regex: fix -w when regex can match empty string

This is a weird bug where our optimization for handling -w more quickly
than we would otherwise failed. In particular, if the original regex can
match the empty string, then our word boundary detection would produce
invalid indices to the start the next search at. We "fix" it by simply
bailing when the indices are known to be incorrect.

This wasn't a problem in a previous release since ripgrep 13 tweaked how
word boundaries are detected in commit efd9cfb2.

Fixes #1891

1 files changed, 8 insertions, 0 deletions

diff --git a/tests/regression.rs b/tests/regression.rs
index c905e736..e35e953f 100644
--- a/tests/regression.rs
+++ b/tests/regression.rs
@@ -1029,3 +1029,11 @@ rgtest!(r1878, |dir: Dir, _: TestCommand| {
     let args = &["-U", "--mmap", r"\Abaz", "test"];
     dir.command().args(args).assert_err();
 });
+
+// See: https://github.com/BurntSushi/ripgrep/issues/1891
+rgtest!(r1891, |dir: Dir, mut cmd: TestCommand| {
+    dir.create("test", "\n##\n");
+    // N.B. We use -o here to force the issue to occur, which seems to only
+    // happen when each match needs to be detected.
+    eqnice!("1:\n2:\n2:\n", cmd.args(&["-won", "", "test"]).stdout());
+});