v1.6.1 release candidate

[simonpcouch]

The CRAN check issue was on macos systems—that's now resolved.

Closes #902.

EDIT: FYI that I'll be OOO the 12th-14th. The submission deadline is Friday the 14th.

[simonpcouch]

Pro Driver failures existed with the odbc submission currently on CRAN without issue, so ignoring.

@hadley, #902 is up-to-date with the current status of the release process; should be ready to send out!

[simonpcouch]

The failure we see is:

  ── Failure ('test-utils.R:180:3'): locate_install_unixodbc() returns reasonable values ──
  file.exists(res[1]) is not TRUE
  
  `actual`:   FALSE
  `expected`: TRUE 

#899 did not resolve it, apparently (though note that it did resolve the same failure on mac-builder).

locate_install_unixodbc() either returns:

1. c(system_safely("odbc_config --lib-prefix"), system_safely("pkg-config --variable=libdir odbc")) if that vector is length > 0, or
2. Results from list.files(..., full.names = TRUE)

Call the results from either of the above res. In the failing test, file.exists(res[1]) is FALSE. Notably, it's not a subset error, so res is at least length 1. This makes me believe that locate_install_unixodbc() likely doesn't return from 2) on CRAN, as I can't think of an explanation for file.exists(list.files(..., full.names = TRUE)[1]) being FALSE other than some sort of tempfile local to list.files() or some path normalization wonk. So I'm inclined to believe that locate_install_unixodbc() returns from 1), and CRAN has set odbc_config cflags in such a way that they point to files that don't exist. Opting to just skip the test on CRAN.

[simonpcouch]

Was able to replicate the error_install_unixodbc() errors on an M1 Mac via removing the homebrew unixodbc install and its reverse dependencies. There are now quite a few new test skips here.

[detule]

Was able to replicate the error_install_unixodbc() errors on an M1 Mac via removing the homebrew unixodbc install and its reverse dependencies. There are now quite a few new test skips here.

Wait, we think there is no unixodbc installed at all on the CRAN machine, despite it being part of the recipes here: https://github.com/R-macos/recipes ?

[simonpcouch]

My hypothesis from #903 (comment) is that c(system_safely("odbc_config --lib-prefix"), system_safely("pkg-config --variable=libdir odbc")) is length > 0, though the files that one of odbc_config or pkg-config is pointing us to doesn't actually exist, making for a unixodbc installation that's not actually usable. Very much open to the idea that I'm wrong or that we ought to instead troubleshoot the recipe!

[detule]

Thanks Simon!

Ugh, i really wish there was a simpler mechanism to test releases against their build environment.

Apologies for asking, I should know, but how are you getting feedback that the tests error out in their build environment?
Provided there is a way to throw tar balls at it, and iterate on this, I can take a look this weekend. I am curious why/where those missing files are.

But generally don't think this should hold the release / think it's OK to skip these tests on CRAN.

[simonpcouch]

Very much agreed.

Apologies for asking, I should know, but how are you getting feedback that the tests error out in their build environment?

Nope, not a silly question at all! The mechanism here is that odbc is currently being flagged for manual review, and a CRAN team member emailed Hadley back with pasted-in testthat output from an M1 system ("M1mac", the one linked here) in response to our submission a couple days ago. This seems to be a different system than Mac-builder—from what I understand, Mac-builder is consistent with the recipes you've linked. I'm unfortunately not aware of a way to throw tarballs at the thing without manual intervention from CRAN folks.

[hadley]

I've submitted it again 🤞