In Q3 2009 we decided that Q1 2010 will be occupied with getting SUMO running on Tiki 5.1. We have both a high level plan and a detailed implementation plan. In summary, the plan was to upstream our patches and then upgrade.
During the upstreaming process, we have developed some concern about the viability of building on top of Tiki 5.1, as the upstreaming and initial testing has not gone as smoothly as planned. We will hence re-evaluate our options once the upstreaming is finished, before we actively port our production instance.
We had originally intended porting our production instance to 5.1, running through a couple of release cycles, and then evaluating. However, we can reduce time to evaluate and possible investment by performing this evaluation earlier. Finishing upstreaming gets us to the point of having paid our ethical debt to TikiWiki by giving back our changes and improvements to the community, so this was important to complete.
The evaluation criteria are as follows:
- TikiWiki 5.1 Codebase: This can be evaluated by looking at the code and running tests against tiki-trunk.mozilla.com.
- Is the new codebase easier to understand and develop further than the old one? That is, will it be easier/more enjoyable for our webdevs to do their jobs? [all devs, subjective]
- Will we be able to get more done in a release cycle? [all devs, subjective]
- Is the new codebase more robust? [all devs, subjective]
- Is the new codebase at least as performant as the old one? [?, objective: run perf tests]
- Is the new codebase actively good on each of these criteria, as opposed to just better than before? [all devs]
- How does the current codebase perform versus our litmus tests? [WebQA, objective: litmus tests] (We will not be able to run Selenium tests without porting themes as it depends on the DOM.)
- Our implementation
- How would we implement our build process in order to keep our TikiWiki up to date? [laura, [Draft Build Process]
- How will we separate out our local changes that are not going to be upstreamed to TikiWiki? [laura, [Draft Build Process]
- What is the scope of these local changes? [james /MaintainedLocalChanges]
- What is the complexity and time required to create and maintain a local codebase consisting of templates, modules, and overridden files?
- How would this affect the deployment process and our desire to move towards continuous integration?
- What new local code do we need to write or rewrite in order to run Tiki 5.1?
- How much work would it be to update our SUMO themes to work with Tiki 5.1?
- Related to above question: how much additional work would it be to keep the new templates consistent with the old w.r.t. user/contributor experience?
Option 1: continue with Tiki
One of the rationales for upgrading Tiki to current was to enable us to rewrite parts of the code as we see fit. This includes improving existing sections of core code and modules.
We need to identify which parts of Tiki we plan to rewrite or refactor. The criteria for choosing these are as follows:
- Existing poor code quality and/or fragility
- Impact of refactoring: choose areas that will have the biggest impact for users and developers
- Ease of refactoring: While this would never be the only criteria, there is certainly value in choosing low hanging fruit
Option 2: Redevelop SUMO platform
In Q3 2009 we discussed three options: Upgrade, Fork, or Port (see []).
Having upstreamed the vast majority of our changes, the Fork option differs from Upgrade in the following respects:
- No need to work out complex build and deployment process
- No need to support two code repositories
If we have reached this point by giving the TikiWiki 5.1 codebase a negative evaluation, then the Fork option is less attractive than it once was. The main advantages are
- User and contributor experience would remain the same
- We have working code and would not need to invest time reimplementing features
The outcome of the Porting options matrix we developed in Q3 2009 ([]) was that none of the other existing platforms look significantly better than TikiWiki. That means that this option represents reimplementing the platform from scratch.
When considering this option we should remember the pros and cons of porting to a new platform that we analyzed previously.
If we were to implement from scratch we would use a modern web framework such as Zend Framework, Kohana, or Django. For this application I would recommend one of the more full featured frameworks (ie probably not Kohana) for the ready availability of reusable components, which would reduce implementation time.
We would need to work up an SRS of features that require/actually use and prioritize these. It may be possible to move parts of the code over before the new system is complete. For example, we have discussed reimplementing forums. We could implement forums in the new system and then migrate them over.
As a result of the SUMODEV meeting on 01/25/10, we decided not to move ahead with the TikiWiki upgrade.
- Looking at the new code, while some architectural improvements have been made and many features added, it is apparent that improvements have not been made in the parts of the code that we use.
- Some portions of the code have become even more bloated, in particular tiki-js.js which has ballooned from 1200LOC to 1700LOC, and tikilib, which has been partially split off into wikilib but managed to grow nonetheless. These were two of our original pain points in Q3.
- Each feature has been harder to upstream than anticipated. A larger number than anticipated have not and will not be upstreamed and would require local maintenance.
- Some upstreamed features have been adapted to work differently (for example CSAT) and we would have to either support our local versions or rewrite our code to work with the Tiki versions. Some existing features have this problem as well (themes).
- Having to work around this local adaptation of features will slow our release velocity.
- It still seems difficult if not impossible to incorporate unit tests. We cannot see a situation where it would be possible to have decent coverage. The new code is not more robust and definitely not more testable.
- Perceived performance is much slower. This is already a pain point for our contributors and we cannot afford to have it get worse.
- The increased complexity of releases and adaptations needed would flow on to other projects using SUMO (MoMo and Firefox for Mobile).
We plan to undertake a rewrite of SUMO, likely component by component. The dev team will meet ASAP to formulate an initial plan.