-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA512 Tim Cross <theophil...@gmail.com> writes:
> David Masterson <dsmaster...@gmail.com> writes: > >> Tim Cross <theophil...@gmail.com> writes: >> >>> David Masterson <dsmaster...@gmail.com> writes: >>> >>>> "Michel Schinz" <mic...@schi.nz> writes: >>>> >>>>> Just for the record, I also ran into problems when installing Org 9.6 >>>>> using Emacs' package system on top of an older version that came with >>>>> Emacs. If I tried to install it as usual (M-x list-packages, then >>>>> install the package from there), I had errors during compilation related >>>>> to `org-assert-version`, and then if I restarted Emacs, I would get a >>>>> fatal error in an unrelated package. >>>>> >>>>> I managed to solve that problem by: >>>>> 1. uninstalling Org 9.6 and exiting Emacs, >>>>> 2. starting Emacs with -q, >>>>> 3. installing Org 9.6 from there (using M-x list-packages as usual), >>>>> 4. restarting Emacs. >>>> >>>> Interesting! I tried this (essentially) and it worked for my case. In >>>> my case, I had a built-in Org-9.3 and I was trying to use list-packages >>>> to install Org-9.6. I checked that using -q still added Org-9.3 to >>>> the >>>> load-path, but, since Org wasn't loaded, the install via list-packages >>>> worked. >>>> >>>> The question is what's the proper way of doing this without '-q'? This is what is in my 'init.el' file - ===== ;;; init.el --- sharon's config -*- eval: (read-only-mode 1) -*- ;; Make sure that Git version of Org mode is being loaded instead of the built-in version. ;;; (add-to-list 'load-path (expand-file-name "/home/boudiccas/.emacs.d/elpa/org-9.5.5")) (add-to-list 'load-path (expand-file-name "/home/boudiccas/.emacs.d/elpa/org-9.6")) (add-to-list 'load-path (expand-file-name "/home/boudiccas/git/org-contrib/lisp")) (require 'package) (setq package-enable-at-startup nil) (package-initialize) (require 'ob-tangle) (org-babel-load-file "/home/boudiccas/.emacs.d/config22-2.org") ===== And the first thing that is loaded is org-9.6 *before* anything else! And you could equally have org-mode loading from a git repository on your hard drive, just by changing the filepath. Super-easy without having to faff about with 'emacs -q' etc. And as you can see I'm using org-9.6 from elpa, and I didn't move over to it until this morning, so far so good. Thanks Sharon. >> >> [...] >> >>> I don't think there is any safe way to install an updated version of >>> org-mode other than >>> >>> 1. Use the -q approach outlined above >> >> Thinking about it, this only works if Org is in elpa as melpa (etc.) are >> not added to package-archives. You'd have to do some handwritten elisp >> out of *scratch* to setup package-archives if Org-9.6 was still coming >> out of melpa. That's why this can only be labeled as a hack and not a >> solution. >> > > Well, yes, if your going to use this technique to load a package which > is not in the default package archives you would need to add that > archive first. People who use this technique often just have a > 'update.el' file which they load/evaluate when starting Emacs with -q. > > >>> 2. Craft your init.el file such that org functionality is only loaded >>> when explicitly requested and always update as the first action after >>> starting emacs. >> >> In this case, something happened in package-install when trying to >> install Org-9.6 with a built-in Org-9.3. During the compilation check >> (.el -> .elc) many files failed because the new 'org-assert-version' >> macro was not defined. Sort of like, after package-install started >> working on Org-9.6, org-macs.el (where org-assert-version should be) got >> loaded *before* the new load-path had been set causing it to load the >> old one from 9.3. Thereafter, everything went awry. >> > > You must have some custom code in your init.el or are using something > like use-package as package.el doesn't try to install or upgrade > packages during init by default. > > Where people can come undone is when they are using use-package and set > the :ensure t option. In this case, use-package will not know about the > bundled version and will attempt to install org from ELPA. If use > package runs after org has already been loaded (possibly because some > other package has been loaded which depends on/requires org mode and has > loaded the bundled version) then things will break because you will end > up with a mixed version install. This is why I always ensure org is the very > first use-package in my init.el and it comes before any other code which > loads or initialises anything. > >>> The first approach is actually the easiest. The second is hard to get >>> right and very fragile because packages like use-package and more >>> specifically, other packages with leverage off org functionality, make it >>> impossible to reliably know exactly when org is loaded. >> >> Using ':after" in use-package is supposed to help that, but I'm not sure >> it is reliable. Packages are often incomplete about what other packages >> it depends on. >> > > You cannot rely on :after. The problem is, other packages may require > org functionality and will load org when they are loaded. This can be > very subtle as there are a lot of packages which only make very small > use of org mode, but even that requires that org mode is loaded. > >>> An approach used by many 'canned' distributions is to postpone package >>> updates. You have a function you run to check for updates which >>> generates a list of packages to update and writes that list to a >>> file. Each time emacs is started, it looks for this update list and if >>> it finds it, it installs packages updates at the very beginning of the >>> init process (before any of your other init.el code or custom >>> blocks). The process also looks for org in the list of packages to >>> update and if it is found, updates it first. >> >> Probably doesn't work in this case as you would need to be able to use >> package.el suggesting that the load-path has been updated for all >> built-ins already. The thing I note is that the load-path has already >> been updated for built-ins at the beginning of >> '~/.emacs.d/early-init.el', but the libraries haven't been loaded yet >> (unless needed). That's okay if the newer version of a package then >> cleanly replaces all the files in the old version. >> > > It does work. Distributions like Sapcemacs use this approach and do not > have the mixed versions issue. I've not looked closely at their > implementation, but they well might do all this work as part of the > early-init.el file. > > >>> I don't think there is a safe way to load org mode after the init >>> process i.e. after booting emacs by M-x package-update. >> >> Where is package-update called in the boot process of emacs? I don't >> see package-update in Emacs v2.7. >> > > Sorry, I wasn't clear. > > What I meant is that there is no clean and reliable way of running the > package update process after emacs has initialised. The package.el API > has changed, but previously, you would run M-x package-update-packages > to update installed packages. Now it is M-x package-update-all. With > many packages, you can run this command and update your installed > packages. However, more complex packages, like org-mode, cannot handle > this due to potential mixed version problems - a problem which has > existed for many years. > > By default, there is no automated package update setting (unless that > has been added in recent versions). There are some add-on packages which > can do this and the elisp to make this happen is trivial (but much > harder to do it and make it robust). > > >>> I've had good success using straight.el. I had to be careful regarding >>> how I structured my init.el file (ensuring any straight stuff happens >>> first and the first use package stanza is for org. The main reason >>> straight works well for me is that my work flow is to do a M-x >>> straight-pull-all when I want to update my packages. This does a git >>> pull for all the sources, but does not do any build/install. This occurs >>> when I next start Emacs and because I have all the straight stuff at the >>> start and because org mode is the first straight-use-package, the update >>> and install happens before any other org functionality is loaded, >>> avoiding mixed version issues. >> >> Where do you get straight.el? I don't see it in [m]elpa. > > No, to use straight.el, you add some code to your init.el which will > retrieve the current straight.el file from the github > repository. Straight.el does not use the ELPA/NONGNU ELPA/MELPA archives > - at least not directly. It does use information from those archives to > build recipes which it sues to obtain the code. Essentially, straight.el > clones the package source repositories and builds it locally. It give a > lot more control and is really great if you want to modify the code > (basically, it creates a local fork). Of course, with all that extra > power, there is also additional responsibility placed on the user. You > may need to tell it which versions you want to use, which ones to 'pin' > to, which branches or fix merge conflicts when you have made changes and > want to update from the upstream soruces etc. I've been using it for > about 12 months and find it really good. YMMV. - -- A taste of linux = http://www.sharons.org.uk TGmeds = http://www.tgmeds.org.uk DrugFacts = https://www.drugfacts.org.uk Debian 11.5, fluxbox 1.3.7, emacs 29.0.50, org 9.6 -----BEGIN PGP SIGNATURE----- iQJRBAEBCgA7FiEELSc/6QwVBIYugJDbNoGAGQr4g1sFAmOTWQodHGJvdWRpY2Nh c0Bza2ltYmxlMDkucGx1cy5jb20ACgkQNoGAGQr4g1sBchAAkByRu8RZ6EV9WkAi tbYOMCW9+rBQ9Oib2i6pj4XuGC1opjjjgMr84y2+UfgFm1bbcFo6bv1hi33UFvM2 9D8ORzSPW/wcPjGILcwHS6w1hbwkflB8qwRDwL7TxUFuZqCZ3mr9q7aLZR4ZVuf9 3WFzUHd30O61H6rLKrJWRPJgB2zYcUS4pbsiurOTK35sYoiogAshoV4akTNxQefV TmLFcoHmJvK0M+L/5BhPJblvrG0jkudUyMCYPeAHGxDfDpXiW/aFdYJuibqhk4mm nblCIe1EFBa5Lxu3m0lTkY7X9I/XtS1cf92gkj7qA9Y/ZYtVfWO54FyDA5sJFOZ2 aHcQrGS2vqoU1HNYv9u0CVj+/QJQbUSGrI60lrwIikOe+Xt+iHVlyx6dmmgKHLzt YHuZVKwMZxOEJtUjVwHc23+MzDmHc1WbJ8o5yOomFnYivFg/P0GQpY0o9O3CDJCV yV/ux7qPSPlyQZcJqKVHjrg05y/t76i5CPsiYAeUUQPa8jdU3F5SHxtpmuf9KPgQ +5EmUYkYodcNyss93YAuPEamKI491nLuWkc+F4t+fi24QS/t3FJaemkfW+UIIBuk gpg8/fgMkOdrs8E4IEwb7wZkCEHl52J0WypsGv9Of+vcrb5YXdPk7x/dJoHUmADm SremPednv0xBoZCKg12orlIQzIM= =cfTm -----END PGP SIGNATURE-----
