Advise on building R on OSX without optimization for debugging

On Jul 13, 2020, at 7:59 PM, Prof Brian Ripley <ripley at stats.ox.ac.uk> wrote:

On 08/07/2020 21:38, Simon Urbanek wrote:

Dmitriy,
due to permissions and the various limitations on passing environment variables across processes it is often easier to simply run R and attach the debugger to it:
$ R
[...]

Sys.getpid()

[1] 89955

$ sudo lldb
Password:
(lldb) attach 89955
[...]
(lldb) c
Process 89955 resuming

On Catalina I can do that for a version of R I compiled, but not for a notarized binary distribution (which also refuses to be run under a debugger).  The lldb error message is maximally uninformative:

(lldb) attach 16682
error: attach failed: Error 1

or

(lldb) run
error: process exited with status -1 (Error 1)

I presume that is intentional on Apple's part and there is no way round it other than weakening security (e.g. disable SIP)?

Correct. It is annoying enough that it made to to the FAQ:

https://cran.r-project.org/bin/macosx/RMacOSX-FAQ.html#I-cannot-attach-debugger-to-R

So the recommended way is to use the non-notarised builds (you can also disable SIP but that is not recommended). And, yes, it would be nice if lldb provided a more useful error...

I suspect that it would be actually sufficient to provide just a exec/R binary that doesn't use hardened run-time, but it couldn't be distributed in the Apple Installer ... unless we create it with the post install script ... something to test I suppose...

Best,
Simon

My memory was that it did work on High Sierra (I sometimes use it to investigate packages which segfault under your distribution but not with my builds) and I have just re-checked there with the CRAN distribution of 4.0.2.

Brian Ripley

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