Contributing to documentation [Was: Installing gfortran]

Re git and empty directories... git is structurally incapable of recording them in the repo. A common workaround is to touch a .gitkeep file in the directory, but I suspect this will never become an automated feature of git because it cannot be hidden from the user without making the chosen filename off-limits to the user.

That seems like a design flaw, but a pretty easy one to work around.

On 30/04/2022 11:07 a.m., Jeff Newmiller wrote:

Generating patch files is one of the most fundamental capabilities of git. Changes to the Linux kernel are (almost?) universally submitted via patch files generated from git.

svn uses nearly standard patch files, but they record svn revision numbers.  I'd guess directly applying a git patch file to an svn working copy would almost always work, but I'm not sure it would always be applied correctly in the case where the patch was created from rev X and applied to rev Y.

Re git and empty directories... git is structurally incapable of recording them in the repo. A common workaround is to touch a .gitkeep file in the directory, but I suspect this will never become an automated feature of git because it cannot be hidden from the user without making the chosen filename off-limits to the user.

That seems like a design flaw, but a pretty easy one to work around.

Duncan Murdoch

On April 30, 2022 7:45:40 AM PDT, Duncan Murdoch <murdoch.duncan at gmail.com> wrote:

On 30/04/2022 9:53 a.m., Patrick Schratz wrote:

     If that is the case, why not contribute to the documentation? That
     is the whole point of an open source project after all.

Because often it is not easily accessible, e.g. living in an ancient SVN repo or lacking (an easy) and clear contribution guide.

There's a mirror of that repo at https://github.com/wch/r-source .  It is of course unofficial and not maintained by R Core so I could understand you might worry about using it, but as far as I know it is well maintained.  The only difference that I ever heard about in the past was that the official svn repo had an empty directory somewhere or other, and git at the time didn't support empty directories.  I don't know if either of those is still true.

WRT to the Mac dev instructions, I can see that the source lives in https://github.com/R-macos/R-mac-dev <https://github.com/R-macos/R-mac-dev> which is definitely a good start.
Yet I think it needs way more cross-linking between the repos, more ?official? pointers and ?how-tos? to really also encourage people to contribute.
The README could give more detailed contribution instructions, such as whom to tag for a PR, what should go there and what not, possibly stating that it?s the official documentation and define it from other ?random? orgs on developer portals - all of these could e.g. go into a |CONTRIBUTING.md| which is a widely known source for such information.
Just some personal thoughts though which could potentially considered to improve things.

To be clear, I acknowledge your effort in opening things up to platforms like GH - which not all parts of R/CRAN are doing at the moment AFAIK.
And yes, when complaining about things not being optimal, one should also put in effort to make things better.
So I?ll see if I can put some time in to improve things and see how the experience is.

If you're happier working in git than in svn, what you could do is fork the mirror repo to your own git repo, and make your proposed changes there.  If they are good changes it won't be hard for someone (maybe even you) to convert into the appropriate format to merge into svn.

The way R development changes is when a change makes things easier for the devs.  I suspect whether it's easier for you is only important to them if you've got a history of making helpful contributions: they like help, they don't like arguments about how to do things.  (I'm saying this as a former member of R Core.)

Duncan Murdoch

     The problem is that generally they cannot. You are looking something
     up, because you don't know about it so you can't judge whether it is
     a good answer (SO is good example proving why crowd-souring the
     definition of truth doesn't generally work). At best you may know
     the person and thus judge by that, but even then you may not know if
     the information is still accurate.

I see your point here and generally agree that it?s hard making such judgements in this position.
Yet I disagree on referring to Stackoverflow as a ?crowd-souring the definition of truth doesn't generally work?. Without SO, we would be nowhere where we are today and I?d argue it has done a lot more positive things than negative ones to every single person who ever accessed it. >
Cheers
Patrick

On 25 Apr 2022, at 1:04, Simon Urbanek wrote:

         On Apr 23, 2022, at 7:44 PM, Patrick Schratz
         <patrick.schratz at gmail.com> wrote:

         FWIW blog posts which explain such things usually have a (good)
         reason - they aim to help people getting started when the
         official documentation is either unclear, hard to find or
         incomplete.

     If that is the case, why not contribute to the documentation? That
     is the whole point of an open source project after all.

     The problem with random blogs is that many of them are written by
     people trying to find an answer without much knowledge on the
     subject and often post very bad advice that does not necesarily
     address the actual issue. There are rare exceptions of knowledgeable
     people posting explanatory blogs, but if you search for an answer
     you have no way of knowing whether it is of the good kind. In
     addition, blogs tend to get out of date quickly, so what used to be
     a good advice may not be anymore (prime example was the R 4.0.0
     release which made a lot of the "hacks" obsolete and the well-meant
     advice out there has only led to more problems).

         It?s on the readers themselves to decide whether such blog posts
         are trustworthy or useful.

     The problem is that generally they cannot. You are looking something
     up, because you don't know about it so you can't judge whether it is
     a good answer (SO is good example proving why crowd-souring the
     definition of truth doesn't generally work). At best you may know
     the person and thus judge by that, but even then you may not know if
     the information is still accurate.

         I have personally profited so often from blog posts of others
         already and therefore find the general advice to not consult
         such resources quite shortsighted.
         Of course the official documentation should always be the first
         point to have a look at - and in this case the required
         information would have been there.

         Apologies for going partly off-topic but I think this point is
         important.

     I agree. That's why I think it would be great if those that have the
     knowledge would help the community to improve the documentation. Of
     all the contributions to R it is the easiest.

     That said, I also agree that complementary information is very
     useful, in particular if it explains the "why" as well - which may
     be too far out of scope of the canonical documentation. In that case
     it is easier to spot mismatches, e.g., if it becomes out of date. It
     is not without precedent to reference such external documentation if
     it is maintained.

     Anyway, I'd like to encourage everyone to contribute - it may be
     pointing out issues in the documentation or by sending PRs with
     proposed updates or posting here. Some did in the past (like you,
     Jan or Bob, thanks!), but the more contribute the better for the
     community. Often this may also uncover genuine issues that should be
     addressed rather than worked around (like the lack of the symlink in
     the gfortran-12 tar-ball discovered just this morning...).

     Cheers,
     Simon

         Cheers
         Patrick

         On 23 Apr 2022, at 2:13, Simon Urbanek wrote:

             For posterity - please always consult

             https://mac.r-project.org/tools/
             <https://mac.r-project.org/tools/> (linked from CRAN)

             The old locations like libs* are no longer updated and have
             been deprecated in favor of /tools and /bin which are
             maintained for all builds. Similarly, I would strongly
             discourage following any advice from blogs as they tend to
             be outdated, wrong or both.

             Cheers
             Simon