Merge pull request #58 from ConvertAPI/fix-parallel-uploads

Upload Files in threads, not processes; test on Python 3.14

Author: tomasr78

.github/workflows/build.yml

@@ -31,7 +31,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        python: ['3.10', '3.11', '3.12', '3.13']
+        python: ['3.10', '3.11', '3.12', '3.13', '3.14']
     steps:
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:

CHANGELOG.md

@@ -4,13 +4,19 @@ All notable changes to this project will be documented in this file.
 
 The format is based on Keep a Changelog and this project adheres to Semantic Versioning.
 
-## [2.0.1] — Unreleased
+## [2.0.1] — 2026-10-20
+
+### Fixed
+- Converting several local files at once (`Files`) failed with `TypeError: can only concatenate str (not
+  "NoneType") to str` on macOS and on Python 3.14+ on Linux: the parallel uploads ran in child processes,
+  which do not inherit `api_credentials` under the spawn and forkserver start methods. They now run in
+  threads, honouring `max_parallel_uploads` as before.
 
 ### Changed
 - Packaging moved from `setup.py` to `pyproject.toml`. The PyPI page now shows this README and links to
   the source, changelog and issues; the wheel is `py3-none-any` (it was tagged `py2.py3`, although
   Python 2 has not been supported since 1.7.0). No change to the library's API or dependencies.
-- Tested on Python 3.10–3.13 (was 3.7 and 3.9), with pytest instead of nose. The tests run without network
+- Tested on Python 3.10–3.14 (was 3.7 and 3.9), with pytest instead of nose. The tests run without network
   access; the ones that call the real service are marked `live`.
 - Releases are now cut from a GitHub Release `v{version}`; the package version comes from the tag and is
   uploaded with PyPI Trusted Publishing. See [RELEASING.md](RELEASING.md).

README.md

@@ -20,7 +20,7 @@ Install from source with:
 
 ### Requirements
 
-* Python 3.10+ (tested on 3.10–3.13)
+* Python 3 with `requests`; tested on Python 3.10–3.14
 
 ## Usage
 

RELEASING.md

@@ -2,7 +2,7 @@
 
 `main` is the only long-lived branch. Changes reach it through pull requests, merge commits only, and
 every pull request and push to `main` must pass the `build` check (`.github/workflows/build.yml`: the
-tests on Python 3.10–3.13, then build and check the sdist and wheel). Merging publishes nothing.
+tests on Python 3.10–3.14, then build and check the sdist and wheel). Merging publishes nothing.
 
 A release is a **GitHub Release** whose tag is the version, published by `tomasr78`:
 

convertapi/utils.py

@@ -1,12 +1,10 @@
-import multiprocessing
+from concurrent.futures import ThreadPoolExecutor
 
+# Uploads wait on the network, so threads are enough, and unlike processes they share the caller's
+# configuration (api_credentials, base_uri, ...) whatever multiprocessing start method Python uses.
 def map_in_parallel(f, values, pool_size):
     if pool_size < 2:
-        return map(f, values)
+        return list(map(f, values))
 
-    pool = multiprocessing.Pool(pool_size)
-    results = pool.map_async(f, values)
-    pool.close()
-    pool.join()
-
-    return results.get()
+    with ThreadPoolExecutor(max_workers = pool_size) as pool:
+        return list(pool.map(f, values))

pyproject.toml

@@ -22,6 +22,7 @@ classifiers = [
     "Programming Language :: Python :: 3.11",
     "Programming Language :: Python :: 3.12",
     "Programming Language :: Python :: 3.13",
+    "Programming Language :: Python :: 3.14",
     "Topic :: Software Development :: Libraries :: Python Modules",
 ]
 

tests/test_client.py

@@ -130,3 +130,22 @@ def test_http_error_without_json_body():
 		convertapi.convert('pdf', { 'Url': 'https://example.com' })
 
 	assert '502' in str(error.value)
+
+
+@pytest.mark.parametrize('max_parallel_uploads', [1, 10])
+@responses.activate
+def test_convert_files_uploads_each(max_parallel_uploads):
+	# Each upload must carry the caller's configuration: in 2.0.0 parallel uploads ran in child processes,
+	# which lost api_credentials under the spawn and forkserver start methods (macOS, Python 3.14 on Linux).
+	convertapi.max_parallel_uploads = max_parallel_uploads
+	responses.post(BASE + 'upload', json={ 'FileId': 'abc', 'FileExt': 'docx' })
+	responses.post(BASE + 'convert/any/to/zip', json=conversion('test.zip'))
+
+	convertapi.convert('zip', { 'Files': ['examples/files/test.docx', 'examples/files/test.pdf'] })
+
+	uploads = [call.request for call in responses.calls if call.request.url == BASE + 'upload']
+	assert [r.headers['Authorization'] for r in uploads] == ['Bearer test-token'] * 2
+	assert sorted(r.headers['Content-Disposition'] for r in uploads) == [
+		"attachment; filename*=UTF-8''test.docx", "attachment; filename*=UTF-8''test.pdf"]
+	sent = form(responses.calls[-1])
+	assert sent['Files[0]'] == 'abc' and sent['Files[1]'] == 'abc'