[R-SIG-Mac] Tcl/Tk issues in arm64 build of R 4.1.0

[Prof Brian Ripley]

On 09/09/2021 21:34, John Fox wrote:
> Dear Brian,
> 
> Thank you for responding so quickly. Please see below:
> On 2021-09-09 2:59 p.m., Prof Brian Ripley wrote:
>> On 09/09/2021 19:03, John Fox wrote:
>>> Dear Simon, Philippe, and list members,
>>>
>>> I've encountered some Tcl/Tk-related issues with the Apple silicon 
>>> arm64 build of R 4.1.0 for macOS. These issues affect the Rcmdr 
>>> package, although they don't prevent it from working:
>>>
>>> (1) Some fonts that are available for the Intel build appear to be 
>>> missing from the arm64 build. Compare the screen shots of the Rcmdr 
>>> main window at 
>>> <https://socialsciences.mcmaster.ca/jfox/.Pickup/x86_64-apple-darwin17.0.png> 
>>> (Intel build) and 
>>> <https://socialsciences.mcmaster.ca/jfox/.Pickup/aarch64-apple-darwin20.png> 
>>> (arm64 build). This problem is purely aesthetic.
>>
>> It would be helpful to know what those fonts are.  This is likely to 
>> be a fontconfig issue and needs to be debugged by name.
> 
> I query the Tk default and fixed-width fonts via
> 
>      tclvalue(.Tcl("font actual TkDefaultFont -family")
>      tclvalue(.Tcl("font actual TkFixedFont -family"))
> 
> and use those -- in the past, that produced pleasing font choices on all 
> platforms. In the macOS Intel build, the resulting font families are 
> "Verdana" and "Andale Mono"; in the arm64 build, "helvetica" and 
> "courier". Perhaps the Intel and arm64 builds were just configured 
> differently.

Yes, I think fontconfig was, and that may help Simon dig into this.

>>
>>> (2) The Tcl/Tk Tktable package is apparently absent from the arm64 
>>> build but still present in the Intel build. The Rcmdr detects its 
>>> absence and suppresses some features (i.e., those requiring the Rcmdr 
>>> data editor). I understand that the Tktable package is problematic 
>>> and it may have been removed intentionally.
>>
>> https://sourceforge.net/projects/tktable/files/tktable/2.10/Tktable2.10.tar.gz 
>> (apparently the latest version from 2008)
>> does not compile against recent Tcl/Tk (8.6.11 seems to be used): it 
>> uses 'panic' which should be 'Tcl_Panic'.    (The Intel build appears 
>> to be much older, 8.6.6.)
>>
>> I have a patched version: AFAICS it installs entirely into 
>> /opt/R/arm64/lib I could tar it up and make it available.
>>
>> I could also send the patch to Simon for future use, but note that 
>> https://mac.r-project.org/libs-arm64/ contains an aqua build of Tcl/Tk 
>> (not the X11 build in the CRAN build of R 4.1.[01]) and which does not 
>> support Tcl/Tk.
> 
> I'm sorry, but I don't quite understand that -- perhaps there's a typo 
> here? Do you mean to say that an aqua build of Tcl/Tk will replace the 
> X11 build after R 4.1.1, and that Tktable will be incompatible with it?

There was, so let me expand.

There are two Tk implementations on macOS,

- using aqua, which is included in https://mac.r-project.org/libs-arm64/
- using X11, included in the CRAN distributions of R 4.1.[01] on amd64, 
and always used for Intel. So I do not have access to the details of 
that build.

As I recall, TkTable does not install for the aqua implementation, but 
does (when patched) for the X11 implementation.  My recollection is that 
XQuartz became available for amd64 not long before R 4.0.0 and Simon 
switched to X11 Tk, although pre-releases had aqua Tk.

> If so, first, I think that an aqua Tcl/Tk would be preferable to the 
> current X11 version. If I understand correctly, for example, aqua would 
> put Tk menus where Mac users expect them.

aqua Tk does not work correctly with R.app, and is not supported by 
package tkrplot (which lots of other packages depend on).  Personally I 
usually use aqua Tk in my own builds, but then I only use R.app if 
needed to debug something.

> Second, given the problems that Tktable seems to be producing, wouldn't 
> it be better to make Tablelist available as a substitute? I'm happy to 
> rewrite the Rcmdr data editor using Tablelist (actually, it's mostly 
> done), and, as I said, if I knew how to incorporate the Tcl/Tk Tablelist 
> package in the Rcmdr package (if it's undesirable simply to make it 
> available generally), I'd do it.

That's your choice.  The sources are at 
https://www.nemethi.de/tablelist/tablelist6.15.tar.gz and contain a 
README.txt with a 'How to Install It?'.  At least on a Unix-alike you 
just need to unpack that into somewhere on the Tcl path.  So to 
distribute with an R package, have the  tablelist6.15 dir in inst/ and 
arrange to add the installed location to the Tcl search path with 
tcltk::addTclPath() .  (I don't see why Windows would be different.)


-- 
Brian D. Ripley,                  ripley using stats.ox.ac.uk
Emeritus Professor of Applied Statistics, University of Oxford