Le lundi 30 avril 2018, 03:19:20 CEST sebb a écrit : > On 29 April 2018 at 22:11, Hervé BOUTEMY <[email protected]> wrote: > > better than a discussion: a demo > > > > here is the website: > > https://asf-attic.github.io/ > > > > its source: https://github.com/asf-attic/asf-attic.github.io/tree/source > > its output: https://github.com/asf-attic/asf-attic.github.io/tree/master > > > > the only thing that is not included here is the code for the CI to > > checkout > > the output branch and update content after rebuild > > All I see is two disjoint branches. yes, that's what it's about
> It's not obvious how a user is supposed to update the master from the > source. people who know GitHub pages know: that's now part of common culture for many people and from an operational point of view, that's why in general there is a CI server doing the official build from source branch then committing output to output branch > > The build.sh script can be used to create a docs/ tree, but then what? when you build locally, it's just for yourself, because you want to check a change that is not as trivial as usual then the change in output branch is usually done by a CI server When used with Maven, there is scm-pulbish plugin for that: it's a common convention, and that's what is used by Apache Cayenne for example. When used with Jekyll or anything else as rendering engine, I don't know how people script the output commit: Apache Freemarker just tells in their documentation "To publish the built site, commit the output into the "asf- site" branch". And Apache Accumulo writes a serie of shell commands that they tell are launched by a Git hook (and not a CI server). And again, in general, there is a central official setup does the job in a central and official way, to avoid subtle differences when rendering from multiple personal configurations > > > But what is done here with GitHub GitPubSub equivalent can be done exactly > > the same way at Apache Software Foundation > > > > Regards, > > > > Hervé > > > > Le dimanche 29 avril 2018, 19:13:43 CEST Hervé BOUTEMY a écrit : > >> Le dimanche 29 avril 2018, 14:46:09 CEST sebb a écrit : > >> > On 29 April 2018 at 11:33, Hervé BOUTEMY <[email protected]> wrote: > >> > > Le dimanche 29 avril 2018, 11:04:44 CEST sebb a écrit : > >> > >> On 29 April 2018 at 09:41, Hervé BOUTEMY <[email protected]> wrote: > >> > >> > first, I want to reassure everybody: this is a discussion, to get > >> > >> > common > >> > >> > knowledge of how things work in other projects then may work in > >> > >> > the > >> > >> > future for Attic if we decide to do an equivalent setup > >> > >> > >> > >> +1 > >> > >> > >> > >> > Le dimanche 29 avril 2018, 07:50:21 CEST sebb a écrit : > >> > >> >> On 28 April 2018 at 12:48, sebb <[email protected]> wrote: > >> > >> >> > On 28 April 2018 at 12:37, Hervé BOUTEMY > >> > >> >> > <[email protected]> > >> > >> wrote: > >> > >> >> ... > >> > >> >> > >> > >> >> >> In Git, this would naturally be in a separate branch named > >> > >> >> >> "asf-site" > >> > >> >> > >> > >> >> How would that work for Attic? > >> > >> >> > >> > >> >> Where would the source files used to generate the site be held? > >> > >> > > >> > >> > There are multiple ways of doing, and GitHub documented it as > >> > >> > clearly > >> > >> > as > >> > >> > possible [2] (yes, what we do at ASF with GitPubSub is exactly > >> > >> > what > >> > >> > GitHub calls "GitHub Pages", with marketing bells turned on and > >> > >> > technical > >> > >> > details on the build solution turned off) > >> > >> > > >> > >> > The 2 common ways are: > >> > >> > 1. publish html from separate branch (which would be by default > >> > >> > "asf-site" > >> > >> > at ASF, and is "gh-pages" at GitHub) 2. publish html from a > >> > >> > subdirectory > >> > >> > on master branch (you see Attic current pattern?) > >> > >> > > >> > >> > I find the first option a lot more clear from a build+scm > >> > >> > perspective > >> > >> > than > >> > >> > the second one. This will avoid the exact same discussion we have > >> > >> > currently at Attic with svn to know who commits the generated > >> > >> > content > >> > >> > (& > >> > >> > when as a consequence): - CI after source-only commit? > >> > >> > - or user who builds on his machine then commits simultaneously > >> > >> > source > >> > >> > and > >> > >> > generated content? > >> > >> > > >> > >> > Then looking at ASF gitwcsub configuration [2], I had a look at > >> > >> > many > >> > >> > ASF > >> > >> > projects: the 2 ways are used. I picked Cayenne [3] case to show a > >> > >> > case > >> > >> > where: > >> > >> > - master branch is a source branch, with markup and a build script > >> > >> > - asf-site branch is a completely separate branch that contains > >> > >> > generated > >> > >> > html It uses Maven scm-publish plugin to update asf-git branch > >> > >> > with > >> > >> > generated html [6] > >> > >> > > >> > >> > But there is also Freemarker [4], that has a simple README telling > >> > >> > "To > >> > >> > publish the built site, commit the output into the "asf-site" > >> > >> > branch". > >> > >> > > >> > >> > Or Accumulo [5] which uses Jekyll and has some instructions to > >> > >> > publish > >> > >> > generated output to asf-site branch with a git-hook that I don't > >> > >> > fully > >> > >> > understand, but that maybe Attic members will prefer since it > >> > >> > seems > >> > >> > it's > >> > >> > more the common culture here > >> > >> > > >> > >> > Notice: I'm a Maven guy, I co-wrote the Maven scm-publish plugin > >> > >> > used > >> > >> > by > >> > >> > Cayenne, and I use it in many projects, initially with svn as > >> > >> > target > >> > >> > source control (in 2012, for svnpubsub & Apache CMS) then with git > >> > >> > also, > >> > >> > when GitHub pages became popular. But I see that it's not the > >> > >> > right > >> > >> > choice at Attic because it's not the most common Attic culture. > >> > >> > >> > >> Attic does not produce source code. > >> > >> The only output from its SCM is the website. > >> > > > >> > > yes, like any other website that I showed: source code here is a > >> > > markup > >> > > language (be it Markdown, xdoc, static content, or anything else) > >> > > Attic is really exactly the same > >> > > > >> > > What makes Attic different is that Attic does not have any other repo > >> > > for > >> > > "programming" code: that's true, but does not change anything > >> > > regarding > >> > > site> > >> > > > >> > >> AFAICT, all the above examples have a branch which contains the > >> > >> source > >> > >> for building the website. > >> > > > >> > > yes, I explained I chose them exactly for that reason > >> > > > >> > > I can show you Airavata site, which is quite simple and did the other > >> > > choice: https://github.com/apache/airavata-site > >> > > > >> > > source is in source, output is in content in the same branch > >> > > >> > This is equivalent to Attic. > >> > > >> > > it could have been: source in master branch, content in asf-site > >> > > branch > >> > > which is the most common setup in GitHub pages (= what people > >> > > nowadays > >> > > know a lot) > >> > > > >> > >> 1) The source is edited. > >> > >> 2) Run the build script to create the output in a clean subdirectory > >> > >> 3) Copy the subdirectory tree to the asf-site branch > >> > >> 4) commit the asf-site branch > >> > >> 5) The entire asf-site branch is then published via pubsub. > >> > >> > >> > >> What Attic does currently is: > >> > >> 1) & 2) as above > >> > >> 3) commit the changes > >> > >> 5) as above > >> > >> > >> > >> i.e. there is no need to copy the generated output anywhere because > >> > >> it > >> > >> is part of the same repo. > >> > >> > >> > >> This works because svnpubsub is set up to get its source from the > >> > >> docs/ subdirectory > >> > > > >> > > yes, the setup with source and output in the same svn repo or Git > >> > > branch > >> > > makes it simple to checkout, but it mixes 2 types of files (source > >> > > and > >> > > generated) > >> > > > >> > > separating source and generated in 2 separate locations (separate svn > >> > > root > >> > > or different branches in the same git repo) makes things more clear, > >> > > at > >> > > the cost of an extra step to check out the generated content then > >> > > update > >> > > with the updated content > >> > > >> > The workspace still contains both source and generated output in the > >> > examples I have seen. > >> > > >> > I assume it is ignored by SVN/Git so does not get committed or show up > >> > as a local change. > >> > >> I showed you Cayenne, Freemarker and Accumulo that are not like this. > >> Here we go back to Freemarker: > >> - source: https://github.com/apache/freemarker-site > >> - output: https://github.com/apache/freemarker-site/tree/asf-site > >> > >> > > yes: choose your issue > >> > > personally, I prefer the second setup (clear but a little harder to > >> > > setup) > >> > > I don't like having mixed content in one repo (source and generated > >> > > output) > >> > > > >> > > If everybody understands that these 2 setups a completely equivalent > >> > > but > >> > > really prefer the mixed one (just to avoid a second checkout), I'll > >> > > let > >> > > you > >> > > go: I don't have any problem myself, I make a strong difference > >> > > between > >> > > source directory and output directory > >> > > >> > There's still mixed content in local workspaces unless you generate > >> > the output in a separate tree. > >> > > >> > > But if people start to edit output directory instead of source (like > >> > > it > >> > > is > >> > > so easy to do in the mixed content setup), you're at risk > >> > > >> > It's also possible to checkin the generated output if it's not > >> > properly ignored in a 2 branch version. > >> > And then wonder why the site does not get updated. > >> > >> that's why in general there is a .gitignore or svn:ignore that is > >> properly > >> configured > >> > >> > >> I don't know if gitpubsub can take its input from a subdirectory of > >> > >> a > >> > >> branch. > >> > > > >> > > it can: see Airavata site > >> > > >> > Ah - I see now. > >> > > >> > The webserver defines the site to be under content/, so the branch can > >> > contain other files in parallel directories. > >> > > >> > >> If not, then we will have to change strategy in order to use Git. > >> > >> Otherwise, we have a choice. > >> > > > >> > > we have a choice > >> > > >> > Yes. > >> > > >> > I prefer the status quo, not least because it involves fewer changes > >> > (I think only renaming docs/ to content/ if we move to Git). > >> > Using multiple repos would involve updating instructions as well. > >> > >> no, it's not multiple repos but multiple branches of the same repo > >> > >> > But if the majority want to change I won't object. > >> > >> since you are the guy who does the buidbot script, that does the commit, > >> you'll have to be confident that you can code the multi-branch option > >> > >> > > Regards, > >> > > > >> > > Hervé > >> > > > >> > >> > Regards, > >> > >> > > >> > >> > Hervé > >> > >> > > >> > >> > > >> > >> > [1] > >> > >> > https://help.github.com/articles/configuring-a-publishing-source-f > >> > >> > or-> > >> > gi th > >> > >> > ub-pages/ > >> > >> > > >> > >> > [2] > >> > >> > https://github.com/apache/infrastructure-puppet/blob/deployment/mo > >> > >> > dul > >> > >> > es > >> > >> > /g > >> > >> > itwcsub/files/config/gitwcsub.cfg > >> > >> > > >> > >> > [3] https://github.com/apache/cayenne-website/ > >> > >> > > >> > >> > [4] https://github.com/apache/freemarker-site > >> > >> > > >> > >> > [5] https://github.com/apache/accumulo-website > >> > >> > > >> > >> > [6] > >> > >> > https://maven.apache.org/plugins/maven-scm-publish-plugin/various-> > >> > >> > >> > >> > tip > >> > >> > s. > >> > >> > ht > >> > >> > ml#Git_branch
