Implementing an effective build check for CRAN

Question

Implementing an effective build check for CRAN

martinamorris opened this issue 5 years ago · 11 comments

From Carter:

BTW, per the email Martina forwarded, it would be ideal if we had a way to duplicate the checks CRAN has started running using rchk - these are trying to ferret out PROTECT issues (and related things), but they seem to use a lot of heuristics that are both overzealous and that do not interact well with the backend code. Last I checked, there was no way for developers to duplicate CRAN's rchk checks; I tried building my own local version (which was a PITA), but even after doing that I was unable to replicate what CRAN was doing. It is possible that they've recently added functionality to assist with this, but if so that has been recent.

Answer 1 · 2019-10-16T19:30:02.000Z

Try the new rhub package.

rhub::check_for_cran()

Answer 2 · 2019-10-23T19:45:51.000Z

@chad-klumb, has there been any progress on this?

Answer 3 · 2019-10-23T19:56:47.000Z

No.

I was told we would not be pushing a new version of network to CRAN until pre-Sunbelt next year.

Has that now changed?

Answer 4 · 2019-10-23T20:27:12.000Z

I was told we would not be pushing a new version of network to CRAN until pre-Sunbelt next year. Has that now changed?

what is changing is the statnet server that we've been building on -- we've talked about this a couple of times. it's being retired at the end of December, so we need to develop a new build/test environment. @krivit let's put this on the agenda for this Thursday's call.

Answer 5 · 2019-10-23T20:39:45.000Z

My understanding of the op was that it related to rchk which is something we need to worry about for CRAN but typically not otherwise unless we're modifying the C backend of network (which I don't anticipate doing).

The question of where/how to run R CMD check is important (if we're losing the server) but not the same issue.

Answer 6 · 2019-10-23T20:46:12.000Z

There is this, which I currently know nothing about:

https://github.com/kalibera/rchk

@CarterButts Do you have any familiarity with this software/repo? I sounds like the sort of thing we're looking to do. Quoting front page:

This project consists of several bug-finding tools that look for memory protection errors in C source code using R API, that is in the source code of R itself and packages. The tools perform whole-program static analysis on LLVM bitcode and run on Linux. About 200-300 memory protection bugs have been found using rchk and fixed in R. rchk is now regularly used to check CRAN packages.

Answer 7 · 2019-10-23T20:48:07.000Z

My understanding of the op was that it related to rchk which is something we need to worry about for CRAN but typically not otherwise unless we're modifying the C backend of network (which I don't anticipate doing). The question of where/how to run R CMD check is important (if we're losing the server) but not the same issue.

as in many cases, a specific issue links to a more general issue, and i use that as a vehicle to start on the general issue. let's talk on Th.

Answer 8 · 2019-10-23T22:21:06.000Z

rhub did not want to work on my laptop but apparently worked on my desktop

it appears to do what it claims and offers a variety of platforms to test on

as far as I can tell it will not run reverse dependencies

Answer 9 · 2019-11-15T17:07:58.000Z

Pavel noted in #18

R CMD check --use-valgrind on the package fails for network because tests/speedTests.R uses specific numbers for time limits on its tasks. This means that running inside a virtual machine like Valgrind, which slows down the execution, can trigger false positives.

Answer 10 · 2019-11-21T00:23:49.000Z

Update: see #19

Answer 11 · 2019-11-21T00:25:51.000Z

Closing this, with a followup issue in #19