♻️ switch to 'Transfer-Encoding: chunked' to prevent Net::HTTP from writing temporary files (#189)

Author: sebastianMindee

lib/mindee/http/endpoint.rb

@@ -128,6 +128,8 @@ def predict_req_post(input_source, opts)
         form_data.push ['include_mvision', 'true'] if opts.all_words
 
         req.set_form(form_data, 'multipart/form-data')
+        req['Transfer-Encoding'] = 'chunked'
+
         Net::HTTP.start(uri.hostname, uri.port, use_ssl: true, read_timeout: @request_timeout) do |http|
           return http.request(req)
         end
@@ -163,6 +165,7 @@ def document_queue_req_post(input_source, opts)
         form_data.push ['include_mvision', 'true'] if opts.all_words
 
         req.set_form(form_data, 'multipart/form-data')
+        req['Transfer-Encoding'] = 'chunked'
 
         Net::HTTP.start(uri.hostname, uri.port, use_ssl: true, read_timeout: @request_timeout) do |http|
           return http.request(req)

lib/mindee/http/workflow_endpoint.rb

@@ -69,6 +69,7 @@ def workflow_execution_req_post(input_source, opts)
         form_data.push ['priority', opts.priority.to_s] if opts.priority
 
         req.set_form(form_data, 'multipart/form-data')
+        req['Transfer-Encoding'] = 'chunked'
 
         response = nil
         Net::HTTP.start(uri.hostname, uri.port, use_ssl: true, read_timeout: @request_timeout) do |http|

sig/custom/net_http.rbs

@@ -15,6 +15,7 @@ module Net
       def initialize: (untyped, Hash[String, String]?) -> void
       def set_form: (untyped, String?) -> void
       def new: (untyped, untyped) -> void
+      def []=: (?untyped, ?untyped) -> bool
     end
 
     # Stub for the HTTP GET request class.

spec/input/sources/url_input_source_integration.rb

@@ -3,9 +3,9 @@
 require 'mindee'
 
 describe Mindee::Input::Source::URLInputSource do
+  let(:client) { Mindee::Client.new(api_key: ENV.fetch('MINDEE_API_KEY')) }
+
   it 'retrieves response from a remote file' do
-    api_key = ENV.fetch('MINDEE_API_KEY', nil)
-    client = Mindee::Client.new(api_key: api_key)
     remote_input = Mindee::Input::Source::URLInputSource.new('https://github.com/mindee/client-lib-test-data/blob/main/products/invoice_splitter/invoice_5p.pdf?raw=true')
 
     local_input = remote_input.as_local_input_source
@@ -14,4 +14,17 @@
     result = client.parse(local_input, Mindee::Product::Invoice::InvoiceV4)
     expect(result.document.n_pages).to eq(5)
   end
+
+  it 'streams with chunked transfer‐encoding without creating temp files' do
+    remote_input = Mindee::Input::Source::URLInputSource
+                   .new('https://upload.wikimedia.org/wikipedia/commons/1/1d/Blank_Page.pdf')
+    allow(Tempfile).to receive(:new).and_call_original
+    allow(Tempfile).to receive(:create).and_call_original
+
+    result = client.parse(remote_input, Mindee::Product::Invoice::InvoiceV4)
+
+    expect(result.document.n_pages).to eq(1)
+    expect(Tempfile).not_to have_received(:new)
+    expect(Tempfile).not_to have_received(:create)
+  end
 end