[RRFC] .npmignore pragma to include .gitignore

[ljharb]

Motivation ("The Why")

While opinions differ, I consider "files" to be dangerous, since I have zero concern for accidental publishing but I have infinite concern for accidental breakage. (Let's avoid debating this)

When using .npmignore, one has to remember to start by cp .gitignore .npmignore, and then later, to keep those in sync, since unfortunately npm ignores .gitignore when .npmignore is present. This is a maintenance burden and is error-prone over time.

Example/How/Desired Behavior

These examples show a hypothetical where "coverage data" is ignored by both git and npm, but the "dist" folder (build output) is ignored by git but included in npm.

Current:

.gitignore

coverage/ # coverage data
dist/ # build output

.npmignore

coverage/ # coverage data

Potential:

.gitignore

coverage/ # coverage data
dist/ # build output

.npmignore

<include gitignore> # ideally something that would make the npm CLI crash prior to supporting this feature
!dist/ # unignore build output

The advantage would be that additions to .gitignore no longer need to be propagated to .npmignore. The disadvantage would be that removals from .gitignore now do need to be propagated, but the explicit opt-in would mean that's expected.

Thoughts?

[isaacs]

I think this is pretty useful. The challenge will be to find a syntax that would indeed make the npm CLI crash prior to supporting this feature. Globs are very accommodating, unfortunately, since filenames on posix systems can contain almost any character. I don't see anything offhand that will make ignore-walk throw an error. Anything that doesn't parse as a glob is just treated as part of the filename being matched.

[ljharb]

Another alternative would be to make a new filename, that when present causes npmignore to be ignored - but i have no brilliant suggestions for a name for it :-/

[isaacs]

Another alternative would be to make a new filename, that when present causes npmignore to be ignored - but i have no brilliant suggestions for a name for it :-/

Hm, not seeing how that would fix the make-work of adding something to both .gitignore and .npmignore. I'm reminded somewhat of https://xkcd.com/927/ 😂

[ljharb]

Yeah, fair point - for npm back compat i'd still need to maintain an npmignore in that case, so that's probably not a viable solution.