rcmdr under R 4.2.0 patched and HTML 
help works fine (as expected).

> I wonder whether we should re-visit native Tcl/Tk now, since if I recall the main reason it didn't work before were deadlocks in the event loop, so it's even possible that those could be fixed along similar lines...

You, of course, will know much better than I what's possible. I believe 
that Brian Ripley told me some time ago that he was able to get R to 
work on macOS with native Tcl/Tk. That would be desirable for several 
reasons, including (I believe) integration of Tk menus with the Mac menu 
bar and eliminating the necessity for users to install XQuartz.

Best,
  John

> 
> Cheers,
> Simon
> 
> 
> 
>> On Jun 10, 2022, at 2:32 AM, John Fox <jfox using mcmaster.ca> wrote:
>>
>> Dear Tomas,
>>
>> On 2022-06-09 4:14 a.m., Tomas Kalibera wrote:
>>> On 6/8/22 18:39, John Fox wrote:
>>>>
>>>>
>>>> On 2022-06-08 12:36 p.m., John Fox wrote:
>>>>> Dear Thomas,
>>>>>
>>>>> Thanks for this additional information.
>>>>>
>>>>> I added some of my own defensive code to the development version of the Rcmdr package, to set the "help_type" option to "text" for macOS. I'll
>>>>
>>>> I should have added that help_type is set to "text" for R version 4.2.0 only!
>>> I think it is a coincidence that this was seen in R 4.2.0. The http server's handler has a safeguard against re-entrant execution since 59338 (year 2012). As that safeguard was needed at that time for clients connecting in rapid sequence, the problem with the live-lock with tcltk event loop could have been triggered already then, but probably nobody ran into this with tcltk package loaded. PR#17773 shows that attempts for re-entrant/concurrent handling of http requests were seen in R 4.0.0 with the HTML help.
>>
>> My experience of this issue is just with HTML help from the Rcmdr under macOS, but I've confirmed that the problem occurs in R 4.2.0 and not in R 4.1.3, possibly because the Rcmdr doesn't load tcltk (i.e., imports tcltk)?
>>
>> Best,
>> John
>>
>>> Best
>>> Tomas
>>>>
>>>>> modify that to all non-Windows platforms, and also include this in the CRAN version of the package.
>>>>>
>>>>> I currently test on macOS and Windows systems but when I have a chance will get a Linux system working as well.
>>>>>
>>>>> Best,
>>>>>    John
>>>>>
>>>>> On 2022-06-08 12:16 p.m., Tomas Kalibera wrote:
>>>>>> For reference, this problem is not macOS specific, it also exists on Linux (but not on Windows which use a different code path) and the core problem is how to prevent re-entrant execution of R event handlers when run from the R/tcltk event loop, itself running from an R event handler.
>>>>>>
>>>>>> R-patched now contains a defensive work-around from Simon.
>>>>>>
>>>>>> We have added a more efficient and cleaner fix to R-devel, but there is some risk of breakage: this form is "the documented way" but didn't work on some systems ~22 years ago. So it would be great if people using Tcl/Tk in R on Unix, and particularly those with rarely used window managers, could test their applications or the tcltk package demos and report any issues well before the next release.
>>>>>>
>>>>>> Thanks
>>>>>> Tomas
>>>>>>
>>>>>> On 5/30/22 01:27, John Fox wrote:
>>>>>>> Dear Simon,
>>>>>>>
>>>>>>> Thank you for addressing this problem so promptly.
>>>>>>>
>>>>>>> Best,
>>>>>>>   John
>>>>>>>
>>>>>>> On 2022-05-29 6:51 p.m., Simon Urbanek wrote:
>>>>>>>> Actually, I take back the reference to the commit - the deadlock is true, and removing the line is a work-around, but r78421 was not necessarily the regression. The issue seems to be caused by a combination of Tcl/Tk and KaTeX (the latter causes the browser to issue multiple parallel http-requests and the former causes the deadlock in its event loop) which has not been there in earlier R versions.
>>>>>>>>
>>>>>>>> Cheers,
>>>>>>>> Simon
>>>>>>>>
>>>>>>>>
>>>>>>>>> On 30/05/2022, at 10:09 AM, Simon Urbanek <simon.urbanek using r-project.org> wrote:
>>>>>>>>>
>>>>>>>>> John,
>>>>>>>>>
>>>>>>>>> thanks, this is a regression caused by r78421 (deadlock when TclTk's event loop is called in http processing). Removing L875 in src/modules/internet/Rhttpd.c restores the previous behavior, but the underlying problem is more complex and will require more investigation.
>>>>>>>>>
>>>>>>>>> Cheers,
>>>>>>>>> Simon
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>> On 29/05/2022, at 5:36 AM, John Fox <jfox using mcmaster.ca> wrote:
>>>>>>>>>>
>>>>>>>>>> Dear R-sig-mac list members,
>>>>>>>>>>
>>>>>>>>>> I've discovered that loading the tcltk package apparently causes R 4.2.0 (including the current patched version) to hang on an M1 Mac.
>>>>>>>>>>
>>>>>>>>>> Try, e.g.,
>>>>>>>>>>
>>>>>>>>>> library("tcltk")
>>>>>>>>>> help("lm")
>>>>>>>>>>
>>>>>>>>>> My session info:
>>>>>>>>>>
>>>>>>>>>> ------- snip -------
>>>>>>>>>>
>>>>>>>>>>> sessionInfo()
>>>>>>>>>> R version 4.2.0 Patched (2022-05-28 r82413)
>>>>>>>>>> Platform: aarch64-apple-darwin20 (64-bit)
>>>>>>>>>> Running under: macOS Monterey 12.3.1
>>>>>>>>>>
>>>>>>>>>> Matrix products: default
>>>>>>>>>> BLAS: /Library/Frameworks/R.framework/Versions/4.2-arm64/Resources/lib/libRblas.0.dylib
>>>>>>>>>> LAPACK: /Library/Frameworks/R.framework/Versions/4.2-arm64/Resources/lib/libRlapack.dylib
>>>>>>>>>>
>>>>>>>>>> locale:
>>>>>>>>>> [1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
>>>>>>>>>>
>>>>>>>>>> attached base packages:
>>>>>>>>>> [1] tcltk     stats     graphics  grDevices utils datasets  methods
>>>>>>>>>> [8] base
>>>>>>>>>>
>>>>>>>>>> loaded via a namespace (and not attached):
>>>>>>>>>> [1] compiler_4.2.0
>>>>>>>>>>
>>>>>>>>>> ------- snip -------
>>>>>>>>>>
>>>>>>>>>> Some additional details: The problem occurs both in the R macOS console and, if options(help_type="html"), when R is run in a terminal window on macOS, but not when options(help_type="text"). The former is the default in the Mac R console, the latter when R is run in a terminal.
>>>>>>>>>>
>>>>>>>>>> The problem is apparently new in R 4.2.0 -- it doesn't, e.g., occur in R 4.1.3. My apologies for not turning it up earlier.
>>>>>>>>>>
>>>>>>>>>> I discovered the problem when accessing help in the Rcmdr GUI, which uses tcltk, caused R to hang.
>>>>>>>>>>
>>>>>>>>>> Has anyone else encountered this problem?
>>>>>>>>>>
>>>>>>>>>> Best,
>>>>>>>>>> John
>>>>>>>>>> -- 
>>>>>>>>>> John Fox, Professor Emeritus
>>>>>>>>>> McMaster University
>>>>>>>>>> Hamilton, Ontario, Canada
>>>>>>>>>> web: https://socialsciences.mcmaster.ca/jfox/
>>>>>>>>>>
>>>>>>>>>> _______________________________________________
>>>>>>>>>> R-SIG-Mac mailing list
>>>>>>>>>> R-SIG-Mac using r-project.org
>>>>>>>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-mac
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> _______________________________________________
>>>>>>>>> R-SIG-Mac mailing list
>>>>>>>>> R-SIG-Mac using r-project.org
>>>>>>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-mac
>>>>>>>>>
>>>>>>>>
>> -- 
>> John Fox, Professor Emeritus
>> McMaster University
>> Hamilton, Ontario, Canada
>> web: https://socialsciences.mcmaster.ca/jfox/
>>
>> _______________________________________________
>> R-SIG-Mac mailing list
>> R-SIG-Mac using r-project.org
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mac
>>
> 
-- 
John Fox, Professor Emeritus
McMaster University
Hamilton, Ontario, Canada
web: https://socialsciences.mcmaster.ca/jfox/