Hi Graham,
On 24 January 2021 at 18:06, Graham Inggs wrote: | Hi Dirk | | On Sun, 24 Jan 2021 at 14:47, Dirk Eddelbuettel <e...@debian.org> wrote: | > R code can promote warnings to errors on a per-session basis: | | Thanks! This could be useful in the r-cran-tmb autopkgtest. | | > I am still not quite sure what the best way here is. With hard '==' warnings | > we get ourselves the need for a transition within a transition. My | > instictive preference would be to just rely on rebuilds of *TMB after Matrix | > updates majors (which is only every so and so many years). We likely to not | > need a '==' on minor versions. | | Right, so we just need a way to know when r-cran-tmb and any other | affected packages need a rebuild. How do upstream know this? Well "trial and error" or more generously: "strongly suggested to test". When a package 'A' is uploaded to CRAN, the maintainer asserts that the package is in good standing, including 'no changes to worse' (i.e. breakage) on dependent packages. At CRAN we have a full dependency graph as it is one repo, just how Debian knows what's in it. You can walk that graph. One could then walk down that graph and look at every 'reverse depency' and how it encoded the relationship. But I am using conditional language here as this just about never has '=' constraint. CRAN had closed for the holidays. Matrix 1.3.0 was rushed out just before that in late December, and created a lot of damage. (For context, I am 'upstream' for Rcpp and test it 2200+ reverse depends on an old Debian testing box. I saw a lot of breakage and reported that to the mailing list.) Martin, the author/maintainer of Matrix, is a long time friend and R Core member. (And, per historical anecdote, so into GNU + FSF that around 1994 or so he ensure the R system became GNU R ...) He knows what he is doing, this one was a very unusual release. "Normally" when we see required changes before an upload to CRAN we coordinate -- I had to do just that for my (upstream, containing parts of Boost) "BH" package from December to January as six or so packages "broke", some for trivial reasons that Boost Geometry now needs C++14 etc. So in short: *normally* we do not need to do anything. *Normally* these things are caught at CRAN before they get to us. So I would recommend not to put other formalisms and red tape on this. Let's wait and see -- we can rebuild the two *TMB packages and move on. | I'm awaiting autopkgtest results on the binNMU'd r-cran-tmb. I will | update this bug and #980809 when they are in. That is excellent. Thanks for doing that. Dirk | | Regards | Graham -- https://dirk.eddelbuettel.com | @eddelbuettel | e...@debian.org
