rfcs
rfcs copied to clipboard
npm
add "obey user specifier" RFC
When I type npm install package@specifier, a normalized specifier is what should be saved to package.json, no more, no less.
Motivation
Currently, if i type npm install --save-dev eslint@=8.8.0 other, it installs v8.8.0, but saves ^8.8.0 to package.json (with the default save-prefix of ^). (note: i'm intentionally not using --save-exact here, because i want other to install using the default save-prefix).
When no @version is provided, or when a dist-tag is provided, it makes perfect sense to resolve that to a version, and save it with save-prefix. However, when I've provided a version range - =1.2.3, ~1.2.3, >= 1.2 - I know what I want, I’ve explicitly told npm I want it, and npm should obey my wishes not just in the reification in node_modules (which I believe it does), but also in what it saves to package.json.
agreed 👍 if the user is specifying a valid semver prefix in along with the package spec when running npm install we should use that prefix instead of respecting the save-prefix config option.
👍🏻 removing Agenda flag as we're all in favor of this & have queued this up for v9 (we'll land this as accepted) - https://github.com/npm/statusboard/issues/443
In some case, save tag is expected. Antd mark the stable version with conch dist-tag, when i install antd with npm i antd@conch, i also want save it with conch tag.
And latest-x, beta, next is also i want to save.
with npm i antd@conch, i also want save it with conch tag.
I think there are two modes of operation here, and I would like to see both continued to be supported. Since resolving a version from a dist-tag has explicit behavior today I would love to see a new flag to achieve this behavior, which might need to be a separate RFC, I don't know if @ljharb (or the npm team) would want to bundle that together into this one since it already has agreement.
I agree they're two different things - one is this RFC, which is that the version specifier range should be obeyed; the other is "save the tag vs resolve the tag", and i agree there should be a flag for that.