Web   ·   Wiki   ·   Activities   ·   Blog   ·   Lists   ·   Chat   ·   Meeting   ·   Bugs   ·   Git   ·   Translate   ·   Archive   ·   People   ·   Donate
summaryrefslogtreecommitdiffstats
path: root/docs/rationale.txt
blob: 3c8d0598d52b3b362abfb021156de61d527ecded (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
= Rationale =

== The current situation ==

* We need more people to hack on the core sugar modules. A lot of maintenance is necessary and so many features that are a fundamental part of the initial vision are still unimplemented.
* Building sugar is currently too complicated and unrealiable. While this is just an annoyance for an experienced hacker, it can a blocker for newcomers. If they can't even get the build to work they are likely to give up and never contribute.
* It is a difficult problem to solve because we are depending on system components that, while having reasonable ABI policies, are moving very fast (recently with gobject-introspection and gtk3). The number and the level of our dependencies is also much higher than the average project.
* It's not just about building and running glucose. A good development environment needs to include all the system components activities are allowed to expect.
* Unrealiability brings more unrealiablity. Build breakages are apparently going unnoticed for several days. I suspect part of the reason is that people just expect things to be broken and deal with it, without reporting or fixing the issues for everyone.

== How sugar-build tries to improve the situation ==

* Rather then trying to adapt to the underlining system, it ensures we have a consistent system on all the distributions we support, so that the same identical build process works on all of them. The downside is obvious, we can't expect to work on old distributions. But I think that's a price we can pay for easier maintenance and especially for higher reliability. It's better to support only a few distributions and work flawlessly 99% of the time on the time, then promise support for everything but never work realiably.
* To verify that all the system components we need are installed we use cross-distribution checks. Though to make life easier for developers we map these to distribution-specific packages and install all of them automatically before starting to build.
* We only build master of the sugar modules. That's what we want developers to work with. The less build configurations we need to maintain the more likely they are to be realiable. For the same reason we only support the most recent version of a distribution.
* We build only a couple of essential activities, browse and terminal, to improve the initial develop experience. The rest can be downloaded from activities.sugarlabs.org and they should work because we are ensuring consistent system dependencies.
* We are using jhbuild internally to keep the amount of custom code to the minimum, but we are not exposing it in the build commands, which are a just few make targets. jhbuild is very flexible but too complicated for us. People should be able to contribute to large parts of the sugar code base without much linux or GNOME experience.
* Ideally we would only build sugar modules and depend on distributions for everything else. And I hope that's what we will do at some point in the future. Unfortunately in the current development cycle we will need a few components that are not yet packaged, not even written or upstreamed. I believe we can reduce the impact on realiability at the minimum by building only modules that are absolutely required and by checking out a specific commit rather than master. We don't want to get new code from upstream without first verifying that it works well on all the distributions we support. The same is true for jhbuild itself.
* To ensure reliability of the build scripts and of the sugar code itself, we are going to setup buildbot slaves for all the supported distributions. They will notify the irc channel and perhaps the committers when a build breaks. When it happens I think fixing it should be everyone top priority. Ideally we would also improve our review process so that no code is committed before it has gone through buildbot, gerrit works very well for that but there might be less invasive ways to do it. Regardless of the exact process and tools, patches that add new dependencies to the codebase should come with a corresponding sugar-build patch. Nothing that is pushed should break the buildbots.
* Ensuring the build works is an important first step. Though integration tests, even if just to verify that stuff actually run, will go a much longer way.