BRENNEN

Industries / Technology / Software and SaaS / Developer tools

Build and deploy

Turning source code into a program that runs, fetching the outside code it depends on, and running those steps automatically on shared machines each time the code changes.

7

Milestones achieved, from 1978 to 2025.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1978

    Rebuild only what changed

    Stuart Feldman of Bell Labs wrote Make. A description file lists the files each output depends on and the commands that produce it, and Make "does a depth-first search of this graph" to redo only the work a change makes necessary. Feldman's paper describing Make, shipped with the manuals of Seventh Edition Unix, is dated August 1978.

    wolfram.schneider.org

    Credit

    Stuart Feldman, Bell Laboratories

  2. 1995

    One archive for shared code

    CPAN, the Comprehensive Perl Archive Network, has been online since October 1995. It collects modules, packages of reusable code, in one archive copied to mirror servers around the world, so a programmer can install what others wrote instead of writing it again. By its own count it holds 52,943 distributions by 14,761 authors. Later languages built central registries on the same idea, such as PyPI for Python and npm for JavaScript.

    cpan.org

    Credit

    Jarkko Hietaniemi and the CPAN contributors

  3. 2011

    Developers rename their project

    Kohsuke Kawaguchi started Hudson at Sun Microsystems: a server that builds and tests code each time it changes, a practice called continuous integration. Sun sponsored it and trademarked the name. After Oracle bought Sun it claimed control of the name, and in January 2011 the developers voted 214 to 14 to rename the project Jenkins and carry on outside Oracle's control. Oracle kept Hudson and later moved it to the Eclipse Foundation, which lists the project as archived.

    LessonA trademark controls a project's name, not the people who build it; they can leave with the code.

    i-programmer.infoprojects.eclipse.org

    Credit

    Kohsuke Kawaguchi and the Jenkins developers

  4. 2013

    Programs packed with their parts

    At PyCon US in March 2013, Solomon Hykes, chief executive of the hosting company dotCloud, showed Docker in a lightning talk, and dotCloud released it as open source. Docker packs a program together with the libraries, files and settings it needs into a container, which runs the same way on a laptop or a server, isolated from other programs by features of the Linux kernel. Stack Overflow's 2025 developer survey says Docker "has moved from a popular tool to a near-universal one", with a 17-point rise in use over 2024, the largest of any technology in the survey, partly because the survey merged some categories.

    pyvideo.orginfoq.comsurvey.stackoverflow.co

    Credit

    Solomon Hykes and dotCloud

  5. 2016

    A tiny package breaks builds

    After npm, the JavaScript package registry, gave the package name kik to the messaging company Kik in a naming dispute, the developer Azer Koçulu removed his kik package and 272 others from the registry. One was left-pad. From shortly after 2:30 p.m. Pacific time on 22 March 2016, npm saw hundreds of failures per minute as projects that depended on left-pad, directly or through packages such as babel and atom, failed to install. A volunteer published a replacement within ten minutes, but builds that asked for the exact version 0.0.3 kept failing until npm restored the original from a backup; the failures lasted 2.5 hours. npm now lets an author remove an older package only if no other package in the registry depends on it, it had fewer than 300 downloads in the past week, and it has a single owner.

    LessonCode that thousands of builds pull in is shared infrastructure, and one person withdrawing it should not be able to stop them.

    blog.npmjs.orgdocs.npmjs.com

    Credit

    npm, Inc., for the account; Cameron Westland, who published the replacement

  6. 2019

    CI built into GitHub

    GitHub began supporting continuous integration through GitHub Actions in August 2019 and released it publicly that November. A workflow is a file in the repository that runs builds, tests and releases when code is pushed, and it can reuse shared steps, called actions, that others publish. Researchers at the University of Mons found that it became the most used CI service on GitHub 18 months after its launch, and that 43.9% of the 68,000 repositories they studied used it.

    orbi.umons.ac.be

    Credit

    GitHub; Alexandre Decan, Tom Mens, Pooya Rostami Mazrae and Mehdi Golzadeh, University of Mons, for the measurement

  7. 2025

    A moved tag leaks secrets

    Attackers changed tj-actions/changed-files, a shared GitHub Action that lists the files changed in a pull request and that GitHub says was used in more than 23,000 repositories. They "retroactively modified multiple version tags", the labels such as v35 that workflows use to name a release, to point at a malicious commit that printed the build's secrets, such as access keys and tokens, into the workflow logs, which are public for public repositories. The compromise ran on 14 and 15 March 2025; CISA, the US cybersecurity agency, says it may have been enabled by the compromise of another action, reviewdog/action-setup, and added both to its catalog of known exploited flaws. A 2022 study had found that 9 in 10 steps in public workflows named their action by version tag rather than by commit ID, the fingerprint of the exact code.

    LessonA version tag is a pointer its owner, or an attacker, can move, so a build that must not change should name the exact commit.

    github.comcisa.govorbi.umons.ac.be

    Credit

    GitHub Advisory Database and CISA, for the account

Watch for

  1. Hardened base images, now free

    A base image is the starting layer of a container, usually a small operating system. Docker launched Docker Hardened Images in May 2025, base images cut down to what an app needs to run, and said they have an attack surface "up to 95% smaller" and are kept at "near-zero known CVEs" (a CVE is a published, numbered security flaw). In December 2025 it made the catalog, more than 1,000 images and Helm charts by its count, free under the Apache 2.0 license; near-zero CVEs are "guaranteed" only in the paid Enterprise tier.

    Achieved whenan outside party scans the free images and their usual counterparts with public scanners over several months and publishes the counts.

    docker.comdocker.com

    Claimed by

    Docker, May and December 2025

  2. A compiler ten times faster

    TypeScript 7 is Microsoft's port of the TypeScript compiler to the Go language. Microsoft says it "typically" builds 8 to 12 times faster: the VS Code codebase in 10.6 s instead of 125.7 s (11.9 times) and Sentry's in 15.7 s instead of 139.8 s (8.9 times), with less memory, and that code which compiles cleanly with TypeScript 6.0 "should compile identically". An outside test the next day on an M4 Mac mini, against each project's own pinned version, measured 7.3 times on lobe-chat but 5.3 times on Effect and 3.9 times on drizzle-orm, both heavy users of advanced types.

    Achieved whenoutside measurements on many public codebases, with the hardware stated, fall in Microsoft's range and show the same output from both compilers.

    devblogs.microsoft.comlilting.ch

    Claimed by

    Microsoft, July 2026

Open

  1. 01

    Hardened images, scanned by outsiders

    A monthly public table of known flaws in hardened base images (Docker Hardened Images, Chainguard Images and Google's distroless images) beside the Docker Official Images for the same software: Python, Node.js, nginx and PostgreSQL. Each image, identified by its digest, is scanned on the same day by two scanners, and a small test app is run on each to check it still works.

    The brief
    Credit · unclaimed
  2. 02

    TypeScript 6 and 7, compared

    A public table that builds 300 open-source TypeScript projects with TypeScript 6.0 and 7.0 on the same machine and records, for each project, build time, peak memory, whether the two error lists match, and whether the JavaScript and type declaration files they write are byte for byte the same. Each real difference is filed as an issue.

    The brief
    Credit · unclaimed

Open 01 · the brief

Hardened images, scanned by outsiders

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A monthly public table of known flaws in hardened base images (Docker Hardened Images, Chainguard Images and Google's distroless images) beside the Docker Official Images for the same software: Python, Node.js, nginx and PostgreSQL. Each image, identified by its digest, is scanned on the same day by two scanners, and a small test app is run on each to check it still works.

Why it is still open

The comparisons found come from vendors. Chainguard, which sells hardened images, compared 28 images slimmed by a rival, RapidFort, with its own, using one scanner, Grype. Docker's figures for its images are its own. A search found no independent table that scans the same images with two scanners over time, and none that includes Docker's free images.

chainguard.devdocker.com

Built with what exists

Trivy and Grype, open-source vulnerability scanners; Docker or Podman to pull and run the images; GitHub Actions on a monthly schedule, with every action named by commit ID; a public CSV file.

  1. First step

    Pull the Python image from each of the four sources, scan each with Trivy and Grype on the same day, and publish the eight counts with the image digests.

  2. Done looks like

    Four image sources, four kinds of software, six months. For each image and month: flaw counts by severity from each scanner, the scanner and database versions, and whether the test app ran. A second person reruns one month from the published digests and gets the same counts.

Open 02 · the brief

TypeScript 6 and 7, compared

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public table that builds 300 open-source TypeScript projects with TypeScript 6.0 and 7.0 on the same machine and records, for each project, build time, peak memory, whether the two error lists match, and whether the JavaScript and type declaration files they write are byte for byte the same. Each real difference is filed as an issue.

Why it is still open

Microsoft says it tests TypeScript 7 against projects on GitHub to find regressions, but it publishes timings for five selected codebases and reports from partner companies. The outside test found covered three repositories on one machine. A search found no public table of output differences across many projects.

devblogs.microsoft.comlilting.ch

Built with what exists

TypeScript 6.0 and 7.0 from npm, which Microsoft says can run side by side; the GitHub search API to pick projects that have a tsconfig file; Docker to fix the machine setup; a short script to time each build and compare outputs; an AI coding agent such as Claude Code to fix project setup failures.

  1. First step

    Build ten projects from Microsoft's table and the outside test with both versions, and check that your timings land near the published ones.

  2. Done looks like

    300 projects, each built three times with each version, with timings, memory and output comparisons published with the scripts. Every mismatch is either filed upstream with a link or matched to a documented TypeScript 7.0 change.