On 8 June 2026 at 06:26, Jeff Newmiller via R-help wrote: | Don't upgrade packages just because upgrade.packages says there are new packages every day... that way leads to obsessive-compulsive disorder (OCD). As a general rule, this is objectively questionable advice. CRAN _excels_ at being _always being buildable / installable_. To use a software analogy, we could call this 'build at @HEAD' as one would at a large mono-repo. Which is what CRAN essentially is. I have updated more or less daily / multiple times per week since CRAN started, and I have done so mostly on Linux where it is "even more expensive" as one generally instals from source (unless newer services like r2u or cran2copr are used). But if you know what you are doing, and design you deployments well with packages you trust, and maybe avoid some others you trust less / have shown a tendency to change abruptly then this is as stable as it gets. I sometimes reference some examples for that: CRANberries has been running more-or-less unalted since 2007, on a machine that gets the daily (or close to daily) updates and has stayed alive near-constantly. Same for some things I run that are less visible to the outside. So ... "your mileage may vary". CRAN supports "continuous" updates just fine. It does it to itself too. Dirk -- Dirk Eddelbuettel | edd at debian.org | http://dirk.eddelbuettel.com Support my Tour de Shore 2026 ride benefiting Maywood Fine Arts! More info at https://dirk.eddelbuettel.com/blog/2026/04/03#sponsor_tour_de_shore_2026
You claim that this position is "objectively questionable advice" ... but Stack Overflow moderators concluded that this was an opinion-based argument and shut it down. [1] While CRAN is remarkably robust at ensuring that packages compile and pass checks together, user-facing API changes and behavioral shifts still happen frequently. For an active research project, updating daily introduces unnecessary moving targets. CRAN's strength as a monorepo is excellent for developers, but for end-users running data pipelines, code stability and isolated lockfiles usually trump tracking the bleeding edge. The very stability that you rely on also means that you can also afford to delay updates until you have time to consider the implications of api shifts and algorithm missions that package authors inject on their own schedule. I will take the SO lead and not respond further on this topic. On June 9, 2026 5:24:26 AM PDT, Dirk Eddelbuettel <edd at debian.org> wrote:> >On 8 June 2026 at 06:26, Jeff Newmiller via R-help wrote: >| Don't upgrade packages just because upgrade.packages says there are new packages every day... that way leads to obsessive-compulsive disorder (OCD). > >As a general rule, this is objectively questionable advice. > >CRAN _excels_ at being _always being buildable / installable_. > >To use a software analogy, we could call this 'build at @HEAD' as one would >at a large mono-repo. Which is what CRAN essentially is. I have updated >more or less daily / multiple times per week since CRAN started, and I have >done so mostly on Linux where it is "even more expensive" as one generally >instals from source (unless newer services like r2u or cran2copr are used). > >But if you know what you are doing, and design you deployments well with >packages you trust, and maybe avoid some others you trust less / have shown a >tendency to change abruptly then this is as stable as it gets. I sometimes >reference some examples for that: CRANberries has been running more-or-less >unalted since 2007, on a machine that gets the daily (or close to daily) >updates and has stayed alive near-constantly. Same for some things I run that >are less visible to the outside. So ... "your mileage may vary". CRAN >supports "continuous" updates just fine. It does it to itself too. > >Dirk > >-- >Dirk Eddelbuettel | edd at debian.org | http://dirk.eddelbuettel.com > >Support my Tour de Shore 2026 ride benefiting Maywood Fine Arts! More info at >https://dirk.eddelbuettel.com/blog/2026/04/03#sponsor_tour_de_shore_2026-- Sent from my phone. Please excuse my brevity. [[alternative HTML version deleted]]
Dr. Eddelbuettel, Thank you for your message, quite reassuring for me. I agree with you, I see no problem in updating frequently the packages I use here. On system, Windows 10, with a good Internet connection, this takes just about one or two minutes, depending on the number of packages, and by that way I know I am using the most recent and corrected versions of the packages. Best regards, Paulo Barata Rio de Janeiro - Brazil ------------------------------------------------------- On 09-Jun-26 9:24, Dirk Eddelbuettel wrote:> > On 8 June 2026 at 06:26, Jeff Newmiller via R-help wrote: > | Don't upgrade packages just because upgrade.packages says there are new packages every day... that way leads to obsessive-compulsive disorder (OCD). > > As a general rule, this is objectively questionable advice. > > CRAN _excels_ at being _always being buildable / installable_. > > To use a software analogy, we could call this 'build at @HEAD' as one would > at a large mono-repo. Which is what CRAN essentially is. I have updated > more or less daily / multiple times per week since CRAN started, and I have > done so mostly on Linux where it is "even more expensive" as one generally > instals from source (unless newer services like r2u or cran2copr are used). > > But if you know what you are doing, and design you deployments well with > packages you trust, and maybe avoid some others you trust less / have shown a > tendency to change abruptly then this is as stable as it gets. I sometimes > reference some examples for that: CRANberries has been running more-or-less > unalted since 2007, on a machine that gets the daily (or close to daily) > updates and has stayed alive near-constantly. Same for some things I run that > are less visible to the outside. So ... "your mileage may vary". CRAN > supports "continuous" updates just fine. It does it to itself too. > > Dirk >