Development Collaboration and Good Practices
Working with a large number of development teams across many projects and organizations we found that it is useful to collect some of our insights. Some of those are related to AEM, but the majority are related to general purpose frontend development or are just general guidelines on how to collaborate in a team of developers.
You may read some of those items and think that it is generally understood as common sense amongst developers. We agree, and that’s a great sign that you are ready to work in a collaborative way on AEM projects together with other developers.
At this point this is just a collection of lessons learned from our engagements on a growing set of projects.
Create Pull Requests
If you work on a project with multiple developers it is rarely a good idea to push directly to
main. When your project is in production code changes that are merged or pushed to
main often means that they are released to production. Protecting your
main branch is a good mechanism to ensure that people don’t push to
main by accident, which is especially advisable with a site that is in production. Keep the scope of your Pull Request to what’s in the title / description of the PR.
Pull Request etiquette
If you open a pull request, make sure that you include a URL to a page (or a number of links to pages) on your branch, where the reviewer can see your code in action. If you are updating code of an existing block, make sure to include the link that features the block you are updating as the reviewer may not know where this block is in use to test its functionality.
AEM Github Bot
The AEM Github Bot and standard Boilerplate project setup will check linting of your CSS/JS and PageSpeed Insights for Web Performance, Accessibility, SEO and Best Practices. Do not submit a Pull Request for Review that has linting errors or doesn’t have a Lighthouse Score from Page Speed Insights of 100.
It is good practice to have your code reviewed by the maintainer or main developer of the project you are working on. For the reviewer to see a realistic PageSpeedInsight performance score, include links to the cached
If you are a maintainer or lead developer on a project that is not in production and you are adding code or are changing your own code, it sometimes is unnecessary overhead to have your PRs reviewed by peers and it is often ok to merge your own PRs without a peer review.
Merging Pull Requests
Merge your own pull requests only. If the person who opened the Pull Request has the ability to merge their own Pull Request, the author of the code is the ideal person to merge. There are situations where the author specifically states that this can and should be merged by someone else, and in those cases a maintainer (main developer) of the project should feel free to merge a Pull Request.
Even if a Pull Request is approved, you should always check with the author of the Pull Request if they are ready to merge.
For work still in progress, opening A Draft Pull Request also helps prevent accidental merging.
It is good practice to consider branches for individual Pull Requests as initially a private location of the developer that created the branch. Do not just push into other developers’ branches without having been invited to do so. There are situations where people are collaborating on Pull Requests but it should be an explicit agreement.
(Scaled) Trunk-based Development
AEM works great in a scaled trunk-based development with built-in devops and CI/CD. Meaning you merge a large number of small pull requests into main (production) but the Quality Assurance / Review efforts can be tailored to the impact of a small change. Nobody wants to review and test large pull requests, and long-lived branches with lots of changes tend to be difficult (and dangerous) to merge.
Don’t change the linting configurations (eslint airbnb-base and stylelint) unless you have a really good reason. Personal preference is not a good reason. Changing linting makes it very hard to move developers from project to project. Arguing if something is a good reason to change the linting configuration is usually a lot more effort than just categorically saying no.
CSS Selector Block Isolation
AEM blocks most often operate as components collaboratively in the same DOM / CSSOM. This means that you should write your CSS selectors in a block
.css in a way that isolates your CSS from impacting layout of elements outside your block. The easiest way to do this is by making sure that every CSS selector in the
.css of a block only applies to the block.
Cascade in CSS
Use your CSS classnames wisely. Some CSS classes and variables are used across different blocks, and others are not expected to be used outside your block. Prefixing classes and variables that are private to your block with the block name is good practice. Conversely if there are CSS classes and CSS context that should be inherited (often those can be authored) those classes and variables should not be prefixed.
CSS indentation and property order
Outside of a CSS refactoring PR, don’t change sequencing on properties or the indentation across the CSS files you touch in a functional PR. Every developer has different preferences on sort order of properties or indentation. Make sure the diff that you see in your PR is isolated to the changes you actually want reviewed before submitting it.
CSS Selectors Complexity
Don’t let your CSS Selectors become complex and unreadable. Often it is better to decorate additional CSS classes / Elements onto your DOM and write readable CSS instead. Complex CSS selectors also often are harder to maintain and more brittle than the equivalents in JS.
Naming your classes simple and intuitive is helpful for other developers. Avoid namespacing unless it is necessary within the scope of a project. There is often no need to specify the type or the origin (eg. the name of your design system) of a CSS variable that is to be used across the entire project.
Leverage ARIA Attributes for styling
In many situations you will add ARIA Attributes for accessibility. Since those have well defined semantics like
hidden that are understood by all developers, in most cases there is really no need to come up with additional classes in your vocabulary that have unknown semantics.
Generally Web projects should be developed “Mobile First”. This means that your CSS without media query should render a mobile site. Add media queries to extend the layout for tablet and desktop.
1200px as your breakpoints, all
min-width. Don’t mix
max-width. Only use different breakpoints in exceptional cases where you have good understanding why that’s needed.
Less, Sass, PostCSS, Tailwind and friends.
If you are working in the context of a bigger organization, make sure that you don’t introduce a dependency to any CSS preprocessor or Framework of your personal preference without getting the buy-in from the entire team and organization. As there are a lot of differing personal preferences in this area, it makes code hard to maintain if every project or every block inside a project uses different technologies.
The simplest solution is to rely on the growing CSS featureset which is supported by the browsers.
Modern CSS features
Make sure the features you are using are well supported by evergreen browsers. Depending on the features more or less pervasive support may be acceptable.
On most web sites, frameworks are overkill for simple layout problems outside of application-like functionality. Frameworks often introduce web performance issues (Lighthouse and Core Web Vitals), particularly if they are in the pathway of the
LCP or introduce
TBT, while trying to address trivial problems. Keep simple things simple.
The simplest solution is to rely on the growing featureset supported by browsers.
Build tool chain
Differing build tool chains from project to project make it hard for new developers to be onboarded and often introduce additional complexity. Make sure that you don’t introduce a dependency of your personal preference without getting the buy-in from the entire team and organization.
The simplest solution is to keep the entire project build-less.
Make sure the features you are using are well supported by evergreen browsers. Depending on the features more or less pervasive support may be acceptable. While AEM can be used with any browser,
lib-franklin.js has a dependency on browsers that support dynamic
import(). Any feature that is supported by the set of browsers that support dynamic
import() should be considered safe. Technically of course older browsers (eg. IE) can be supported by AEM projects, but those require heavy customization.
Not all features have the same consequences if a browser doesn’t support them, some may be cosmetic and others may stop the site from working. A common example is “optional chaining”. If a browser doesn’t support “optional chaining”, a single usage can have fatal consequences for the entire page.
Loading 3rd party libraries
Don’t add 3rd party libraries to your
head.html as they will be in the critical path of loading content above the fold and will often be loaded when they are not needed. Add the dependencies where needed via
loadScript() to the specific block that has the corresponding requirement.
In case of larger 3rd party libraries, you may even want to consider using an
IntersectionObserver to make sure you only load them when the block depending on it is actually being scrolled into view.
<head> that is delivered from the server as part of the HTML markup should remain minimal and free of marketing technology like Adobe Web SDK, Google Tag Manager or other 3rd party scripts due to performance impacts. Adding inline scripts or styles to
<head> is also not advisable for performance and code management reasons.
Content in Sharepoint / Google Drive
Start your development with Content
Before writing a line of code, create your (sample) content in a Word or Google Doc (or spreadsheet). Make sure that it feels good for authoring and share it with people on your team that have experience supporting authors. It requires support experience to understand what content structures are easy for authors to understand and recreate. Once you have settled on a content structure that contains all elements you need for your block, and have it reviewed you can get started developing your CSS and JS code.
The content lifecycle is very different from the code life cycle. If you are proposing changes to an existing content structure in your code or come up with a new block, don’t just make those changes on the page you are working on. Copy the page into your
/drafts/<yourname>/ folder and make changes there.
Once your code changes are merged to
main , you can have authors copy or merge your content with the content outside of your
Backwards compatibility of new features
Especially in production environments it is important to keep your anticipated changes to the content structure backwards compatible with existing content. Ideally code that is being merged should not have an impact on the website or require refactoring the content. Only when new content is put in place through a preview and publish cycle, the new functionality becomes available. This of course doesn’t apply to things like design changes across the existing content or functional bug fixes.
Use Content for “static” resources
Generally, it is not a good idea to commit binaries into your GitHub repo. Even text-based static resources for example HTML files or SVGs should only be put into GitHub in exceptional cases. A good reason to add an SVG to your git repo is if it is referenced from code. Don’t commit anything that is related to the content authoring process, or could be a part of an authoring process. There are some exceptions (usually related to legacy and non-browser clients) that require a certain set of fixed resources that cannot be produced and manipulated dynamically by AEM, but in general if you find a large set of static resources (eg. images, etc.) or an HTML file in a PR it is most likely not a good practice.
Use Content for Strings / Literals
Strings that are displayed to end-users and could possibly be translated or changed at some point should always be authorable and sourced from content (eg.