[R-sig-ME] errors building lme4 from source: Indexis not a member of 'lme4::lme4CholmodDecomposition...

[Boylan, Ross]

Further news on this problem below.

-----Original Message-----
From: Boylan, Ross 
Sent: Friday, June 28, 2024 8:03 PM
To: r-mixed <r-sig-mixed-models using r-project.org>
Subject: errors building lme4 from source: Indexis not a member of 'lme4::lme4CholmodDecomposition...

While trying to build lme4 from source with current master (SHA a7239d63) on MS-Windows I get the errors shown at the bottom of this message; the first one is
----------------------------------------------
lme4CholmodDecomposition.h: In instantiation of 'lme4::lme4CholmodDecomposition<_MatrixType, _UpLo>::solveInPlaceonst Eigen::MatrixBase<OtherDerived>&, int) const [with OtherDerived = Eigen::Matrix<double, -1, -1>; _MatrixType = Eigen::SparseMatrix<double, 0, int>; int _UpLo = 1]
predModule.cpp:127:37: required from here
lme4CholmodDecomposition.h:59:39:error: Indexis not a member of 'lme4::lme4CholmodDecomposition<Eigen::SparseMatrix<double, 0, int> >::Base{aka 'Eigen::CholmodDecomposition<Eigen::SparseMatrix<double, 0, int>, 1>
   59 |             eigen_assert((Base::Index)(factor()->n)ther.rows());
      |                          ~~~~~~~~~~~~~^~~~~~~~~~~~~
------------------------------------------------
which looks as if I may have some version mismatch, but I'm not sure exactly where the culprit is or what to do about it.  Any ideas?
This particular build was triggered while requesting a run of the test suite in RStudio, but I got roughly the same when doing a straight build in RStudio.
[snip]
-------------------------------------------

This seems like a repeat of the situation and error a few months ago (https://stat.ethz.ch/pipermail/r-sig-mixed-models/2024q2/030610.html), except that this time I kinda do need to build from source.  As before, installing from the tarball outside of RStudio works fine.  Installing from my development directory works fine.  One peculiarity is that 3 binaries built on the first run (from the tarball) were not built the second time around: `external.o`, `glmFamily.o`, and `optimizer.o`.

And after this (as before), building seems fine inside of RStudio.

I do not understand what is causing this behavior, and don't even have a plausible theory.

BTW with LME4_TEST_LEVEL=2 there are enough test failures (in testthat) to produce:
Error ('test-methods.R:650:3'): (code run outside of `test_that()`)
<testthat_abort_reporter/rlang_error/error/condition>
Error in `stop_reporter(c("Maximum number of failures exceeded; quitting at end of file.",
i Increase this number with (e.g.) `testthat::set_max_fails(Inf)`

Many, but not all, relate to uses of deprecated or non-existent alternatives to re.form in simulate().