[R-pkg-team] Bug#972074: [Help] r-cran-bayestestr: autopkgtest regression

[Andreas Tille]

Control: tags 972074 pending
Control: block 972074 by 963392

On Thu, Nov 19, 2020 at 07:53:06AM -0600, Dirk Eddelbuettel wrote:
> | 
> | I do not have the slightest idea what this might mean.
> 
> ABI/API slippage in the stack. An interface changed but a package didn't recompile.

It seems it is caused by r-cran-rstanarm which in turn has a test
suite issue itself:

...
g++ -std=gnu++14 -shared -L/usr/lib/R/lib -Wl,-z,relro -o sourceCpp_2.so file8f23649f86.o /usr/lib/R/site-library/rstan/lib/libStanServices.a -L/usr/lib/R/site-library/StanHeaders// -lStanHeaders -L/usr/lib/R/site-library/RcppParallel// -ltbb -L/usr/lib/R/lib -lR
/usr/bin/ld: cannot find -lStanHeaders
collect2: error: ld returned 1 exit status
make: *** [/usr/share/R/share/make/shlib.mk:10: sourceCpp_2.so] Error 1
── ERROR (test_stan_functions.R:70:1): (code run outside of `test_that()`) ─────
Error: $ operator is invalid for atomic vectors
Backtrace:
    █
 1. └─rstan::expose_stan_functions(stanc_ret, rebuild = TRUE, verbose = TRUE) test_stan_functions.R:70:0

══ testthat results  ═══════════════════════════════════════════════════════════
ERROR (test_stan_functions.R:70:1): (code run outside of `test_that()`)

[ FAIL 1 | WARN 0 | SKIP 0 | PASS 1 ]
Error: Test failures
Execution halted


This is actually the reason why r-cran-rstanarm 2.21.1 is in Git for
several weeks but not uploaded yet.
 
> | If nobody has any clue we should probably ask upstream about this.
> 
> That risks making you look foolish as upstream is (once again) 100% clean at
> CRAN which is (to a first approximation) the only measure a CRAN author cares
> about.  Copied and pasted from the results page [1] and lightly edited for
> column alignment:

We do not test the code provided by somebody else but what we package
for Debian.  You know we are sometimes replacing code copies you are
yourself maintaining r-cran-bh which is something else than what is
running on CRAN.  Thus I think its a sensible approach to test what we
package.
 
> CRAN Package Check Results for Package bayestestR
> Last updated on 2020-11-19 13:50:51 CET.
> 
> Flavor	Version	Tinstall		Tcheck	Ttotal	Status	Flags
> r-devel-linux-x86_64-debian-clang	0.7.5	12.67	406.54	419.21	OK	
> r-devel-linux-x86_64-debian-gcc		0.7.5	8.37	295.67	304.04	OK	
> r-devel-linux-x86_64-fedora-clang	0.7.5			498.86	OK	
> r-devel-linux-x86_64-fedora-gcc		0.7.5			487.49	OK	
> r-devel-windows-ix86+x86_64		0.7.5	16.00	636.00	652.00	OK	
> r-patched-linux-x86_64			0.7.5	12.72	388.61	401.33	OK	
> r-patched-solaris-x86			0.7.5			725.40	OK	
> r-release-linux-x86_64			0.7.5	12.04	390.98	403.02	OK	
> r-release-macos-x86_64			0.7.5				OK	
> r-release-windows-ix86+x86_64		0.7.5	13.00	652.00	665.00	OK	
> r-oldrel-macos-x86_64			0.7.5				OK	
> r-oldrel-windows-ix86+x86_64		0.7.5	11.00	500.00	511.00	OK

Not sure what information this table should provide to us.  The fact
that things are running on those other systems is great but does not
answer why it is not running for us. 
 
> 'OK' on every platform is as clean as it gets. It is an ambitious package
> with a lot of Suggests. If you insist on loading all Suggests (which are,
> after all, optional) then you simply have to make sure they are all
> current. CRAN does that.  If you want to autotest you need to as well.

In most cases the error log was verbose about what is missing or what
is not current.  My try with the latest r-cran-rstanarm (that fails its
own test) shows that this is probably the cause for the test issue.

Kind regards

     Andreas.
 
> [1] https://cran.r-project.org/web/checks/check_results_bayestestR.html

-- 
http://fam-tille.de