fix: return raw bytes from S3 and AzureBlob downloads (#26)

Author: vasspilka

lib/ash_storage/service.ex

@@ -43,10 +43,18 @@ defmodule AshStorage.Service do
   @doc """
   Download a file from the storage service.
 
-  Returns the file contents as binary data.
+  By default, services built on `Req` (S3, AzureBlob) run Req's `decode_body`
+  step, so the returned body reflects the stored object's `content-type` —
+  `application/json` comes back as a decoded map, `text/csv` as parsed rows,
+  `application/zip` already unzipped, and so on. File-based services (Disk,
+  Mirror) always return raw bytes.
+
+  Callers that need the exact uploaded bytes — writing to disk, streaming to
+  a client, verifying a checksum — should pass `decode_body: false` via the
+  service options on the `Req`-based services.
   """
   @callback download(key(), Context.t()) ::
-              {:ok, binary()} | {:error, term()}
+              {:ok, term()} | {:error, term()}
 
   @doc """
   Delete a file from the storage service.

lib/ash_storage/service/azure_blob.ex

@@ -49,6 +49,10 @@ if Code.ensure_loaded?(Req) do
       (default: `"2020-12-06"`)
     - `:signed_protocol` - SAS protocol restriction. Defaults to `"https"`, or
       `"https,http"` for `http://` endpoints
+    - `:decode_body` - whether `download/2` runs Req's content-type response
+      decoding (JSON → map, CSV → rows, gzip → unzipped, etc.). Defaults to
+      `true` to match Req's own default; pass `false` when you need the raw
+      uploaded bytes.
 
     ## Azure setup
 
@@ -109,7 +113,8 @@ if Code.ensure_loaded?(Req) do
         expires_in: [type: :integer],
         direct_upload_expires_in: [type: :integer],
         service_version: [type: :string],
-        signed_protocol: [type: :string]
+        signed_protocol: [type: :string],
+        decode_body: [type: :boolean]
       ]
     end
 
@@ -148,8 +153,11 @@ if Code.ensure_loaded?(Req) do
     def download(key, %AshStorage.Service.Context{} = ctx) do
       full_key = prefixed_key(key, ctx)
 
+      decode_body? = Keyword.get(ctx.service_opts, :decode_body, true)
+
       with {:ok, url} <- signed_blob_url(full_key, ctx, permissions: "r", expires_in: 900),
-           {:ok, %{status: 200, body: body}} <- Req.get(url, headers: base_headers(ctx)),
+           {:ok, %{status: 200, body: body}} <-
+             Req.get(url, headers: base_headers(ctx), decode_body: decode_body?),
            :ok <- verify_md5(body, ctx.expected_md5) do
         {:ok, body}
       else

lib/ash_storage/service/s3.ex

@@ -23,6 +23,10 @@ if Code.ensure_loaded?(ReqS3) do
     - `:secret_access_key` - AWS secret access key (falls back to `AWS_SECRET_ACCESS_KEY` env var)
     - `:endpoint_url` - custom endpoint URL for S3-compatible services (e.g. MinIO, Tigris)
     - `:prefix` - optional key prefix (e.g. `"uploads/"`)
+    - `:decode_body` - whether `download/2` runs Req's content-type response
+      decoding (JSON → map, CSV → rows, gzip → unzipped, etc.). Defaults to
+      `true` to match Req's own default; pass `false` when you need the raw
+      uploaded bytes.
     """
 
     @behaviour AshStorage.Service
@@ -35,7 +39,8 @@ if Code.ensure_loaded?(ReqS3) do
         access_key_id: [type: :string],
         secret_access_key: [type: :string],
         endpoint_url: [type: :string],
-        prefix: [type: :string]
+        prefix: [type: :string],
+        decode_body: [type: :boolean]
       ]
     end
 
@@ -59,7 +64,10 @@ if Code.ensure_loaded?(ReqS3) do
     def download(key, %AshStorage.Service.Context{} = ctx) do
       full_key = prefixed_key(key, ctx)
 
-      with {:ok, %{status: 200, body: body}} <- Req.get(req(ctx), url: "/#{full_key}"),
+      decode_body? = Keyword.get(ctx.service_opts, :decode_body, true)
+
+      with {:ok, %{status: 200, body: body}} <-
+             Req.get(req(ctx), url: "/#{full_key}", decode_body: decode_body?),
            :ok <- verify_md5(body, ctx.expected_md5) do
         {:ok, body}
       else

test/ash_storage/service/azure_blob_integration_test.exs

@@ -142,6 +142,32 @@ defmodule AshStorage.Service.AzureBlobIntegrationTest do
       assert {:error, :not_found} = AzureBlob.download(unique_key(), ctx())
     end
 
+    test "auto-decodes JSON bodies by default" do
+      key = unique_key()
+      json = ~s({"name":"Alice","score":95})
+
+      assert :ok = AzureBlob.upload(key, json, ctx(content_type: "application/json"))
+
+      assert {:ok, %{"name" => "Alice", "score" => 95}} =
+               AzureBlob.download(key, ctx())
+    end
+
+    test "decode_body: false returns raw CSV bytes" do
+      key = unique_key()
+      csv = "Name,Score\nAlice,95\nBob,87\n"
+
+      assert :ok = AzureBlob.upload(key, csv, ctx(content_type: "text/csv"))
+      assert {:ok, ^csv} = AzureBlob.download(key, ctx(decode_body: false))
+    end
+
+    test "decode_body: false returns raw JSON bytes" do
+      key = unique_key()
+      json = ~s({"name":"Alice","score":95})
+
+      assert :ok = AzureBlob.upload(key, json, ctx(content_type: "application/json"))
+      assert {:ok, ^json} = AzureBlob.download(key, ctx(decode_body: false))
+    end
+
     test "accepts upload when ctx expected_md5 matches the body" do
       key = unique_key()
       data = "checksum-verified payload"

test/ash_storage/service/s3_integration_test.exs

@@ -88,6 +88,24 @@ defmodule AshStorage.Service.S3IntegrationTest do
       assert {:error, :not_found} = S3.download(unique_key(), ctx())
     end
 
+    test "auto-decodes JSON bodies by default" do
+      key = unique_key()
+      json = ~s({"name":"Alice","score":95})
+
+      :ok = raw_put_with_content_type(key, json, "application/json")
+
+      assert {:ok, %{"name" => "Alice", "score" => 95}} =
+               S3.download(key, ctx())
+    end
+
+    test "decode_body: false returns raw bytes for content-typed objects" do
+      key = unique_key()
+      csv = "Name,Score\nAlice,95\nBob,87\n"
+
+      :ok = raw_put_with_content_type(key, csv, "text/csv")
+      assert {:ok, ^csv} = S3.download(key, ctx(decode_body: false))
+    end
+
     test "accepts upload when ctx expected_md5 matches the body" do
       key = unique_key()
       data = "checksum-verified payload"
@@ -361,6 +379,24 @@ defmodule AshStorage.Service.S3IntegrationTest do
 
   # -- Helpers --
 
+  defp raw_put_with_content_type(key, body, content_type) do
+    url = "#{@service_opts[:endpoint_url]}/#{@bucket}/#{key}"
+
+    {:ok, %{status: status}} =
+      Req.put(url,
+        body: body,
+        headers: %{"content-type" => content_type},
+        aws_sigv4: [
+          service: :s3,
+          region: @service_opts[:region],
+          access_key_id: @service_opts[:access_key_id],
+          secret_access_key: @service_opts[:secret_access_key]
+        ]
+      )
+
+    if status in 200..299, do: :ok, else: {:error, status}
+  end
+
   defp unique_key do
     "test/#{System.unique_integer([:positive])}-#{:crypto.strong_rand_bytes(4) |> Base.encode16(case: :lower)}"
   end