ignore: WalkBuilder.add_ignore() patterns are not rooted in the path being walked

[dragonmaus]

Alternative title: No way to specify "root path" for WalkBuilder.add_ignore()

Describe your question, feature request, or bug.

When building an instance of Walk using ignore::WalkBuilder, the patterns in files added by .add_ignore() are matched against the current working directory of the process, instead of the root of the tree being Walked.

If this is a bug, what are the steps to reproduce the behavior?

After downloading the files in this gist, perform the following steps:

mkdir src
mv main.rs src
cargo build --release
target/release/test -I ignore .              # this is "Test 1"
cd ..
test/target/release/test -I test/ignore test # this is "Test 2"

If this is a bug, what is the actual behavior?

Test 1 produces the expected output:

.
./Cargo.lock
./Cargo.toml
./ignore
./src
./src/main.rs

Test 2, however, also includes the contents of the target directory (full output here).

If this is a bug, what is the expected behavior?

Test 2 should produce the following output:

test
test/Cargo.lock
test/Cargo.toml
test/ignore
test/src
test/src/main.rs

Changing the pattern in the file ignore to /test/target/ produces the desired results, but is not what I would expect (or want) to have to do.

[tommilligan]

I'd like to take this issue - I've reproduced it and should have a fix this week.

Minimal test case:

diff --git a/ignore/src/walk.rs b/ignore/src/walk.rs
index e00b29a..f765b7f 100644
--- a/ignore/src/walk.rs
+++ b/ignore/src/walk.rs
@@ -1919,6 +1919,22 @@ mod tests {
         );
     }
 
+    #[test]
+    fn explicit_ignore_absolute_path() {
+        let td = tmpdir();
+        let igpath = td.path().join(".not-an-ignore");
+        mkdirp(td.path().join("a"));
+        wfile(&igpath, "/a");
+        wfile(td.path().join("foo"), "");
+        wfile(td.path().join("a/foo"), "");
+        wfile(td.path().join("bar"), "");
+        wfile(td.path().join("a/bar"), "");
+
+        let mut builder = WalkBuilder::new(td.path());
+        assert!(builder.add_ignore(&igpath).is_none());
+        assert_paths(td.path(), &builder, &["foo", "bar"]);
+    }
+
     #[test]
     fn gitignore_parent() {
         let td = tmpdir();

[BurntSushi]

@dragonmaus Thank you very much for the detailed bug report! Unfortunately, I'm quite unsure how to fix this. Please see my comment in #1399: https://github.com/BurntSushi/ripgrep/pull/1399#issuecomment-586613865

[dragonmaus]

@BurntSushi Thank you for looking into this, and thank you @tommilligan for doing the legwork.

I will have to re-evaluate how I am using WalkBuilder; perhaps there is some cognitive disconnect on my end, or maybe a novel solution will present itself.

[BurntSushi]

I don't think you are doing anything wrong. I think your use case is valid. I just don't know how quite to do it with the current API. We could use @tommilligan's solution with a new method, e.g., add_ignore_relative, and just panic if add is called to add additional roots. That would make your use case work. I wouldn't be too opposed to doing this since IMO the ignore API is an unmitigated disaster at this point anyway. So adding another quirk isn't that big of a deal. Once I summon the courage to redo the crate, I can rethink everything from first principles. The ignore crate has gone through a lot of requirements evolution and has badly needed an overhaul for quite some time now. But it will a long time before I get to doing it. (I'd say on the order of a year.)

So if you want a work-around until then, then I think I'd be okay with the aforementioned tweaks to @tommilligan's PR. (Apologies for closing it, but I want to completely clear the PR queue before doing the next release.)