llm_providers 0.1.0 → 0.1.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: fd68399074379aa3cca94d766217abe7270698fa764b1656d1fc585977c47e49
-  data.tar.gz: 6b82fe10d4ae4bb414317c1fbbca2252976032032fcc846c2864154cb54aee62
+  metadata.gz: f2f9351f5b2fcd03887b26b61aad5a7efdd810207d9ae96577455b5450730c96
+  data.tar.gz: 17aa0c0c838045102727f5650ba5be88d232ac6e54b7bf283eb5d8bcc7b1b2d4
 SHA512:
-  metadata.gz: 77759c7d037da031e31db5ce76456279d5c1ed104b7df60d19289ceb6cba0ad26b7de1aa80f1e78b1afcc18b41d780ffd34718cb318b6cf905a1947adde806e0
-  data.tar.gz: 598b707ba1f7160b7b6f03b3d291a7d2bf0d726afe55d6fd86b60747c8ebc440438c5e5753597eb36016b0012552404aedc98f3cc3dc28e5ba93465c78a8b73a
+  metadata.gz: 6ec192e3c3bd4d66d29e7dd619f7821cd2a79d41bd0df3ee61376dee03bf5fa0fd78b1fb1fb9f0ee2a4481c0f419139decade2f70767bc7493d6c59f619bc158
+  data.tar.gz: 5482d3ed05b2dedc3285cfc6b337e08ed1334826eefa4549e79a64a1effabe4c97078fd550b0946db24b4b5b9f5f14885636a683352cd6f9559e9982808758e1

data/CHANGELOG.md

@@ -1,5 +1,13 @@
 # Changelog
 
+## [0.1.1] - 2026-02-27
+
+### Fixed
+
+- Fix SSE streaming parser losing data when HTTP chunk boundaries split SSE data lines mid-line
+  - Added line buffering to `stream_response` in Anthropic, OpenAI, and OpenRouter providers
+  - Fixes tool call `input_json_delta` events being dropped, which caused tool calls with empty `{}` input
+
 ## [0.1.0] - 2026-02-13
 
 - Initial release

data/lib/llm_providers/providers/anthropic.rb

@@ -131,64 +131,19 @@ module LlmProviders
         payload[:stream] = true
         started_at = Time.now
 
+        puts "[Anthropic] stream_response payload tools: #{payload[:tools]&.size || 0}"
+        puts "[Anthropic] payload[:tools] = #{payload[:tools].inspect}" if payload[:tools]
+
         full_content = ""
         tool_calls = []
         usage = {}
+        line_buffer = ""
 
-        conn = Faraday.new do |f|
-          f.options.open_timeout = 10
-          f.options.read_timeout = 300
-          f.options.write_timeout = 30
-          f.adapter Faraday.default_adapter
-        end
-
-        response = conn.post(API_URL) do |req|
-          req.headers["Content-Type"] = "application/json"
-          req.headers["x-api-key"] = api_key
-          req.headers["anthropic-version"] = API_VERSION
-          req.body = payload.to_json
-          req.options.on_data = proc do |chunk, _|
-            process_stream_chunk(chunk, full_content, tool_calls) do |parsed|
-              if parsed[:content]
-                full_content += parsed[:content]
-                block.call(content: parsed[:content])
-              end
-              usage = parsed[:usage] if parsed[:usage]
-            end
-          end
-        end
-
-        unless response.success?
-          error_message = begin
-            if response.body.is_a?(Hash)
-              response.body["error"]&.dig("message")
-            else
-              response.body.to_s
-            end
-          rescue StandardError
-            response.body.to_s
-          end
-          raise ProviderError.new(
-            (error_message && !error_message.empty? ? error_message : nil) || "API error",
-            code: "anthropic_error"
-          )
-        end
-
-        {
-          content: full_content,
-          tool_calls: tool_calls,
-          usage: usage,
-          latency_ms: ((Time.now - started_at) * 1000).to_i,
-          raw_response: { content: full_content, tool_calls: tool_calls }
-        }
-      end
-
-      def process_stream_chunk(chunk, _full_content, tool_calls)
-        chunk.each_line do |line|
+        process_sse_line = proc do |line|
           next unless line.start_with?("data: ")
 
           data = line.sub("data: ", "").strip
-          next if data == "[DONE]"
+          next if data == "[DONE]" || data.empty?
 
           begin
             event = JSON.parse(data)
@@ -196,7 +151,11 @@ module LlmProviders
             case event["type"]
             when "content_block_delta"
               if event.dig("delta", "type") == "text_delta"
-                yield(content: event.dig("delta", "text"))
+                text = event.dig("delta", "text")
+                if text
+                  full_content += text
+                  block.call(content: text)
+                end
               elsif event.dig("delta", "type") == "input_json_delta"
                 if tool_calls.any?
                   tool_calls.last[:input_json] ||= ""
@@ -204,7 +163,9 @@ module LlmProviders
                 end
               end
             when "content_block_start"
+              puts "[Anthropic] content_block_start: #{event["content_block"]&.dig("type")}"
               if event.dig("content_block", "type") == "tool_use"
+                puts "[Anthropic] Tool use started: #{event.dig("content_block", "name")}"
                 tool_calls << {
                   id: event.dig("content_block", "id"),
                   name: event.dig("content_block", "name"),
@@ -223,17 +184,65 @@ module LlmProviders
               end
             when "message_delta"
               if event["usage"]
-                yield(usage: {
+                usage = {
                   input: event.dig("usage", "input_tokens"),
                   output: event.dig("usage", "output_tokens"),
                   cached_input: event.dig("usage", "cache_read_input_tokens")
-                })
+                }
               end
             end
           rescue JSON::ParserError
             # Skip invalid JSON
           end
         end
+
+        conn = Faraday.new do |f|
+          f.options.open_timeout = 10
+          f.options.read_timeout = 300
+          f.options.write_timeout = 30
+          f.adapter Faraday.default_adapter
+        end
+
+        response = conn.post(API_URL) do |req|
+          req.headers["Content-Type"] = "application/json"
+          req.headers["x-api-key"] = api_key
+          req.headers["anthropic-version"] = API_VERSION
+          req.body = payload.to_json
+          req.options.on_data = proc do |chunk, _|
+            line_buffer += chunk
+            lines = line_buffer.split("\n", -1)
+            line_buffer = lines.pop || ""
+
+            lines.each(&process_sse_line)
+          end
+        end
+
+        # Process any remaining data in the buffer
+        process_sse_line.call(line_buffer) unless line_buffer.empty?
+
+        unless response.success?
+          error_message = begin
+            if response.body.is_a?(Hash)
+              response.body["error"]&.dig("message")
+            else
+              response.body.to_s
+            end
+          rescue StandardError
+            response.body.to_s
+          end
+          raise ProviderError.new(
+            (error_message && !error_message.empty? ? error_message : nil) || "API error",
+            code: "anthropic_error"
+          )
+        end
+
+        {
+          content: full_content,
+          tool_calls: tool_calls,
+          usage: usage,
+          latency_ms: ((Time.now - started_at) * 1000).to_i,
+          raw_response: { content: full_content, tool_calls: tool_calls }
+        }
       end
 
       def sync_response(payload)

data/lib/llm_providers/providers/openai.rb

@@ -121,6 +121,52 @@ module LlmProviders
         usage = {}
         raw_chunks = ""
         stream_error = nil
+        line_buffer = ""
+
+        process_sse_line = proc do |line|
+          next unless line.start_with?("data: ")
+
+          data = line.sub("data: ", "").strip
+          next if data == "[DONE]" || data.empty?
+
+          begin
+            event = JSON.parse(data)
+
+            if event["error"]
+              stream_error = event.dig("error", "message") || event["error"].to_s
+              next
+            end
+
+            if event["usage"]
+              usage = {
+                input: event.dig("usage", "prompt_tokens"),
+                output: event.dig("usage", "completion_tokens"),
+                cached_input: event.dig("usage", "prompt_tokens_details", "cached_tokens")
+              }
+            end
+
+            choice = event.dig("choices", 0)
+            next unless choice
+
+            delta = choice["delta"]
+            next unless delta
+
+            if delta["content"]
+              full_content += delta["content"]
+              block.call(content: delta["content"])
+            end
+
+            delta["tool_calls"]&.each do |tc|
+              idx = tc["index"]
+              tool_calls[idx] ||= { id: "", name: "", arguments: "" }
+              tool_calls[idx][:id] = tc["id"] if tc["id"]
+              tool_calls[idx][:name] = tc.dig("function", "name") if tc.dig("function", "name")
+              tool_calls[idx][:arguments] += tc.dig("function", "arguments").to_s
+            end
+          rescue JSON::ParserError
+            # Skip invalid JSON
+          end
+        end
 
         conn = Faraday.new do |f|
           f.options.open_timeout = 10
@@ -135,17 +181,17 @@ module LlmProviders
           req.body = payload.to_json
           req.options.on_data = proc do |chunk, _|
             raw_chunks += chunk
-            process_stream_chunk(chunk, full_content, tool_calls) do |parsed|
-              if parsed[:content]
-                full_content += parsed[:content]
-                block.call(content: parsed[:content])
-              end
-              stream_error = parsed[:error] if parsed[:error]
-              usage = parsed[:usage] if parsed[:usage]
-            end
+            line_buffer += chunk
+            lines = line_buffer.split("\n", -1)
+            line_buffer = lines.pop || ""
+
+            lines.each(&process_sse_line)
           end
         end
 
+        # Process any remaining data in the buffer
+        process_sse_line.call(line_buffer) unless line_buffer.empty?
+
         raise ProviderError.new(stream_error, code: "openai_error") if stream_error
 
         unless response.success?
@@ -174,50 +220,6 @@ module LlmProviders
         }
       end
 
-      def process_stream_chunk(chunk, _full_content, tool_calls)
-        chunk.each_line do |line|
-          next unless line.start_with?("data: ")
-
-          data = line.sub("data: ", "").strip
-          next if data == "[DONE]"
-
-          begin
-            event = JSON.parse(data)
-
-            if event["error"]
-              yield(error: event.dig("error", "message") || event["error"].to_s)
-              next
-            end
-
-            if event["usage"]
-              yield(usage: {
-                input: event.dig("usage", "prompt_tokens"),
-                output: event.dig("usage", "completion_tokens"),
-                cached_input: event.dig("usage", "prompt_tokens_details", "cached_tokens")
-              })
-            end
-
-            choice = event.dig("choices", 0)
-            next unless choice
-
-            delta = choice["delta"]
-            next unless delta
-
-            yield(content: delta["content"]) if delta["content"]
-
-            delta["tool_calls"]&.each do |tc|
-              idx = tc["index"]
-              tool_calls[idx] ||= { id: "", name: "", arguments: "" }
-              tool_calls[idx][:id] = tc["id"] if tc["id"]
-              tool_calls[idx][:name] = tc.dig("function", "name") if tc.dig("function", "name")
-              tool_calls[idx][:arguments] += tc.dig("function", "arguments").to_s
-            end
-          rescue JSON::ParserError
-            # Skip invalid JSON
-          end
-        end
-      end
-
       def sync_response(payload)
         started_at = Time.now
 

data/lib/llm_providers/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module LlmProviders
-  VERSION = "0.1.0"
+  VERSION = "0.1.1"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: llm_providers
 version: !ruby/object:Gem::Version
-  version: 0.1.0
+  version: 0.1.1
 platform: ruby
 authors:
 - kaba