[BUG] emplace is triggered by non global npm about emplace HOT 6 OPEN

I think a has_required_flag member makes a lot of sense for this situation. Sounds look a good clean solution!

from emplace.

It might make more sense to have a hook which is given the entire command line and is able to reject it - then the hook would be usable for other special cases where a declarative approach is a lot of effort

from emplace.

Hi @jayvdb, unfortunately I'm not maintaining this application anymore, so I won't pick up this issue. Any PR's are greatly appreciated though!

from emplace.

I am happy to build the fix for this.

I've had a poke around, and notice the problem of local project package installs is unique to npm. All others are either global/system or per-user installs. Also none of the open issues are proposing another package manager which would have the same problem.

A quick fix is to change npm to be similar to rustup which has a sub_commands of "component add".

i.e. sub_commands for npm could be install -g and install --global. However that would not catch npm install foo -g / npm install foo --global.

To properly solve his without regressions, IMO the package managers need a member has_required_flag or similar, which for npm would return -g and --global. This feels a bit like overkill because there is no other package manager which currently needs this. Maybe I have missed a better alternative.

from emplace.

Digging a little deeper, I wonder if it would be OK to add a CaptureFlag::Invalidating("--path") to cargo.rs to replace has_invalidating_flag, and CaptureFlag::Required(["-g", "--global"]) to npm.rs which would tell catch that the line should be ignored unless it was found.

Invalidating is easy as it presence in the command args could be handled in handle_capture_flags, which could return a bool to tell capture to ignore the line. Multiple Invalidating would be OR'd.

Required is slightly more involved, as capture would need to see if Required is present in the package managers capture_flags() return value before processing the command args, and verified that it was seen. Required would take a vec of equivalent flags that are OR'd (e.g. "-g" or "--global" are required), and multiple Required would be AND'd.

This approach would mean that flag processing is still declarative, but doesn't involve creating new members like has_required_flag that are unnecessary for most package managers.

from emplace.

This sounds like a good solution! I think this would make the implementation a lot cleaner.

from emplace.