fix: accept GCS gzip responses without Content-Length

src/aws/client.rs

@@ -922,6 +922,7 @@ impl GetClient for S3Client {
     const HEADER_CONFIG: HeaderConfig = HeaderConfig {
         etag_required: false,
         last_modified_required: false,
+        stored_size_header: None,
         version_header: Some(VERSION_HEADER),
         user_defined_metadata_prefix: Some(USER_DEFINED_METADATA_HEADER_PREFIX),
     };

src/azure/client.rs

@@ -1075,6 +1075,7 @@ impl GetClient for AzureClient {
     const HEADER_CONFIG: HeaderConfig = HeaderConfig {
         etag_required: true,
         last_modified_required: true,
+        stored_size_header: None,
         version_header: Some(VERSION_HEADER),
         user_defined_metadata_prefix: Some(USER_DEFINED_METADATA_HEADER_PREFIX),
     };

src/client/get.rs

@@ -445,6 +445,7 @@ mod tests {
     const CFG: HeaderConfig = HeaderConfig {
         etag_required: false,
         last_modified_required: false,
+        stored_size_header: None,
         version_header: None,
         user_defined_metadata_prefix: Some("x-test-meta-"),
     };
@@ -507,6 +508,54 @@ mod tests {
         assert_eq!(err.to_string(), "Requested 2..6, got 2..4");
     }
 
+    #[test]
+    fn test_get_missing_content_length() {
+        // Mirrors the GCS config: size falls back to x-goog-stored-content-length.
+        const RELAXED: HeaderConfig = HeaderConfig {
+            stored_size_header: Some("x-goog-stored-content-length"),
+            ..CFG
+        };
+        let path = Path::from("test");
+
+        let resp = |headers: &[(&str, &str)]| {
+            let mut builder = http::Response::builder().status(StatusCode::OK);
+            for (k, v) in headers {
+                builder = builder.header(*k, *v);
+            }
+            builder.body(()).unwrap().into_parts().0
+        };
+
+        // No Content-Length, stored-size header present -> best-effort size from fallback.
+        let r = resp(&[("x-goog-stored-content-length", "355")]);
+        let (range, meta) = get_range_meta(RELAXED, &path, None, &r).unwrap();
+        assert_eq!(meta.size, 355);
+        assert_eq!(range, 0..355);
+
+        // No Content-Length and the stored-size header also absent -> still an error.
+        let r = resp(&[]);
+        let err = get_range_meta(RELAXED, &path, None, &r).unwrap_err();
+        assert_eq!(
+            err.to_string(),
+            "Content-Length Header missing from response"
+        );
+
+        // A present Content-Length always wins over the fallback.
+        let r = resp(&[
+            ("content-length", "10"),
+            ("x-goog-stored-content-length", "355"),
+        ]);
+        let (_, meta) = get_range_meta(RELAXED, &path, None, &r).unwrap();
+        assert_eq!(meta.size, 10);
+
+        // With the strict default (S3/Azure), a missing Content-Length is still fatal.
+        let r = resp(&[]);
+        let err = get_range_meta(CFG, &path, None, &r).unwrap_err();
+        assert_eq!(
+            err.to_string(),
+            "Content-Length Header missing from response"
+        );
+    }
+
     #[test]
     fn test_get_attributes() {
         let resp = make_response(

src/client/header.rs

@@ -36,6 +36,14 @@ pub(crate) struct HeaderConfig {
     /// Defaults to `true`
     pub last_modified_required: bool,
 
+    /// Header to read the object size from when `Content-Length` is absent.
+    ///
+    /// GCS omits `Content-Length` on chunked `Content-Encoding: gzip` responses — large
+    /// bodies, or decompressive transcoding — but always sends `x-goog-stored-content-length`.
+    /// Stores that always send a `Content-Length` (S3, Azure) leave this `None`, so a missing
+    /// `Content-Length` stays an error for them.
+    pub stored_size_header: Option<&'static str>,
+
     /// The version header name if any
     pub version_header: Option<&'static str>,
 
@@ -139,8 +147,12 @@ pub(crate) fn header_meta(
         Err(e) => return Err(e),
     };
 
+    // Prefer `Content-Length`, falling back to a store-provided size header: GCS omits
+    // `Content-Length` on chunked gzip responses (large bodies, or transcoding) but always
+    // sends `x-goog-stored-content-length`. Stores without such a header still require it.
     let content_length = headers
         .get(CONTENT_LENGTH)
+        .or_else(|| cfg.stored_size_header.and_then(|h| headers.get(h)))
         .ok_or(Error::MissingContentLength)?;
 
     let content_length = content_length

src/gcp/client.rs

@@ -52,6 +52,7 @@ const VERSION_HEADER: &str = "x-goog-generation";
 const DEFAULT_CONTENT_TYPE: &str = "application/octet-stream";
 const USER_DEFINED_METADATA_HEADER_PREFIX: &str = "x-goog-meta-";
 const STORAGE_CLASS: &str = "x-goog-storage-class";
+const STORED_CONTENT_LENGTH_HEADER: &str = "x-goog-stored-content-length";
 
 static VERSION_MATCH: HeaderName = HeaderName::from_static("x-goog-if-generation-match");
 
@@ -617,9 +618,12 @@ impl GoogleCloudStorageClient {
 #[async_trait]
 impl GetClient for GoogleCloudStorageClient {
     const STORE: &'static str = STORE;
+    // GCS omits Content-Length on chunked gzip responses (large bodies, or decompressive
+    // transcoding); the size is recovered from x-goog-stored-content-length instead.
     const HEADER_CONFIG: HeaderConfig = HeaderConfig {
         etag_required: true,
         last_modified_required: true,
+        stored_size_header: Some(STORED_CONTENT_LENGTH_HEADER),
         version_header: Some(VERSION_HEADER),
         user_defined_metadata_prefix: Some(USER_DEFINED_METADATA_HEADER_PREFIX),
     };

src/http/client.rs

@@ -359,6 +359,7 @@ impl GetClient for Client {
     const HEADER_CONFIG: HeaderConfig = HeaderConfig {
         etag_required: false,
         last_modified_required: false,
+        stored_size_header: None,
         version_header: None,
         user_defined_metadata_prefix: None,
     };