Versioning and releasing larger Chrome extensions

An approach to plan and implement a robust versioning scheme and release schedule – without Semver

17 minute read


I’ve been developing Chrome extensions for the last few years, and whilst most don’t need a lengthy development phase, if you want to develop alpha or beta versions, the limitations of Chrome’s manifest version (opens new window) (in comparison to Semver (opens new window)’s more flexible schema) can trip you up.

This article is somewhat of a retrospective of my journey through poorly-executed versioning, awkward naming, tagging and releases, to final robust versioning strategy whilst developing Control Space (opens new window).

Chrome extension version format


The version string (opens new window) in Chrome extensions may contain 1 to 4 integers. Most commonly this might be:


It cannot contain text so indicating -alpha or -beta releases is problematic.

As a fallback, you can use a “version name” but that doesn’t fix the numerical conundrum:

  "version": "",
  "version_name": "Control Space 1.0 (Beta 4)"

For those unfamiliar with the build unit, it is intended to be a number that is incremented each time there is a build. Ideally it is set in CI and is used to reference a snapshot of a particular state of the code, no matter the version.

Prerelease numbering

Numbering pre-releases in Chrome can leave you high and dry if you plan badly, or misjudge your launch timeline.

If you make the mistake of setting 1.0 too early and have published on the Chrome Web Store, you can’t replace it with a revised, lower version because the Web Store won’t accept it.

As such, if your aim is a neat-and-tidy 1.0 release (more on this later) you’re going to have to stick to 0.x.y versions. If you screw up and go to 1.0 too early you then may need to hack the patch or build units (like I did for too long):

Semver Chrome Chrome (hack) Work
0.0.0 Initial commit
0.1.0 Prerelease work
1.0.0-alpha Begin work on Version 1
1.0.0 Release Version 1

The problems of thinking in Semver

If you’re thinking in terms of Semver, the lack of prerelease labels has drawbacks:

  • you can’t identify “beta” versions i.e. 1.0.0-beta-05
  • you can’t move from alpha to beta or rc as the format doesn’t support it
  • you may have to abuse minor, patch or build units to achieve your goals


  • version strings for new features may always be out of date:
    • let’s say you’ve released with manifest version 1.0.0
    • you start working and committing code for the 2.0.0 version
    • as there’s no way to specify 2.0.0-beta the manifest states 1.0.0 although the code is technically 2.0-ish
  • Git tags are ambiguous:
    • when I was thinking in terms of a protracted “beta” phase, I used a mix of Chrome x.y.z.0 manifest versions and Semver vX.Y.Z.beta-N Git tags, so up until some theoretical 1.1.0 release, the two would never align
  • it’s confusing for everyone:
    • tagging releases, generating release notes and creating zips was an ongoing headache as there was no clear system on how the incompatible formats were supposed to marry-up

Versioning for product releases

Where versioning matters

To give some context to the rest of the article, let’s review the all versioning touchpoints.

There are various terms and values to keep track of:

  • manifest version, e.g. 1.0.0.n
  • potential manifest version_name, e.g. Control Space 1.0
  • Git tags, e.g. v1.0.0

Which may be used in the following places:

  • Chrome extension manifest
  • local build scripts
  • zipped extension files (sent to testers or uploaded to the store)
  • one or more unzipped extension folders used for local browser testing
  • changelog
  • release notes
  • GitHub issues
  • marketing communications
  • Chrome web store page
  • extension options page
  • support tickets

As you can see, it’s critical to settle on a straightforward and robust system that needs no additional interpretation.

Stepping outside the Semver box

I spent a good few days researching, strategizing and testing ways to move to a new versioning paradigm and build pipeline, and ultimately realised that I had boxed myself in thinking in Semver terms in a Chrome Versioned world.

The epiphany was realising that focusing on major and beta monikers (aka “breaking changes” and “pre-releases”) was missing the point as I was building a product not a library; fixating on how to satisfy these library-oriented constraints was the cause of my product-versioning woes:

  • there isn’t really a concept of a “breaking change" in most products:

    • using a major version bump as a signal to “review before moving forwards” means nothing to users
    • factoring-out the use of major checkpoints misses an opportunity to communicate additional information
  • the idea of “alpha” and “beta” doesn’t make sense; builds are either:

    • private, i.e. a zip file with testers
    • public, i.e. the extension on the web store
  • juggling named versions and .beta-nn monikers is confusing:

    • for end users it’s the major version that matters, i.e. “do I have the newest features?”
    • for beta testers it’s the build, i.e. “is there something for me to review?”

The other big issue was that being in (air-quotes) “beta” for so long, there was no incentive to publicly release until I felt the product was (air-quotes) “one-dot-oh” ready. What I should have done was concentrate on regular stable releases, and communicated properly what each release was about.

Bringing it all together

Versioning scheme

Given the above, I concluded:

  • counting down to 1.0 (i.e. alpha, beta) serves no practical purpose
  • there is no such thing as a beta-nn, there are just “sprints” and releases
  • major versions would be better-used to represent “product goals”, rather than “breaking changes”
  • without major version constraints, minor versions (i.e. features) have more meaning within the release
  • features will be released only when it makes sense, so an initial release could be x.5.2 (vs an aligned x.0.0)
  • there is no requirement to consider beta private and n.0.0 public; it’s either with testers or it’s published
  • the build unit identifies which release is the most recent, no matter what the other units
  • using a manifest version_name offers no real value – and worse – hides the real version string

So for a project like Control Space (opens new window) how does that translate?

Unit Purpose Example
major Sprint Improve window management
minor Feature Add context menu to window header
patch Fix (on main) Fix Settings panel CSS bug
build PRs / test builds Send build to testers to try out new context menu

I decided I wanted to move to this new versioning scheme immediately:

  • what had been (1.0 Beta 16) would jump to 16.0.0
  • individual features would now have proper minor homes in 16.1, 16.2, etc
  • fixes could now be properly recognised with patch releases, i.e. 15.0.1
  • the ever-incrementing build identifier would remove ambiguity from Web Store submissions

Branching strategy

The reality of moving away from a Beta-oriented (i.e. private) workflow to more regular public releases and product sprint cycles has made me decide to switch from trunk-based (opens new window) development to Gitflow (opens new window).

Though it might be seen as unfashionable in the era of CI, I think this should allow me to patch the currently-released major version more easily; main is the released version (15.x) whilst develop is the current sprint (16.x).

I’m no Git expert but as there is only me on the project, I don’t foresee this posing any problems.

Release strategy

Being in a never-ending Beta means that often, projects don’t end; it’s just a constant feeding of issues through the pipe, piling yet more tickets into a hungry GitHub Projects “Todo” column.

Moving forwards, I think it should be much easier to plan sprints and identify what goes in and what gets left.

This should hopefully result in clearer product releases where a single feature-set is executed on, and the communication and marketing around that should be simpler as well, for example:

August 2023

Control Space – Sprint 16

Polished setup, help and first-use experience ready for public release

Updating versions

Updating the major version

As part of trying to tally a version with a sprint, I am experimenting with updating the major version at the start of the sprint rather than the release of the feature.

Taking the “Improve window management” project example above, I might update the manifest from 15.1.0 to 16.0.0 as soon as I create the branch. This means that features should at least show the correct major version (aligning to the current sprint) rather than being linked to 15.1.0 as they might be otherwise.

Additionally, as there’s no longer a requirement to plot intercept courses to x.0.0 versions, features will be released as it makes sense to release them, so the first public release for a sprint branch may be something like 16.3.0, with additional minor releases / features following as they are completed.

Updating the build version

For clarity up until now I have omitted the build version, but in this new paradigm, we might start with: (number of closed issues)

I’ve created a script to bump version units, which I can run manually as I commit and push features, as well as hook it up to a GitHub Action which bumps the build each time a PR is merged.

Additionally, I could run multiple local builds for testers, so the final versions for the 16.x builds might look like:

Even though only the final 16.3.0.n version might be merged.

Fixing bugs and updating the patch version

Fixing bugs should be fairly straightforward by branching off main.

The main thing to check is that build numbers between branches are reconciled:

  • when committing the fix in main, ensurebuild matches the latest in feature
  • when merging the fix to feature, ensure the latest build is used

Though, if you forget it doesn’t matter as the overall version strings should always be different.

Note that the versioning for a theoretical CSS fix in the live 15.x branch bug might go like this:

Branch Version Work Change
main Feature 15 is live -
feature Start “Window Management” project major
feature Add “Context menu” feature minor build
main Fix some CSS issue (update build to latest) patch build and release
feature Pull in the fix (ensure latest build) -
feature Add feature minor build
feature Ready to release build and release

The final releases would be:

  • on main
  • which adds the CSS fix
  • which adds the the 16.x branch along with the fix

And the incrementing build should ensure identical versions are never uploaded to the Chrome Web Store!

Changelogs vs release notes

In recent years I’ve switched from changelogs (opens new window) to release notes (opens new window).

Changelogs are good for developers but no good for customers or marketing as they are too technical and granular. A user doesn’t care that you “refactored some component” they care that you “added support for context menus”.

And creating release notes at the end of a sprint from is hard, as you’ve forgotten the context a month or so later. I prefer to just update the release notes for the sprint as I commit the feature:


  • added support for context menus
  • made shortcut keys combos clearer

If and when the notes make it into your public-facing site, they are much easier to develop in this format. Many products, such as Chrome (opens new window) and Chrome Dev Tools (opens new window) use their blog to provide additional context.

Also, for developers, GitHub already provides a neat log of changes (opens new window) between releases by comparing commits (opens new window).


To summarise:

  • Think vs
  • major versions signify the start (rather than the end) of a block of product features
  • minor versions count from the current major version
  • minor versions are released as and when they make sense
  • Work is beta by definition until published
  • Use build versions to identify changes in code more granular than the feature level
  • Create patch releases as needed

Note that this approach may not suit you and your project (and you should do what suits you) but is something I shall be moving forwards with, for now.

I’ll be sure to update this post if anything changes or I notice any shortcomings!


How do other extensions handle versioning?

Looking at my Extensions page, I was interested to see which extensions had larger major build numbers.

Creating a new profile, I installed around 20 of the most popular extensions, and ordered by version string:

Extension Users Size Builder Major Minor Patch Build
Lighthouse (opens new window) 1m 27 KB Bundler 100 0 0 0
Honey (opens new window) 10m+ 4.5 MB Compiled 16 1 2
Grammarly (opens new window) 10m+ 37 MB Webpack 14 1118 0
Todoist (opens new window) 700K 85 KB - 11 1
Octotree (opens new window) 300K 3.4 MB Bundler 7 9 3
Loom (opens new window) 5m 13 MB Webpack 5 5 26
Google Keep (opens new window) 6m 4 MB Closure 4 23312 540 1
LastPass (opens new window) 10m+ 55 MB Webpack 4 119 0
Tampermonkey (opens new window) 10m+ 1.5 MB Bundler 4 19 0
Dark Reader (opens new window) 5m 600 KB - 4 9 64
Save to Pocket (opens new window) 2m 356 KB Bundler 4 0 6
Save to Google Drive (opens new window) 3m 650 KB Closure 3 0 4
Momentum (opens new window) 3m 14MB Bundler 2 10 0
Google Translate (opens new window) 10m+ 236 KB Closure 2 0 13
OneTab (opens new window) 2m 1.2 MB Bundler 1 76
Google Docs (opens new window) 10m+ 88 KB Closure 1 65 0
Zoom (opens new window) 7m 240 KB Bundler 1 8 20

I’m not sure how conclusive this is, but perhaps:

  • some developers might reconsider their versioning scheme
  • some projects use (Google Keep)

How do other developers handle versioning?

I posted this article on the Chromium Extensions (opens new window) channel and got some great input:

How does Chrome handle versioning?

Chrome is built upon the Chromium project, which has a very specific (opens new window) release and channel lifecycle.

Their versioning scheme works like this:

The web page (opens new window) explains it like this:

MAJOR and MINOR track updates to the Google Chrome stable channel. In this sense, they reflect a scheduling or marketing decision rather than anything about the code itself. These numbers are generally only significant for tracking milestones. In the event that we get a significant release vehicle for Chromium code other than Google Chrome, we can revisit the versioning scheme.

The BUILD and PATCH numbers together are the canonical representation of what code is in a given release. The BUILD number is always increasing as the source code trunk advances, so build 180 is always newer code than build 177. The PATCH number is always increasing for a given BUILD. Developers and testers generally refer to an instance of the product (Chromium or Google Chrome) as BUILD.PATCH. It is the shortest unambiguous name for a build.

For example, the 154 branch was originally released as, but now stands at It’s the same basic code with a lot of bug fixes applied. The fact that it went from a Beta release to several 1.0 stable releases just reflects the decision to call some version ( ‘out of Beta’.

Note that (as of this article) the public stable version Chrome (opens new window) is 115 and the latest Chromium branch (opens new window) is 117.


Finally, links to the resources which helped me reach a conclusion on my own versioning strategy:


I hope you found this article useful or enjoyable.

If you want to engage further, follow me on Twitter or drop a comment or reaction below.

Either way, thanks for reading!