deep reify local links #44
Conversation
Fix: #43 This fixes the bug identified by npm/cli#750 Link deps that point to a location underneath the root package are fully resolved and reified. Links external to the root path are left alone, because they exist in a different project and presumably have their own dep resolution being done separately.
Should be the relative path from the link's containing folder, not from the link path itself. Ie, it should be the actual argument passed to fs.symlink() if one were to try to reify the Link node. This changes a lot of snapshots, but the only one that's really relevant is the handling in the YarnLock class, since all the other instances of using consistentResolve to re-relativize a path were handling Link nodes specially anyway (eg, in reify, shrinkwrap, etc.)
The reify() method now returns the *actual* actualTree, so even though engine-mismatching optional deps remain in the package-lock, they're not being passed in the resolved tree. Filter the offending darwin-only module out of the snapshots so that Travis (which is Linux) doesn't fail tests. In that case, just manually verify that fsevents is actually in the package-lock, even though it isn't in the snapshot.
isaacs
force-pushed
the
isaacs/deep-reify-local-links
branch
from
dcb698a
to
eb03ee9
Compare
I was directly doing this myself. So this is an improvement IMO. The one thing I would like to see is an option to also reify the external links. Your presumption is only valid because that is how other tools have treated it, but as I mentioned to @ruyadorno, that is because they were hacking on top of the filesystem based resolution. If you have something like import maps or yarn style PnP you don't have this requirement anymore. I think there are many use cases for having a one step process to reify across many unrelated "projects" if you need to work on them together. |
|
@wesleytodd Thanks. I can certainly imagine cases where it would be unwelcome (or just impossible), and many others where it would be super convenient. (For myself, most of the times that I didn't want link deps messed with, it was because npm did the wrong thing, but Arborist doesn't, so it's usually going to be fine?) In any event, it's just a boolean check at one point, so it's easy enough to add a flag for. I guess then the question would be whether this should be opt-in or opt-out. Any preference? |
Yeah that logic makes sense, but I am a big fan of it "doing the right thing!
Opt in I think. My guess is that it will take time and usage for folks to see how this all works, and it is a bigger change if it is the default behavior. My use case is such that I am happy to opt-in, and I imagine others who dig in like me will feel the same. |
Handle link deps properly in ideal/virtual/reify
Fix: #43
This fixes the bug identified by npm/cli#750
Link deps that point to a location underneath the root package are fully
resolved and reified. Links external to the root path are left alone,
because they exist in a different project and presumably have their own
dep resolution being done separately.
Also, Link nodes now have their
resolvedfield properly relativized to the directory where the link resides, rather than against the link path itself, since that is how symlinks work.cc: @wesleytodd I believe this may have some impacts on your explorations of workspaces. Not sure if it's helpful or harmful, but most likely something to be aware of.