Trendyol and Tuist: Engineering Apps at Scale
Dive into our exclusive chat with Trendyol, Turkey's e-commerce giant. Explore how they leverage Tuist for expansive iOS development, unravel their tools, team dynamics, and the secrets to managing tech at scale. A must-read for all developers
Interviewee: Trendyol
In the bustling landscape of Turkey’s e-commerce, Trendyol stands as a beacon. As the nation’s foremost e-commerce platform, Trendyol isn’t simply about online shopping. It represents an entire digital ecosystem, a “super app,” dedicated to providing users with an unparalleled, diverse range of services. But, beneath this massive digital edifice, there’s an intricate dance of technology, teamwork, and tools. And one tool that has been instrumental for Trendyol is Tuist. We recently sat down with the developers at Trendyol to gain insights into how they use Tuist to meet the challenges of developing apps on such a monumental scale.
Our mobile team is composed of a talented and dedicated group of 140 professionals who work collectively on both iOS and Android platforms. As a dynamic and innovative organization, Trendyol operates as a super app, offering a diverse range of services through various omnichannels on the platform. To ensure efficient management and specialization, each omnichannel has its own dedicated domain teams. This organizational structure enables us to effectively cater to the unique needs and demands of each omnichannel, fostering a cohesive and agile work environment.
As the platform team, we provide services to five different channels and support ten distinct teams. Our primary focus is to enhance the developer experience, tooling, and performance for these teams. We consider each team as our valued client, working diligently to meet their specific needs and align with their interests in these areas.
We utilize the VIPER architecture in our app, which is composed of View, Interactor, Presenter, Entity, and Router components.
We chose VIPER for its strong alignment with SOLID principles and its modular design. This ensures a clean, maintainable codebase and makes unit testing more straightforward. VIPER’s scalability also makes it easier to onboard new developers who can focus on specific components.
On a practical level, VIPER promotes parallel development. Each developer can specialize in certain components, simplifying code reviews. While there’s an initial learning curve and some boilerplate, we find these are outweighed by long-term benefits like maintainability and scalability. We even have custom Xcode templates to speed up development.
We recently shifted our deployment target to iOS 14. This move opened up the opportunity to integrate SwiftUI into our project. The motivation behind this was multi-fold:
As technologies continue to mature and new architectural patterns emerge, we’re always open to evaluating them. Whether it’s new patterns that mesh well with SwiftUI, or even entirely new paradigms, our team is excited by the potential to improve and optimize.
In summary, while VIPER has served us well, our architecture is not static. It’s influenced by our commitment to best practices, the evolving tech landscape, and our continual desire to deliver the best possible product.
Before using Tuist, we encountered the following problems:
By adopting Tuist, we effectively tackled numerous challenges within our development process.
We have fully integrated Tuist into our development workflow, and the impact has been transformative. Tuist has enabled us to eliminate the need to commit Xcode project and workspace files to version control by utilizing its runtime generation capabilities. This has not only mitigated the risk of merge conflicts but also provided us with greater flexibility in managing project files.
Moreover, Tuist has played a crucial role in ensuring consistency across projects by eliminating the manual configuration of Xcode projects. Its streamlined and intuitive approach to configuration has enabled us to effortlessly enforce best practices and maintain uniformity throughout our codebase.
Additionally, Tuist has simplified the management of Xcode projects and workspaces, significantly reducing complexity. This has allowed us to focus more on development, sparing us the intricacies of project setup tasks.
One of the most valuable features of Tuist has been its templates, which have boosted our productivity. With Tuist’s templates, we can swiftly establish new modules with standardized structures and configurations, saving considerable time and effort.
Furthermore, our overall developer experience has improved significantly as Tuist minimized the number of manual steps required for project configuration and setup. Its efficient and automated processes have streamlined our workflow, empowering us to concentrate on coding and enhancing productivity.
In summary, Tuist has been instrumental in addressing the challenges we previously faced in Xcode project management, dependency handling, consistency, productivity, and developer experience. Its seamless integration into our development process has revolutionized our project execution, leading to smoother and more efficient development.
We have integrated the use of templates to optimize our development process. A particularly noteworthy template that we employ is engineered to generate new modules. This template facilitates the creation of a new Xcode project with customizable parameters, such as its name and path. The resulting project is automatically configured with essential targets, including Interface, Implementation, Test Support, and Tests. This efficient method not only accelerates the initiation of new modules but also guarantees consistency across our projects. It is an invaluable instrument that harmonizes with our development practices and aids us in upholding a superior standard of code quality.
The clean build process takes approximately 295 seconds, whereas the incremental build takes around 40 seconds. We are managing a substantial total of 230 .xcodeproj modules within our project.
The clean build involves recompiling the entire project from scratch, which accounts for the longer duration, whereas the incremental build only compiles the modified or newly added files, resulting in a significantly faster build time.
Given the large number of modules, we have managed to optimize our build times considerably by utilizing incremental builds. This enhancement in build efficiency has had a positive impact on our development workflow, enabling us to devote more time to coding and reducing turnaround time during development iterations.
We segregated concrete modules from interface modules, enabling us to maximize the benefits of incremental building. In this setup, we employ our own dependency container. Concrete modules are not directly interconnected; rather, interface modules serve as intermediaries between them. Consequently, Xcode only compiles the modified sections, thereby minimizing our build time.
We follow a structured release cadence to guarantee the prompt delivery of packages. Our release cycle is bi-weekly, with predetermined release dates set for each cycle. In preparation for these releases, we initiate the process by creating a Release Candidate (RC) branch a few days before the scheduled release date.
The RC branch functions as a dedicated space for finalizing the upcoming release. Once the RC branch is established, our Quality Assurance (QA) team takes the helm and begins regression testing. This rigorous testing phase enables us to comprehensively evaluate the stability and functionality of the package before its live deployment.
Adhering to this approach allows us to confidently ensure the reliability of our releases and minimize the risk of unexpected issues arising. Our well-defined release cadence, complemented by the proactive creation of the RC branch and thorough QA regression tests, empowers us to consistently achieve our bi-weekly release targets while upholding a high standard of software quality.
Our testing strategy encompasses various types of tests, executed intentionally in different scenarios and, when necessary, capable of blocking pipelines. To discuss this quantitatively:
We have over 25k Unit, 1.5k Regression, 250 Smoke Tests, and 500 Snapshots. Both smoke and unit tests are executed on every commit and act as blockers on merge requests.
Unit Tests are indispensable. For many years, it has been mandatory to write and maintain tests for all developments. Recently, we have been using tools (explained in the tools section) that automatically generate these tests, speeding up the process. However, it is crucial to maintain a balance between automation and manual effort to prevent the tests from being neglected, much like unread documents. We continuously fine-tune this balance.
Our UI Testing framework, built on XCUITest, is highly advanced. A significant portion of the code is auto-generated by our internal code generators. Our internal mock server allows us to easily mock the desired services. This setup simplifies the process of writing smoke tests for basic flows during development.
Smoke Tests were initially written to ensure that every page and element worked flawlessly before writing complex regression tests in the production environment. Another purpose was to confirm that the UI test infrastructure wasn’t broken by the developments, by adding these tests as a merge check at an early stage. We quickly achieved close to 100% coverage, and it is now a must to add these tests when new pages are written.
Regression Tests operate in the production environment using real test data and leverage the same architecture created for our smoke tests. These tests ensure existing functionality remains untouched with new updates. This approach not only assures quality but also accelerates our release cycle.
We use Snapshot Tests differently from many teams. Instead of using a unit test style, we use our existing UI Test framework to handle actions and flows while we use the swift-snapshot-testing library to take and compare the screenshots. Architectural reasons and considerations around maintenance responsibilities led us to choose this path. We see it as a low-maintenance tool that effectively catches many bugs, and we share all test results via buttons on Slack.
Lastly, we also have a small number of Integration tests written for our in-house SDK(s). End-to-end testing of certain systems, such as events, can be quite difficult and challenging. In this context, these tests grant us a significant amount of confidence in our systems.
In our pursuit to streamline our development processes and address the unique challenges we face at Trendyol, we’ve invested time and effort in developing several in-house tools. These tools are designed to cater to our specific needs, and have significantly contributed to our efficiency and productivity in developing iOS apps at scale. Here is a list of some of the in-house tools we have developed:
Mockolo is a powerful tool that generates all the necessary mocks for a protocol, conforming to it with just a single button press. It goes a step further by supporting the new async-await structure. Developed on top of Uber’s Mockolo library, it has been tailored to fit the team’s needs and optimized for SwiftKit. The mock output has been fine-tuned to align seamlessly with the project requirements, making it an invaluable asset for the team’s testing and development processes. With Mockolo, generating mocks and staying up-to-date with new Swift features has never been easier.
LatiFlex is an efficient in-app debugging tool designed to streamline our debugging process. With its powerful features, we can easily track network requests, search in network responses, copy Curl commands and responses, monitor analytic events, search events, copy event parameters, execute deeplinks within our application, and seamlessly switch between different environments. LatiFlex simplifies debugging, making it faster and more effective.
MockServer is a powerful request mocking tool that offers flexibility and ease of use. With its user-friendly features, you can effortlessly save and update requests for future use. The tool provides various filtering options such as location (all, path, query, scenario, method, status) to customize your mocking experience. Additionally, you can conveniently list all mocking requests and apply search filters to find specific requests efficiently. MockServer is your go-to solution for seamless and hassle-free request mocking.
GitLab Menu Bar is a handy and feature-rich menubar application designed for an optimized GitLab experience. The application lists merge requests with the user’s approval count, build status, and conflict mark. It also displays merge requests that need to be reviewed, providing date and approval status for easy sorting. The “Create MR” button generates merge requests based on the user’s recently pushed branches. In case of a failed pipeline, the app offers the option to trigger the pipeline again. Additionally, users can switch between branches and create new ones using the macOS Shortcuts app integration. Conflicted files can be opened with Visual Studio Code using the macOS Shortcuts app. The application streamlines the process of sending export IPA requests. For the mobile team, the QA Mode showcases automation tool export options and task status, requiring the board ID. GitLab Menu Bar aims to provide users with convenience and productivity, empowering them to enhance their GitLab workflow. When the user presses the merge button, a confirmation dialog appears, prompting the user to review his actions before proceeding. Creating a new merge request becomes effortless as GitLab automatically formats the merge request name and description for the user. Additionally, the branch name is automatically set as the merge request name, ensuring consistency in identification. For a comprehensive overview, all commit messages are aggregated and set as the merge request description.
A Tuist plugin that analyzes explicit dependencies between modules by using SwiftSyntax.
We developed a tool called SwityTestGenerator to simplify the process of writing UI tests, which previously required setting accessibility identifiers for each IBOutlet individually. This process was repetitive and manual. With SwiftyTestGenerator, we can easily create the necessary elements for UI testing and have made it available for use by all team members.
Collaboration is a cornerstone of our team’s success, and we have established a well-defined process to ensure seamless communication and coordination among various roles, including developers, designers, and product managers.
By nurturing a collaborative environment and involving different roles throughout the development lifecycle, we are able to leverage the unique perspectives and expertise of each team member. This holistic approach results in more well-rounded and user-centric solutions, as well as a stronger sense of ownership and shared accomplishment among our team members.
We hope this insight into Trendyol’s engineering practices for developing iOS apps at scale has been enlightening. Their adept use of Tuist, among other powerful tools and collaborative practices, not only streamlines their development process but also ensures high code quality and consistency across projects. By fostering a collaborative environment, maintaining open lines of communication, and implementing comprehensive testing strategies, the Trendyol team successfully navigates the challenges associated with managing a large number of modules and delivering reliable, user-centric solutions.
We would like to express our gratitude to the engineers at Trendyol for sharing their valuable experiences and strategies. Their commitment to continuous improvement and innovation is truly inspiring. We believe that sharing knowledge and best practices like this contributes to the growth and success of the broader developer community.
Thank you for reading, and we hope you found this interview as informative and inspiring as we did. If you are interested in learning more about Tuist and how it can help streamline your development process, please visit our website or check out our documentation for more details.
For more insights into engineering practices and technical articles, be sure to visit Trendyol Tech’s Medium blog and our additional resources.
