R-devel segmentation fault in combination RStudio/R GUI

On 26 Sep 2024, at 11:42, Simon Urbanek <simon.urbanek at R-project.org> wrote:

Sebastian,

thanks, I can replicate the crash in the R GUI and it?s due to a missing symbol detected at run-time (I?m attaching the traceback even though it?s not very helpful). Unfortunately, the error doesn?t say which symbol and from which library - and the use is inside Apple?s core library, not R itself - so I have to dig deeper to see what triggers it. This could well be some bug in R-devel, because the build of R-4.4-branch from the same nightly run works fine. I?ll try to have a more in-depth look next week.

Cheers,
Simon


* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGABRT
   frame #0: 0x00000001972434c8 dyld`__abort_with_payload + 8
dyld`:
->  0x1972434c8 <+8>:  b.lo   0x1972434e8               ; <+40>
   0x1972434cc <+12>: pacibsp 
   0x1972434d0 <+16>: stp    x29, x30, [sp, #-0x10]!
   0x1972434d4 <+20>: mov    x29, sp
Target 0: (R) stopped.
(lldb) bt
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGABRT
 * frame #0: 0x00000001972434c8 dyld`__abort_with_payload + 8
   frame #1: 0x000000019724e0cc dyld`abort_with_payload_wrapper_internal + 104
   frame #2: 0x000000019724e100 dyld`abort_with_payload + 16
   frame #3: 0x00000001971df7f0 dyld`dyld4::halt(char const*, dyld4::StructuredError const*) + 304
   frame #4: 0x00000001972144fc dyld`dyld4::APIs::_dyld_missing_symbol_abort() + 28
   frame #5: 0x000000019879574c Foundation`__NSFireDelayedPerform + 372
   frame #6: 0x00000001976615b8 CoreFoundation`__CFRUNLOOP_IS_CALLING_OUT_TO_A_TIMER_CALLBACK_FUNCTION__ + 32
   frame #7: 0x000000019766125c CoreFoundation`__CFRunLoopDoTimer + 972
   frame #8: 0x0000000197660d94 CoreFoundation`__CFRunLoopDoTimers + 356
   frame #9: 0x00000001976441cc CoreFoundation`__CFRunLoopRun + 1856
   frame #10: 0x0000000197643434 CoreFoundation`CFRunLoopRunSpecific + 608
   frame #11: 0x00000001a1ded19c HIToolbox`RunCurrentEventLoopInMode + 292
   frame #12: 0x00000001a1decfd8 HIToolbox`ReceiveNextEventCommon + 648
   frame #13: 0x00000001a1decd30 HIToolbox`_BlockUntilNextEventMatchingListInModeWithFilter + 76
   frame #14: 0x000000019aea2cc8 AppKit`_DPSNextEvent + 660
   frame #15: 0x000000019b6994d0 AppKit`-[NSApplication(NSEventRouting) _nextEventMatchingEventMask:untilDate:inMode:dequeue:] + 700
   frame #16: 0x0000000100009628 R`-[RController doProcessEvents:] + 160
   frame #17: 0x0000000100005260 R`-[RController handleReadConsole:] + 80
   frame #18: 0x000000010000c0d8 R`Re_ReadConsole + 192
   frame #19: 0x0000000100b76328 libR.dylib`R_ReplDLLdo1 at main.c:375:6 [opt]
   frame #20: 0x0000000100016ea0 R`run_REngineRmainloop + 260
   frame #21: 0x000000010000e54c R`-[REngine runREPL] + 124
   frame #22: 0x0000000100001c2c R`main + 592
   frame #23: 0x00000001971db154 dyld`start + 2476

On 26 Sep 2024, at 06:12, Sebastian Kreutzer <sebastian.kreutzer at uni-heidelberg.de> wrote:

Hello, 

Thanks again to Prof Ripley and Simon; however, I am afraid I have to get back 
to you on this because my ?I am good for now" was premature.

The R-devel builds now appear again on https://mac.r-project.org/; thank you very much!

Unfortunately, I have the same issue with these builds as with the ones I have produced locally.

- I installed and tried the binary builds from https://mac.r-project.org/big-sur/last-success/ and 
https://mac.r-project.org/ but when I use RStudio (just updated before once more) and try to load
a package, the session crashes immediately. The same goes for the R GUI; it won?t start anymore.

- As before, calling R from the terminal works fine without any issue

- When I return to R-4.4.1-arm64.pkg (always freshly installed, no version switching), 
everything works as expected, with no error or issue. 

- I?ve looked up the RStudio GitHub issue list but found no report

- The console shows the following errors (shorted; I can provide full reports if wanted): 

RStudio:  EXC_CRASH (SIGABRT)
R GUI: codes":"0x0000000000000000, 0x0000000000000000","rawCodes":[0,0],"type":"EXC_CRASH","signal":?SIGABRT"

- To be absolutely sure, I then tried to install the R-devel build from https://mac.r-project.org/ on my private M1 Mac mini, still running macOS 14.7 and 
I got the same crashes. Other than being from Apple, these two machines (the M2 and the private M1) have nothing in common 
regarding the setup. 

This observation makes me somewhat think that it is likely that someone else can perhaps reproduce this "issue"? 

Thanks once more for your support and kind regards, 

Sebastian 

P.S. Sorry for the HTML in my last messages. Prof Ripley was so kind as to point me to the e-mail certificate that caused these HTML tags. It should be fine now. 

On 25. Sep 2024, at 11:50, Prof Brian Ripley <ripley at stats.ox.ac.uk> wrote:

On 25/09/2024 10:26, Sebastian Kreutzer wrote:

Hi Simon,
Many thanks for your prompt reply and the link; it greatly helps, and I appreciate it!
Yes, sorry for not detailing my issue further, but I did not want to
spam anybody with the log and configuration files I am using, so I cut it short.
I will also look more carefully into the differences between ?my? build process and the CRAN builds;
it seems evident that I have overlooked something. My initial thought was just that with all the API
changes going on and me using a more recent version of macOS for the build than CRAN does,

The machine used for the M1mac additional issue is fully up-to-date, see https://www.stats.ox.ac.uk/pub/bdr/M1mac/README.txt

Following the instructions in the R-admin manual and not by some third party is always a good idea before posting.

Do please stop sending HTML, as required in the posting guide.

it would not surprise me that, at some point, something
had changed in macOS, causing this ?user-interface-only? crash.
Anyway, so as not to bother you further, thanks again for the help. I am good for now,
and if I figure out what has to be changed in my configuration to make it work again,  I will post it.
Kind regards,
Sebastian

On 25. Sep 2024, at 01:40, Simon Urbanek <simon.urbanek at R-project.org> wrote:

Sebastian,

if you want to replicate the CRAN builds, you have to also use the same settings, otherwise you may have a build which is not binary compatible. It is unclear from your description how you built R (there are several variants such as framework install vs "unix"-style install and they are incompatible). You can see the flags actually used at the top of ${R_HOME}/etc/Makevars.

As for macOS R binaries, all latest builds are always available from
https://mac.r-project.org/big-sur/last-success/

The fact that the main page itself is not showing R-devel is certainly not intentional since the binaries are there - I?ll look into that, thanks for reposting (you shouldn't wait months to report that ;)).

I did report it a month ago to Simon.

Thanks,
Simon

On 24 Sep 2024, at 02:27, Sebastian Kreutzer <sebastian.kreutzer at uni- heidelberg.de> wrote:

Hello,

I am writing because I have been struggling for a couple of months to get R-devel to work in combination with
RStudio or the R GUI.

In the past, I had been downloading R-devel for macOS from https:// mac.r-project.org/, which
nearly always worked, however, for some (months I have in mind), there aren't daily R-devel builds. So I started building
R from source following https://stackoverflow.com/questions/75595875/ how-do-i-build-r-from-sources-on-macos

Adapted to my system, this worked surprisingly well, so I kept drawing R-devel from the SNV server on a regular
basis and built it from the source. However, it stopped working in mid-August. In a nutshell:

- I can build R-devel from the source without any issue flagged, and when started in the terminal, it works as expected.
- However, it crashes reproducibly when trying to load a package in RStudio (stable/nightly build) and the R GUI (always the latest version)
terminates the R session on start. Error messages in the console I get read as follows:

R GUI: Termination Reason:  Namespace SIGNAL, Code 11 Segmentation fault: 11
RStudio: Exception Type: EXC_CRASH (SIGABRT) ->    /usr/lib/dyld 0x0 - 0xffffffffffffffff ??? (*) <00000000-0000-0000-0000-000000000000> ???

I can share the full logs, but I want to keep it short for now.

My questions are:

- Did anybody encounter such an issue with the latest R-devel and R GUI, RStudio?
- Is this perhaps why an R-devel binary is currently not available on https://mac.r-project.org/?

If I know that this is a known issue, it is all good; however, if it works
For all others without, then the error must be on my end, and I have to keep digging further.

- Tested systems: M2 -> macOS 14.5 to 15.0 with the Xcode on always the latest version available at the time.
- SNV: Always the latest check out


Kind regards,


Sebastian Kreutzer

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

[[alternative HTML version deleted]]