From aaa4b2b7c2f461335078367bd61587708e2edaf8 Mon Sep 17 00:00:00 2001
From: ffccites <99155080+PDGGK@users.noreply.github.com>
Date: Tue, 11 Aug 2026 04:20:07 +1000
Subject: [PATCH 1/2] fix(services/hf): correct the refs/convert revision split

parse_revision sliced rev_and_path[..14 + slash] to recover the revision, but
"refs/convert/" is 13 bytes, so the separator that terminates the revision
segment was pulled into the revision itself. A URI like

    datasets/squad@refs/convert/parquet/default/train/0000.parquet

yielded the revision "refs/convert/parquet/" rather than "refs/convert/parquet",
and that trailing slash is carried into the operator config and percent-encoded
into every request path as refs%2Fconvert%2Fparquet%2F.

Rebuild the revision with format! the way the refs/pr/ arm immediately below
already does, so there is no offset to get wrong.
---
 core/services/hf/src/core.rs | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

diff --git a/core/services/hf/src/core.rs b/core/services/hf/src/core.rs
index dff4d84bdc81..9ffb4e44e1e4 100644
--- a/core/services/hf/src/core.rs
+++ b/core/services/hf/src/core.rs
@@ -1058,10 +1058,8 @@ mod uri {
             // Match special refs: refs/(convert|pr)/<segment>
             if let Some(rest) = rev_and_path.strip_prefix("refs/convert/") {
                 return if let Some(slash) = rest.find('/') {
-                    (
-                        rev_and_path[..14 + slash].to_string(),
-                        rest[slash + 1..].to_string(),
-                    )
+                    let revision = format!("refs/convert/{}", &rest[..slash]);
+                    (revision, rest[slash + 1..].to_string())
                 } else {
                     (rev_and_path.to_string(), String::new())
                 };
@@ -1267,6 +1265,15 @@ mod uri {
             assert_eq!(p.path, "");
         }
 
+        #[test]
+        fn resolve_refs_convert_revision_with_path() {
+            let p = resolve("datasets/squad@refs/convert/parquet/default/train/0000.parquet");
+            assert_eq!(p.repo.repo_type, HfRepoType::Dataset);
+            assert_eq!(p.repo.repo_id, "squad");
+            assert_eq!(p.repo.revision.as_deref(), Some("refs/convert/parquet"));
+            assert_eq!(p.path, "default/train/0000.parquet");
+        }
+
         #[test]
         fn resolve_refs_pr_revision() {
             let p = resolve("username/my_model@refs/pr/10");

From 35f23d305b99a95c52aede57593bef86b9dde2e2 Mon Sep 17 00:00:00 2001
From: ffccites <99155080+PDGGK@users.noreply.github.com>
Date: Tue, 11 Aug 2026 11:13:49 +1000
Subject: [PATCH 2/2] refactor(services/hf): use split_once for the special-ref
 split

Applies the reviewer's suggestion. Both the refs/convert/ and refs/pr/
arms now use split_once instead of find + manual slicing, which removes
the index arithmetic entirely and keeps the two adjacent arms symmetric.
No behaviour change: split_once yields exactly the same two halves the
find + slice pair did.
---
 core/services/hf/src/core.rs | 16 ++++++----------
 1 file changed, 6 insertions(+), 10 deletions(-)

diff --git a/core/services/hf/src/core.rs b/core/services/hf/src/core.rs
index 9ffb4e44e1e4..26ab4cdf10da 100644
--- a/core/services/hf/src/core.rs
+++ b/core/services/hf/src/core.rs
@@ -1057,19 +1057,15 @@ mod uri {
 
             // Match special refs: refs/(convert|pr)/<segment>
             if let Some(rest) = rev_and_path.strip_prefix("refs/convert/") {
-                return if let Some(slash) = rest.find('/') {
-                    let revision = format!("refs/convert/{}", &rest[..slash]);
-                    (revision, rest[slash + 1..].to_string())
-                } else {
-                    (rev_and_path.to_string(), String::new())
+                return match rest.split_once('/') {
+                    Some((segment, path)) => (format!("refs/convert/{segment}"), path.to_string()),
+                    None => (rev_and_path.to_string(), String::new()),
                 };
             }
             if let Some(rest) = rev_and_path.strip_prefix("refs/pr/") {
-                return if let Some(slash) = rest.find('/') {
-                    let revision = format!("refs/pr/{}", &rest[..slash]);
-                    (revision, rest[slash + 1..].to_string())
-                } else {
-                    (rev_and_path.to_string(), String::new())
+                return match rest.split_once('/') {
+                    Some((segment, path)) => (format!("refs/pr/{segment}"), path.to_string()),
+                    None => (rev_and_path.to_string(), String::new()),
                 };
             }