"watch" is broken on windows

[Hades32]

The issue seems to be that the call to verify(...) in main.rs:92 expects the path to be in Unix style (forward slashes instead of backslashes), so the later compare breaks.

I fixed this locally by just doing a string replace, but I guess a true Rustacean will fix this much more nicely so I didn't add a patch.

PS: It's also pretty inconvenient that "watch" doesn't continue with the next exercise once you fix the current one.

[komaeda]

@Hades32 it should continue! sounds like it's broken for you, maybe?

[guttume]

I faced the same issue with watch on Windows

[steing]

reverting to tags/1.0.0 and rebuilding seems to have worked for me. compiles on save and advances to next exercise.

[guttume]

@steing noted. Thank you.

[sebsebmc]

I have the same issue and #134 did not fix the problem.

[cjpearce]

Happy to give fixing this a go if it's still broken. I have a windows vm I can debug in.

[komaeda]

@cjpearce That would be much appreciated! I'm honestly a bit stuck on this issue as I have neither the time or the resources to get going on this

[sebsebmc]

I've been debugging the issue and I believe that the problem is that the paths that notify is returning look something like ..../exercises\\variables\\variables1.rs but the value in info.toml looks like exercises/variables/variables1.rs so the ends_with comparison here always fails:

rustlings/src/verify.rs

Line 16 in 04d1d4c

if start_at.ends_with(path) {

I believe the solution may require using something like http://github.com/BurntSushi/same-file

I have a pretty ugly patch that canonicalize()s both paths for comparison and that seems to work on Windows, haven't tested it on Linux yet. I'm hoping there may be a more elegant solution. The other option is to use Path components and checking if the last components match (but I haven't tested if this works yet).

[cjpearce]

@sebsebmc That's in line with what I've found too. I think that canonicalize() might be a pretty good solution, since that path really needs it. But I absolutely agree gets pretty messy if we start constructing and then canonicalizing the paths just before we use them.

I think one way of fixing this is to deal with constructing canonicalized the paths up front before we get into any of the logic. I'm working on a PR that would clean up how we handle the paths, which hopefully will mean it doesn't become as messy. I should have it something to show in the next day that we could test.

If you get to a solution you're happy with first I'd be happy to help on that too.

[sebsebmc]

rust-lang/rust#42869 has a discussion on canonicalization on Windows.

I looked up some projects that use notify and found that alacritty and xi-editor both do canonicalization for their paths.

There is rust-lang/rfcs#2208 that could potentially be used in the future if it ever lands, but it seems like canonicalization should be fine and is expected to work.