Hi — I've been reading getProjectName in src/spec-node/dockerCompose.ts carefully while contributing to the Zed editor's dev container support (zed-industries/zed), which recently switched from shelling out to @devcontainers/cli to a native Rust implementation. (I'm an outside contributor doing small things there; the Zed team does the heavy lifting.)
The port mirrors this function's precedence ladder quite closely: COMPOSE_PROJECT_NAME env → workspace .env file → explicit top-level name: on the merged compose config → ${workspaceFolderBasename}_devcontainer when the first compose file's directory is <workspace>/.devcontainer → otherwise plain basename of that directory, all passed through toProjectName's lowercase-and-strip-[^-_a-z0-9] sanitization — because divergence produces duplicate compose projects for the same folder when both tools touch it. The fix for this just landed in Zed PR #54302 (tracking issue #54255).
The precedence is crisp in the code but mostly implied, rather than asserted by tests. The existing integration coverage in cli.up.test.ts pins three of the five rungs: explicit name: in a fragment, name: with ${env} substitution (including the CUSTOM_NAME=devcontainer edge case that forces the raw-fragment rescan), and the _devcontainer suffix when the compose file sits under .devcontainer/. The remaining rungs — COMPOSE_PROJECT_NAME from the environment, the .env file lookup, and the "plain basename, no suffix" fallback for compose files outside .devcontainer/ — don't appear to have direct tests. Same fortoProjectName's character class as a contract.
One related observation: the spec on containers.dev is silent on compose project-name derivation — the Docker Compose section covers dockerComposeFile, service, and runServices, but nothing about how the project name is chosen, COMPOSE_PROJECT_NAME, or multi-tool interop around it. So getProjectName in this repo is the de facto contract that any other implementation has to match if the two are ever going to coexist on the same folder. That makes pinning it with tests a little more load-bearing than it might look from inside the CLI's own use.
The question: would you be open, in principle, to pinning this behavior with tests? I'd be happy to contribute them. Either:
-
Unit tests alongside dockerComposeUtils.test.ts — possibly needs a small
refactor to pull the env/.env/config-name reasoning out of the
cliHost-dependent shell of getProjectName, or
-
Integration tests with new fixtures for the three missing rungs, matching
the style of the existing compose-with-name* describe blocks in
cli.up.test.ts.
Or both, or a different shape — I'd rather ask than cold-drop a PR with the wrong approach. The motivation is durability: a CI-enforced invariant here is a nicer long-term mitigation than a doc comment in the Rust port saying "mirrors the reference CLI as of commit X".
No rush, and no problem if the answer is "not a priority / won't review" — in that case I'll close this and I figure Zed can keep on mirroring.
Hi — I've been reading
getProjectNameinsrc/spec-node/dockerCompose.tscarefully while contributing to the Zed editor's dev container support (zed-industries/zed), which recently switched from shelling out to@devcontainers/clito a native Rust implementation. (I'm an outside contributor doing small things there; the Zed team does the heavy lifting.)The port mirrors this function's precedence ladder quite closely:
COMPOSE_PROJECT_NAMEenv → workspace.envfile → explicit top-levelname:on the merged compose config →${workspaceFolderBasename}_devcontainerwhen the first compose file's directory is<workspace>/.devcontainer→ otherwise plain basename of that directory, all passed throughtoProjectName's lowercase-and-strip-[^-_a-z0-9]sanitization — because divergence produces duplicate compose projects for the same folder when both tools touch it. The fix for this just landed in Zed PR #54302 (tracking issue #54255).The precedence is crisp in the code but mostly implied, rather than asserted by tests. The existing integration coverage in
cli.up.test.tspins three of the five rungs: explicitname:in a fragment,name:with${env}substitution (including theCUSTOM_NAME=devcontaineredge case that forces the raw-fragment rescan), and the_devcontainersuffix when the compose file sits under.devcontainer/. The remaining rungs —COMPOSE_PROJECT_NAMEfrom the environment, the.envfile lookup, and the "plain basename, no suffix" fallback for compose files outside.devcontainer/— don't appear to have direct tests. Same fortoProjectName's character class as a contract.One related observation: the spec on containers.dev is silent on compose project-name derivation — the Docker Compose section covers
dockerComposeFile,service, andrunServices, but nothing about how the project name is chosen,COMPOSE_PROJECT_NAME, or multi-tool interop around it. SogetProjectNamein this repo is the de facto contract that any other implementation has to match if the two are ever going to coexist on the same folder. That makes pinning it with tests a little more load-bearing than it might look from inside the CLI's own use.The question: would you be open, in principle, to pinning this behavior with tests? I'd be happy to contribute them. Either:
Unit tests alongside
dockerComposeUtils.test.ts— possibly needs a smallrefactor to pull the env/
.env/config-name reasoning out of thecliHost-dependent shell ofgetProjectName, orIntegration tests with new fixtures for the three missing rungs, matching
the style of the existing
compose-with-name*describe blocks incli.up.test.ts.Or both, or a different shape — I'd rather ask than cold-drop a PR with the wrong approach. The motivation is durability: a CI-enforced invariant here is a nicer long-term mitigation than a doc comment in the Rust port saying "mirrors the reference CLI as of commit X".
No rush, and no problem if the answer is "not a priority / won't review" — in that case I'll close this and I figure Zed can keep on mirroring.