Architectural decision records

From LimeSurvey Manual

The purpose of this page is to keep a record of large technical decision, to track their progress and rational.


The template for an architectural decision is as follows [1]:

In the context of <use case/user story u>, facing <concern c> we decided for <option o> and neglected <other options>, to achieve <system qualities/desired consequences>, accepting <downside d/undesired consequences>, because <additional rationale>.


todo, definition of done,

Add new decisions at the top.

TODO: Add Psalm to entire application

TODO: SonarQube

Date Author Topic Decision
2023-06-21 Olle Haerstedt New question editor In the context of new question editor and styling
facing the problem of divering design between legacy admin and QE
we decided to remove Bootstrap
and neglected keeping and overriding Bootstrap
to achieve <pros>
accepting <possible drawback>
because <some reason> TODO
2022-11-06 Olle Haerstedt New question editor In the context of new question editor
facing the problem to choose a new tech stack
we decided for ReactJS + REST API
and neglected Vue, Angular, RPC, SOAP
to achieve a modern and common tech stack
accepting that we will have to rewrite the side-menu Vue component eventually, and write a new REST module
because side-menu is not that much code, and React is a more established framework (maybe?); and REST we need anyway to support more third-party integration
2021-11-05 Olle Haerstedt CI In the context of continuous integration,
facing lack of evidence-based code metrics,
we decided to apply a maximum function size of 60 LOC for new code,
accepting that some exceptions will likely occur over time,
but overall the code quality will improve by such a restriction considering recommendations from NASA code guide and research.
2021-06-16 Olle Haerstedt Fat models In the context of fat models,
facing lack of testability,
we decided to move code to dependency injected service classes
to achieve higher testability with unit-tests,
accepting that readability might be slightly reduced (circular coupling between new service classes and old models, at least temporarily),
because preventing regressions is more important right now.
2021-06-16 Olle Haerstedt New features In the context of new features,
facing high degree of regressions,
we decided to always add integrity tests to new features
to prevent regressions and ensure adherence to specification,
accepting that we might need to help external contributors with scaffolding and guidance, also accepting the extra development time,
because testing is one of the best ways to prevent regressions long term and as the software changes and is touched by multiple developers.
2020-01-01 Team Controller refactoring In the context of controller refactoring,
facing non-idiomatic Yii code (renderWrappedTemplate)
we decided for refactoring of controllers and neglected other technical debt,
to achieve idiomatic Yii code,
accepting a long-term effort,
because applying the Yii 1.1 framework correctly is important to lower the bar for new developers.