R SIG Debian - Jan 2022 - Segfault from c2d4u binary of RandomFields

Hi everyone,

For some time the GitHub Action of spatstat.core has been failing due
to a segfault when calling RandomFields (which calls RandomFieldsUtils
which may be causing the acutal segfault):

https://github.com/spatstat/spatstat.core/runs/4739596930?check_suite_focus=true#step:6:387

I tried changing the action to install a source version of RandomField
and RandomFieldsUtils over the binary c2d4u version and the action now
continues smoothly without any segfault:

https://github.com/spatstat/spatstat.core/runs/4749470559?check_suite_focus=true

Does anyone know why the segfault only happens with the c2d4u version
(Michael and Dirk are probably good guesses, but maybe someone else is
listening out there and can take some workload off their shoulders...)

I'm happy to investigate further and produce a more minimal
reproducible example, but I will not go down the rabbit hole for now if
someone already now knows what the problem is and how it should be
fixed.

A (not so minimal) reproducible example is to install both
RandomFields, RandomFieldsUtils and spatstat.core from c2d4u and then
run the following in R (this segfaults, so consider yourself warned):

library(spatstat.core)
example(kppm)

Cheers,
Ege

-- 
Ege Rubak, Associate Professor,
Department of Mathematical Sciences, Aalborg University
Skjernvej 4A, 9220 Aalborg East, Denmark
Phone: (+45)99408861
Mobile: (+45)30230252
Email: rubak at math.aau.dk

On 8 January 2022 at 20:40, Ege Rubak wrote:
| Hi everyone,
| 
| For some time the GitHub Action of spatstat.core has been failing due
| to a segfault when calling RandomFields (which calls RandomFieldsUtils
| which may be causing the acutal segfault):
| 
|
https://github.com/spatstat/spatstat.core/runs/4739596930?check_suite_focus=true#step:6:387
| 
| I tried changing the action to install a source version of RandomField
| and RandomFieldsUtils over the binary c2d4u version and the action now
| continues smoothly without any segfault:
| 
|
https://github.com/spatstat/spatstat.core/runs/4749470559?check_suite_focus=true
| 
| Does anyone know why the segfault only happens with the c2d4u version
| (Michael and Dirk are probably good guesses, but maybe someone else is
| listening out there and can take some workload off their shoulders...)
| 
| I'm happy to investigate further and produce a more minimal
| reproducible example, but I will not go down the rabbit hole for now if
| someone already now knows what the problem is and how it should be
| fixed.
| 
| A (not so minimal) reproducible example is to install both
| RandomFields, RandomFieldsUtils and spatstat.core from c2d4u and then
| run the following in R (this segfaults, so consider yourself warned):
| 
| library(spatstat.core)
| example(kppm)

The geo stack is complicated. I have no pointers.  But on the Debian testing
machine (i.e. something "close" your Ubuntu setup) I use for Rcpp
reverse
depends, I recently recompiled an entire stack "in bulk and anger"
simply
because I could not pin-point where such a segfault came from.

What you could is try to mimick the Actions setup (by starting with a Ubuntu
20.04 container, Rocker has some as eg rocker/r-ubuntu:20.04 giving you a
head).  You can then start to replace .deb by .deb by hand-compiling and see
which one is 'guilty'. Michael could then rebuild that .deb.

Sadly I do not know of any other shortcuts.

Best, Dirk

-- 
https://dirk.eddelbuettel.com | @eddelbuettel | edd at debian.org