<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Martin Römer on Medium]]></title>
        <description><![CDATA[Stories by Martin Römer on Medium]]></description>
        <link>https://medium.com/@martin.roemer?source=rss-329e55abb77d------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*-KUbRgeokEGfcYU8_kk-tw.png</url>
            <title>Stories by Martin Römer on Medium</title>
            <link>https://medium.com/@martin.roemer?source=rss-329e55abb77d------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 04:22:28 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@martin.roemer/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[From Divergent Scrolls to a Single Grimoire: Embracing Trunk-Based Development]]></title>
            <link>https://medium.com/getaway-group-tech-blog/from-divergent-scrolls-to-a-single-grimoire-embracing-trunk-based-development-ec1a8928cd86?source=rss-329e55abb77d------2</link>
            <guid isPermaLink="false">https://medium.com/p/ec1a8928cd86</guid>
            <category><![CDATA[engineering-culture]]></category>
            <category><![CDATA[continuous-delivery]]></category>
            <category><![CDATA[devops]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[trunk-based-development]]></category>
            <dc:creator><![CDATA[Martin Römer]]></dc:creator>
            <pubDate>Tue, 15 Jul 2025 06:50:14 GMT</pubDate>
            <atom:updated>2025-07-15T06:50:14.839Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CWetGaoedJ5kPi_05LSkOw.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@joeykyber?utm_source=medium&amp;utm_medium=referral">Joey Kyber</a> on <a href="https://unsplash.com/photos/GPxgi4J82E4?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><blockquote>How we unified two release cultures, halved lead time, and made deployments boring — on purpose.</blockquote><h4>TL;DR</h4><ul><li>We merged two engineering teams that released code in completely different ways and were slowly shredding each other’s nerves.</li><li>A trunk-based workflow gave us a neutral, automation-friendly middle ground: code now ships <em>~several times per day</em> — but always through a staging snapshot and manager approval, never straight to prod.</li><li>Everyone stays in the loop.</li><li>The hardest part wasn’t YAML — it was aligning habits and expectations. Tooling just made the truce stick.</li></ul><h3>Why We Had to Change</h3><h4>Two Cultures, One Codebase</h4><p>When two travel-tech companies merged to form GetAway Group, their engineers arrived with laptops, enthusiasm, and release rituals that looked nothing alike. Overnight we had two tribes sharing multiple repositories:</p><p><em>Team A</em> believed in ceremony: two staging environments, gated promos, Jira tickets blessed by managers, and production only after moon-aligned merges.</p><p><em>Team B</em> ran on adrenaline: commit straight to prod - no CI safety net - and live-demo the feature before the hourglass emptied.</p><p>Great for folklore, terrible for collaboration. Pairing sessions started with <em>“Wait, which branch are you on?”</em> The same bug could take a day or a week to fix depending on whose pipeline it hit.</p><h4>Pain in Numbers (Ballpark)</h4><ul><li><em>About a full work-week</em> from first commit to prod in Team A; <em>same-day</em> for Team B.</li><li>Managers pampered Team A through every release, while Team B’s deployments sometimes went unnoticed for a day or two.</li></ul><p>We needed a single workflow — and it couldn’t feel like either side “won”.</p><h3>Life Before: Release Pipelines</h3><h4>Team A — “Double Staging Waterfall”</h4><ol><li>Branch off <strong>staging 1</strong>, open PR, wait for review.</li><li>Managers verify on <strong>staging 1</strong>.</li><li>Tickets promoted to <strong>staging 2</strong> for final integration.</li><li>Jira release created, <strong>staging 2</strong> merged to prod.</li></ol><p>Predictable? Yes. Fast? Not unless geological time is your benchmark. The ceremony made newbies anxious; hotfixes sometimes bypassed staging 1 — promoted directly to staging 2 or, if daring, straight to prod — but were frequently forgotten, causing drift in staging 1.</p><h4>Team B — “Push &amp; Pray”</h4><ol><li>Branch off prod (or sometimes not), code, merge.</li><li>No CI — features go straight live immediately.</li><li>Optional: refresh a staging box for external partners — often only on experimental builds; it might have been close to prod but was regularly out of sync.</li></ol><p>Velocity through the roof — until a typo shipped at 17:45 on Friday. Managers hated the surprise, and staging lived in another dimension that sometimes resembled prod, so testing in a “safe” place meant time-travel.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/635/1*RjlSCpWbMk95vwDgwYq9Yw.png" /></figure><h3>The Clash in the New Squads</h3><p>We reorganised into two squads mixing members from both legacy teams so every squad had representation from each company. Culture shock followed:</p><ul><li><strong>A-folks</strong> froze when asked to push directly to prod.</li><li><strong>B-folks</strong> grumbled when their PR sat in staging purgatory.</li><li>Nobody felt comfortable in the other world’s workflow, and complaints piled up over whose process to follow.</li></ul><blockquote><em>“Why do I need to dance at two staging weddings?” — an ex-Team B engineer<br>“I’m not confident putting unfamiliar code straight into the production cauldron.” — an ex-Team A engineer</em></blockquote><p>Morale dipped. Something had to give — preferably not prod.</p><h3>Decision: Trunk-Based Development</h3><p>We skipped exhaustive spreadsheets of pros &amp; cons. In a late-evening chat between an engineer and a DevOps lead (no divination potions required), they realised trunk-based ticked <em>all</em> the boxes. So we considered the following options:</p><p><strong>We needed</strong></p><ol><li>Single source of truth</li><li>Fast feedback</li><li>Manager sign-off</li><li>Compatibility with future per-branch environments</li></ol><p><strong>Trunk-Based development gave us</strong></p><ol><li>One main branch</li><li>Continuous integration on every merge</li><li>Release PR gate without manual cherry-picks</li><li>Staging as next-prod snapshot; spin-up previews later</li></ol><p>Neutral ground, minimal ceremony, automation-friendly.</p><h3>Implementation Cheat-Sheet</h3><p><strong>1. Rename default branch to</strong><strong>main</strong></p><ul><li>Renames everyone’s mental anchor and signals a cultural reset.</li><li>Breaks tiny ergonomics so engineers pause and notice the workflow shift.</li></ul><p><strong>2. Merge → Deploy pipeline</strong></p><ul><li>Merge to main triggers a deployment to staging (the next-prod snapshot with all new features).</li><li>A bot amends or creates a <strong>Release PR</strong> summarising changes since the last prod tag.</li></ul><p><strong>3. Release PR gate</strong></p><ul><li>Requires one manager approval.</li><li>Merge auto-tags the commit with a new semantic version (patch/minor/major based on PR labels) and pushes to prod.</li></ul><p><strong>4. Runbooks</strong></p><ul><li>A concise “How to Release” guide for engineers and managers — designed to refresh memories after a release hiatus (e.g. a holiday).</li></ul><p>Our pipeline now handles all the incantations, freeing engineers to focus on building.</p><h3>Day-in-the-Life Workflows</h3><h3>Engineer</h3><ol><li>Start work</li><li>git switch -c feature/instant-booking</li><li><em>hack hack hack</em></li><li>Open PR to main with labels for changelog.</li><li>After reviews are done: Press the shiny blue merge button.</li></ol><p>The pipeline deploys to staging, and the bot updates the Release PR.</p><h3>Manager</h3><ol><li>Ping — “Release PR #42 ready for review.”</li><li>Skim auto-generated changelog; coordinate if multiple teams touched the repo.</li><li>Click <strong>Approve</strong>.</li><li>Engineer merges; prod deploys automatically shortly after.</li></ol><p>Managers regain visibility without becoming bottlenecks; engineers keep their flow.</p><h3>Early Results (Ballpark)</h3><p>We’ve been trunk-based for a single sprint on our largest shared repository, so hard numbers are still forming. On another service that migrated first we’ve already seen:</p><ul><li>Deployment frequency jumping from <em>~once a week</em> to <em>~several times per day</em> when needed.</li><li>Lead time to prod <strong>roughly halved</strong>.</li><li>Hotfixes dropping from <em>a handful each sprint</em> to <em>almost none</em>.</li><li>Deployment confidence is now higher, and any engineer can perform deployments.</li></ul><p>We may never replace these ballparks with precise metrics, so consider them our initial impressions rather than promises.</p><h3>What’s Next</h3><ul><li><strong>Feature flags</strong> to replace the last long-lived epic branches.</li><li>Ephemeral preview environments per PR so managers can validate changes before merge.</li><li>Help any team migrating to trunk-based dev — runbooks, pairing, and the occasional pep talk.</li><li>Encourage immediate feedback when something feels uncomfortable so we can iterate on solutions quickly — no formal retros required.</li></ul><h3>Takeaways for Readers</h3><ol><li>Changing release workflows is 80% sociology, 20% YAML. Start with trust, finish with tooling.</li><li>A neutral new process defuses “your way vs my way” stalemates.</li><li>Automate the boring parts — changelogs, tags, staging syncs — so humans focus on code.</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ec1a8928cd86" width="1" height="1" alt=""><hr><p><a href="https://medium.com/getaway-group-tech-blog/from-divergent-scrolls-to-a-single-grimoire-embracing-trunk-based-development-ec1a8928cd86">From Divergent Scrolls to a Single Grimoire: Embracing Trunk-Based Development</a> was originally published in <a href="https://medium.com/getaway-group-tech-blog">GetAway Group Tech Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How we use quality tools at GetAway]]></title>
            <link>https://medium.com/getaway-group-tech-blog/how-we-use-quality-tools-at-getaway-bcd0698ef066?source=rss-329e55abb77d------2</link>
            <guid isPermaLink="false">https://medium.com/p/bcd0698ef066</guid>
            <category><![CDATA[web-development]]></category>
            <category><![CDATA[clean-code]]></category>
            <category><![CDATA[php]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[development]]></category>
            <dc:creator><![CDATA[Martin Römer]]></dc:creator>
            <pubDate>Mon, 17 Mar 2025 07:02:35 GMT</pubDate>
            <atom:updated>2025-03-17T07:02:35.722Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*rTXyiDzey_s30SIl" /><figcaption>Photo by <a href="https://unsplash.com/@benofthenorth?utm_source=medium&amp;utm_medium=referral">Ben Griffiths</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>We software engineers are a curious bunch of people. We strive to find great solutions to our problems, but we also enjoy creating lazy solutions. Especially when certain rules aren’t enforced by the programming language itself.</p><h3>How we got there</h3><p>PHP is one such language. Because it is dynamically typed, it already provides many opportunities for an engineer to be creative. Not only in legacy code, but in new code as well. We quickly realized this when we started working as a team of three engineers, each with a unique style.</p><p>In order to reduce the amount of discussion during code reviews, we realized that we needed a solution that could help us with these issues, while introducing a bit of type safety along the way.</p><p>Since we like to take the path of least resistance, it is necessary that these checks actively help us and prevent us from producing suboptimal solutions.</p><h3>How we started out</h3><p>So we started looking for tools that can aid us with various problems that came up repeatedly in code reviews. Mostly typing and type safety checks, code flow and various kinds of formatting.</p><p>Enter <a href="https://github.com/squizlabs/PHP_CodeSniffer">PHP_CodeSniffer</a>, <a href="https://phpstan.org/">PHPStan</a> and <a href="https://editorconfig.org/">.editorconfig</a>.</p><h4>PHP_CodeSniffer</h4><p>PHP_CodeSniffer has several roles. Using the excellent rules from <a href="https://github.com/slevomat/coding-standard">slevomat</a>, we chose what we thought would help. If we didn’t like a rule, we tweaked it or removed it. If we noticed that something was missing, we added it.</p><p>The great thing about most of these rules is that they have the ability to fix themselves with phpcbf.</p><p>A small selection of things CodeSniffer does for us:</p><ul><li>remove excess whitespace</li><li>number of empty lines (e.g. never more than 1, always 1 before return/continue/break, always 1 between methods)</li><li>early returns instead of if {} else {} constructs</li><li>trailing commas for multiline definitions</li><li>Order of<br>- attributes<br>- class structure<br>- doc comments</li><li>Lengths<br>- max number of character per line<br>- max number of lines per method<br>- max number of lines per class</li><li>single/multiple line arguments for array/functions depending on total line length</li><li>force arrow and static functions</li><li>and many more</li></ul><h4>PHPStan</h4><p>Getting started with PHPStan can be a daunting experience. Things that are <em>obviously not null</em> now have to be <em>explicitly not null</em>. To get used to phpstan, we started at a lower level and increased the level whenever we felt comfortable until we reached the final level — no mixed and explicit array documentation.</p><p>With its extension ecosystem, phpstan is very customizable. With its base functionality we ensure type safety, array documentation and enforce exception handling.</p><p>Currently we use these extensions:</p><ul><li>symfony: for all things symfony</li><li>doctrine: for checking that dql queries have proper type hinting</li><li>phpstan-disallowed-calls: disable certain php calls. e.g.: dump(), compact(), json_encode(), json_decode()</li><li>phpstan-strict-rules: enforce type safety for functions like array_search(), disallow use of !, …</li><li>phpstan-deprecation-rules: disallow usage of anything marked as deprecated</li><li>sfp-phpstan-psr-log: check for error log message format</li><li>phpstan-rules: no magic numbers / magic strings</li></ul><h4>.editorconfig</h4><p>This is used to enforce the same formatting rules in the editor of our choice.</p><p>These are mostly cosmetic and are not yet enforced in any checks.</p><h3>How it’s implemented</h3><p>Initially we faced three problems</p><ol><li>Making the checks available to multiple repositories</li><li>Ignore old and existing errors</li><li>Run them automatically</li></ol><p>The solutions to all of these were surprisingly simple</p><h4>Making the checks available to multiple repositories</h4><p>We already had a workflow for creating and using custom packages, so we could just use that.</p><p>In addition, all of the tools give us the ability to extend their configuration. So we can define a basic configuration depending on the repository and extend and set settings as needed.</p><h4>Ignore old and existing errors</h4><p>Phpstan has built-in baseline support. For Php_CodeSniffer we use <a href="https://github.com/123inkt/php-codesniffer-baseline">php-codesniffer-baseline</a>.</p><p>Creating initial baselines allows us to fix only what we touch. In the beginning, we often had changes bubbling through many classes — for these cases, we carefully judged how much was in scope and what should go into the baseline.</p><h4>Run them automatically</h4><p>When we implemented the checks, we agreed not to put any more strain on the already fragile build server.</p><p>Since we were a small team, we decided to trust each other. Since we were also forgetful, we still needed these checks to run automatically. So we decided to use pre-commit hooks, which we would set via symlinks after installing the package. This worked surprisingly well.</p><p>We now have a testing pipeline in place, but still use pre-commit hooks to catch bugs as early as possible.</p><h3>Which challenges did we face during implementation?</h3><p>As mentioned in <em>Ignore old and existing errors</em>, changes to existing method signatures can bubble up through many classes, especially with array definitions. Finding a good balance between fixing what is needed and not refactoring multiple classes was something we learned over time and experience.</p><p>Some reported problems we did not want to fix at all, such as line length for long string constants. So we decided to ignore certain errors as precisely as possible, by disabling only a single line or a short block, and always adding the specific rule to be ignored.</p><p>Over time, we also increased the level of phpstan rules to include array definitions, which can be quite large and would involve a lot of copying and pasting (e.g., every assignment, method signature, and call). Eventually we stumbled upon @phpstan-type and use it whenever we can’t convert arrays to DTO’s.</p><p>Occasionally we have also had merge conflicts regarding the baseline, which can often be resolved by taking a closer look at the errors and regenerating the baseline.</p><h3>How have the tools improved the dev workflow?</h3><p>Understanding unknown pieces of code becomes easier because we already know what structure to expect.</p><p>No one needs to keep a mental model of how we agreed to do things because the quality tools take care of that and remind us when we run them. Running a rule multiple times will quickly trigger a learning effect and its use will become second nature.</p><p>Code reviews can now be more concise and focused on actual bugs. Code style issues can be discussed separately and fixed at once.</p><p>Thanks to phpstan, we also realized how different it is to have a tool that shows what checks are missing. From missing null checks, to accessing undefined array keys, to working with incompatible types (float vs int), to exceptions that aren’t caught or declared.</p><p>At some point we decided to try out working with yoda case and forcing explicit false checks instead of using the ! operator. While this is certainly controversial, we have learned that explicit false checks make conditions more readable and less ambiguous, resulting in fewer bugs.</p><h3>What do we recommend teams in the same situation?</h3><p>Create a selection of rules and start using them. Don’t shy away from changing them if a rule doesn’t work. Automate recurrent pain points if possible and use them.</p><p>As for phpstan, don’t start on the highest level straight away. Start a few levels lower and go to the next level whenever you are comfortable on the current one, but keep pushing to reach the highest level.</p><p>Running into problems with a specific tool or rule, look into it’s documentation or look at it with a colleague. This can lead to new discoveries on how to use it and propel it usage even more.</p><h3>How we’ll continue</h3><p>We keep a close eye on new releases of phpcs and phpstan and try to implement new changes as quickly as possible. Recently we were introduced to <a href="https://github.com/phparkitect/arkitect">PHPArkitect</a> and <a href="https://github.com/editorconfig-checker/editorconfig-checker">editorconfig-checker</a>, which we plan to integrate at some point.</p><p>Maybe one day PHP will have a nice default like gofmt or rustfmt, but until then we will continue to improve our existing tools and continiously refine our current ruleset as needed.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bcd0698ef066" width="1" height="1" alt=""><hr><p><a href="https://medium.com/getaway-group-tech-blog/how-we-use-quality-tools-at-getaway-bcd0698ef066">How we use quality tools at GetAway</a> was originally published in <a href="https://medium.com/getaway-group-tech-blog">GetAway Group Tech Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>