Interpret imports relative to the location of the file containing them rather than the current working directory

[kostmo]

I'd like to distribute a self-contained demo "Gosling simulator" as a GitHub gist (practically speaking, a .zip file). This demo includes both a scenario (.yaml) file and multiple swarm-lang (.sw) files.

However, there's not a clear way in the scenario file to specify a path to scripts that is "relative" to the scenario file, such that the scenario can be run outside of the swarm/ directory. Paths in the run command seem to be relative to the current working directory.

Describe the solution you'd like

For a file hierarchy like this:

~/Downloads
~/Downloads/myscenario.yaml
~/Downloads/myscript.sw

and when swarm is invoked like this:

stack run -- --scenario ~/Downloads/myscenario.yaml

Support a notation like this in the REPL/swarm-lang code:

run "@myscript.sw"

The @ sign indicates that myscript.sw exists in the same directory as the scenario currently being played (in this case, the myscenario.yaml file).

For a hierarchy like this:

~/Downloads
~/Downloads/myscenario.yaml
~/Downloads/_myscenario/myscript.sw

The command would be:

run "@_myscenario/myscript.sw"

[xsebek]

Where is this notation used? Also, how would players learn about this feature?

Instead of hiding it by magical character, why not have :cd REPL command like GHCi?

   :cd <dir>                   change directory to <dir>

Or just use a normal function like Python:

>>> from os import chdir
>>> chdir("/tmp/")

[kostmo]

I had this feature in mind mainly (strictly?) for scenario authors. For one, it would allow running the Gosling simulator "out of the box" after simply unpacking the .zip download. And two, it would allow simplification of paths like this, which would instead be written as:

run "@hanoi-count.sw"

And after that, for even better organization, all of the hanoi-related files (both .sw and .yaml) could be placed in a subdirectory. With the @-notation, no changes would be required to the scenario file after performing this relocation.

[byorgey]

I am trying to imagine how this would work, practically speaking. Currently, the text value passed to the run command is interpreted at runtime when the run command is executed. I think at that point it is possible to know what scenario is being played and what directory it is in. So this would require modifying the code in Swarm.Game.Step that interprets the run command to strip off the leading @ and look up the current scenario path.

With that said, at some point (hopefully soon!) the run command will go away and be replaced by a much better import mechanism (#495). In that brave new world order, the argument to import will be processed at typechecking time rather than at runtime (see e.g. this comment), and it will be entirely natural for import paths to be interpreted relative to the location of the file containing them, just like Dhall does. In which case we will get exactly the behavior you want without the need for any special syntax.

Personally, I would be inclined to just wait for #495 (I've actually been working on that, although it's tricky and I can't give any specific promises when it will be ready). But if someone gets impatient they could implement this first.

[xsebek]

At this point we could have a tag Feature that works around missing feature. 😆

[byorgey]

Just an update --- I have been working in earnest on #495 and I can confirm that it will definitely have this behavior (interpreting import locations relative to the location of their containing file).