Use pyicu-wheels for tests via PEP 735 dependency groups; remove redundant optional-deps

.github/workflows/deploy-docs.yml

@@ -32,7 +32,9 @@ jobs:
     - name: Install build tools and doc build tools
       run: pip install --upgrade pip
     - name: Install PyThaiNLP
-      run: pip install ".[docs]"
+      run: |
+        pip install .
+        pip install --group docs
     - name: Build sphinx documentation
       run: |
         cd docs && make html

.github/workflows/pypi-test.yml

@@ -34,7 +34,8 @@ jobs:
         SKLEARN_ALLOW_DEPRECATED_SKLEARN_PACKAGE_INSTALL: True
       run: |
         python -m pip install --upgrade pip
-        python -m pip install ".[compact]"
+        python -m pip install .
+        python -m pip install --group test-compact
         python -m nltk.downloader omw-1.4
 
     # Download the sdist from PyPI, extract it, and run the tests inside it

.github/workflows/unittest.yml

@@ -74,7 +74,6 @@ jobs:
 
     runs-on: ${{ matrix.os }}
     env:
-      PYICU_WIN_VER: 2.15 # 2.15 is the last version that supports Python 3.9
       INSTALL_FULL_DEPS: false
       PYTHON_VERSION_LATEST: "3.14"
       PYTHON_VERSION_LATEST_2: "3.13" # Second-latest supported version
@@ -95,31 +94,13 @@ jobs:
           pip install --upgrade pip
           pip install coverage coveralls
 
-      - name: Install ICU (macOS)
-        if: startsWith(matrix.os, 'macos-')
-        run: |
-          brew install icu4c
-          PKG_CONFIG_PATH=$(brew --prefix)/opt/icu4c/lib/pkgconfig
-          echo "PKG_CONFIG_PATH=${PKG_CONFIG_PATH}" >> "${GITHUB_ENV}"
-          ICU_VER=$(pkg-config --modversion icu-i18n)
-          echo "ICU_VER=${ICU_VER}"
-          echo "ICU_VER=${ICU_VER}" >> "${GITHUB_ENV}"
-
-      - name: Install PyICU (Windows)
-        if: startsWith(matrix.os, 'windows-') && (matrix.python-version == env.PYTHON_VERSION_LATEST_2 || matrix.python-version == env.PYTHON_VERSION_EARLIEST)
-        shell: powershell
-        # Get the wheel URL from https://github.com/cgohlke/pyicu-build/releases
-        run: |
-          $PYTHON_WIN_VER = "${{ matrix.python-version }}"
-          $CP_VER = "cp" + $PYTHON_WIN_VER.Replace(".", "")
-          $WHEEL_URL = "https://github.com/cgohlke/pyicu-build/releases/download/v${{ env.PYICU_WIN_VER }}/PyICU-${{ env.PYICU_WIN_VER }}-${CP_VER}-${CP_VER}-win_amd64.whl"
-          pip install "$WHEEL_URL"
-
       - name: Install PyThaiNLP + full testing dependencies
         if: env.INSTALL_FULL_DEPS == 'true'
         env:
           SKLEARN_ALLOW_DEPRECATED_SKLEARN_PACKAGE_INSTALL: True
-        run: pip install ".[full]"
+        run: |
+          pip install .
+          pip install --group test-full
 
       - name: Test (core)
         if: ${{ (matrix.os == 'ubuntu-latest' && matrix.python-version != env.PYTHON_VERSION_LATEST_2 && matrix.python-version != env.PYTHON_VERSION_EARLIEST) || (matrix.os != 'ubuntu-latest' && matrix.python-version == env.PYTHON_VERSION_LATEST) }}
@@ -138,15 +119,17 @@ jobs:
         env:
           PYTHONIOENCODING: utf-8
         run: |
-          pip install ".[compact]"
+          pip install .
+          pip install --group test-compact
           python -m unittest tests.core tests.compact
 
       - name: Test (extra + compact + core)
         if: matrix.os == 'ubuntu-latest' && matrix.python-version == env.PYTHON_VERSION_LATEST_2
         env:
           PYTHONIOENCODING: utf-8
         run: |
-          pip install ".[compact,extra]"
+          pip install .
+          pip install --group test-extra
           coverage run -m unittest tests.core tests.compact tests.extra
 
       # Only submit a report from the "extra" run, to get maximum coverage

CHANGELOG.md

@@ -21,6 +21,15 @@ and this project adheres to
 
 This release focuses on security issues related to corpus file loading.
 
+### Changed
+
+- Migrate development, documentation, and testing dependencies from
+  `[project.optional-dependencies]` to PEP 735 `[dependency-groups]`
+  in `pyproject.toml`; use `pip install --group <name>` to install (#1329)
+- Use `pyicu-wheels` in test dependency groups for pre-built ICU binary
+  wheels, simplifying local development and CI on Windows and macOS (#1329)
+- Simplify CI by removing platform-specific ICU compilation steps (#1329)
+
 ### Security
 
 - thai2fit: Use JSON model instead of pickle (#1325)

pyproject.toml

@@ -68,28 +68,7 @@ dependencies = [
 
 [project.optional-dependencies]
 
-## 1) Development ########################################
-
-dev = [
-    "black>=25.11.0",
-    "build>=1.0.0",
-    "bump-my-version>=1.2.6",
-    "coverage>=7.10.7",
-    "flake8>=7.0.0",
-    "flake8-type-checking>=3.2.0",
-    "mypy>=1.19.1",
-    "pylint>=4.0.0",
-    "ruff>=0.14.14",
-    "tox>=4.30.3",
-]
-
-docs = [
-    "Sphinx>=6.2",
-    "sphinx-copybutton>=0.5.2",
-    "sphinx-rtd-theme>=3.1.0",
-]
-
-## 2) Feature sets #######################################
+## Feature sets ###########################################
 
 abbreviation = ["khamyo>=0.2.0"]
 
@@ -116,7 +95,7 @@ esupar = ["esupar>=1.3.8", "numpy>=1.22", "transformers>=4.22.1"]
 
 generate = ["fastai<2.0"]
 
-icu = ["pyicu>=2.3"]
+icu = ["pyicu>=2.14"]  # PyICU 2.14 is the last version to support Python 3.9
 
 ipa = ["epitran>=1.1"]
 
@@ -174,23 +153,55 @@ wtp = ["transformers>=4.22.1", "wtpsplit>=1.0.1"]
 
 wunsen = ["wunsen>=0.0.3"]
 
-## 3) Testing ############################################
+## Dependency groups (PEP 735) ############################
 
-# Below are dependency groups for each defined test suites
+# Use `pip install --group <group-name>` to install a group.
+# For test groups, pyicu-wheels (https://pypi.org/project/pyicu-wheels/)
+# provides pre-built ICU binary wheels for all platforms without
+# local compilation. Note: pyicu-wheels is from a third-party source
+# (Init7, not the official PyICU project) and is used for testing only.
+# Distribution dependencies still use pyicu
+# (see icu = ["pyicu>=2.3"] in optional-dependencies below).
 
-# Compact dependencies - for tests.compact modules
-# Safe small set of optional dependencies
-# for common tasks like tokenization and tagging.
-compact = [
+[dependency-groups]
+
+# Development tools
+dev = [
+    "black>=25.11.0",
+    "build>=1.0.0",
+    "bump-my-version>=1.2.6",
+    "coverage>=7.10.7",
+    "flake8>=7.0.0",
+    "flake8-type-checking>=3.2.0",
+    "mypy>=1.19.1",
+    "pylint>=4.0.0",
+    "ruff>=0.14.14",
+    "tox>=4.30.3",
+]
+
+# Documentation build tools
+docs = [
+    "Sphinx>=6.2",
+    "sphinx-copybutton>=0.5.2",
+    "sphinx-rtd-theme>=3.1.0",
+]
+
+# Dependencies for tests.core
+# No extra optional dependencies needed beyond the base package.
+test-core = []
+
+# Dependencies for tests.compact
+test-compact = [
     "nlpo3>=1.4.0",
     "numpy>=1.26.0",
-    "pyicu>=2.3",
+    "pyicu-wheels>=2.14",
     "python-crfsuite>=0.9.7",
     "PyYAML>=5.4.1",
 ]
 
-# Extra dependencies - for tests.extra modules
-extra = [
+# Dependencies for tests.extra (includes compact dependencies)
+test-extra = [
+    {include-group = "test-compact"},
     "bpemb>=0.3.2",
     "budoux>=0.7.0",
     "gensim>=4.0.0",
@@ -201,11 +212,8 @@ extra = [
     "tltk>=1.10",
 ]
 
-# Noauto test dependencies - for tests.noauto_* modules
-# These are further grouped by dependency framework to avoid conflicts
-
-# PyTorch-based dependencies - for tests.noauto_torch
-noauto-torch = [
+# Dependencies for tests.noauto_torch
+test-noauto-torch = [
     "attacut>=1.0.6",
     "numpy>=1.26.0",
     "sentencepiece>=0.1.91",
@@ -216,33 +224,36 @@ noauto-torch = [
     "wtpsplit>=1.0.1",
 ]
 
-# TensorFlow-based dependencies - for tests.noauto_tensorflow
-noauto-tensorflow = [
+# Dependencies for tests.noauto_tensorflow
+test-noauto-tensorflow = [
     "deepcut>=0.7.0",
     "numpy>=1.26.0",
 ]
 
-# ONNX Runtime-based dependencies - for tests.noauto_onnx
-noauto-onnx = [
+# Dependencies for tests.noauto_onnx
+test-noauto-onnx = [
     "numpy>=1.26.0",
     "onnxruntime>=1.10.0",
     "oskut>=1.3",
     "sefr_cut>=1.1",
 ]
 
-# Cython-based dependencies - for tests.noauto_cython
-noauto-cython = [
+# Dependencies for tests.noauto_cython
+test-noauto-cython = [
     "phunspell>=0.1.6",
 ]
 
-# Network-dependent tests - for tests.noauto_network
-# These tests require network access but minimal dependencies
-noauto-network = [
+# Dependencies for tests.noauto_network
+test-noauto-network = [
     "huggingface-hub>=0.16.0",
 ]
 
-# Full dependencies - pinned where available
-full = [
+# Dependencies for tests.corpus
+# No extra optional dependencies needed beyond the base package.
+test-corpus = []
+
+# Full pinned dependencies for all features (for reproducible testing)
+test-full = [
     "attacut==1.0.6",
     "attaparse==1.0.0",
     "bpemb>=0.3.6,<0.4",
@@ -266,7 +277,7 @@ full = [
     "pandas>=2.2.0,<3",
     "panphon==0.22.2",
     "phunspell==0.1.6",
-    "pyicu>=2.15.2,<3",
+    "pyicu-wheels>=2.14",
     "python-crfsuite==0.9.12",
     "PyYAML>=5.4.1,<6.0.2",
     "sacremoses==0.1.1",

tests/README.md

@@ -60,10 +60,20 @@ The CI/CD test workflow is at
 ## Compact tests (testc_*.py)
 
 - Run `python -m unittest tests.compact`
-  - Need dependencies from `pip install "pythainlp[compact]"`
+  - Need dependencies:
+
+    ```shell
+    pip install pythainlp
+    pip install --group test-compact
+    ```
+
+    The `test-compact` group uses `pyicu-wheels` for pre-built ICU binary
+    wheels (easier to install on all platforms without local compilation).
+    Note: `pyicu-wheels` is from a third-party source (Init7) and is used
+    for testing convenience only; distribution dependencies use `pyicu`.
 - Test a limited set of functionalities that rely on a stable
   and small set of dependencies.
-- These dependencies are `PyYAML`, `nlpo3`, `numpy`, `pyicu`,
+- These dependencies are `PyYAML`, `nlpo3`, `numpy`, `pyicu-wheels`,
   and `python-crfsuite`.
 - Includes corpus download/remove tests (may require network access).
 - Tested on:
@@ -74,7 +84,14 @@ The CI/CD test workflow is at
 ## Extra tests (testx_*.py)
 
 - Run `python -m unittest tests.extra`
-  - Need dependencies from `pip install "pythainlp[compact,extra]"`
+  - Need dependencies:
+
+    ```shell
+    pip install pythainlp
+    pip install --group test-extra
+    ```
+
+    The `test-extra` group includes all `test-compact` dependencies.
 - Test more functionalities that rely on larger set of dependencies
   or one that require more time or computation.
 - Only tested on Ubuntu using the second-latest Python version.
@@ -116,7 +133,13 @@ By separating tests by dependency group, we can:
 ##### PyTorch-based: tests.noauto_torch
 
 - Run `python -m unittest tests.noauto_torch`
-  - Need dependencies from `pip install "pythainlp[noauto-torch]"`
+  - Need dependencies:
+
+    ```shell
+    pip install pythainlp
+    pip install --group test-noauto-torch
+    ```
+
 - Tests requiring PyTorch and its ecosystem:
   - torch, transformers (PyTorch backend)
   - attacut, thai-nner, wtpsplit, tltk
@@ -129,7 +152,13 @@ By separating tests by dependency group, we can:
 ##### TensorFlow-based: tests.noauto_tensorflow
 
 - Run `python -m unittest tests.noauto_tensorflow`
-  - Need dependencies from `pip install "pythainlp[noauto-tensorflow]"`
+  - Need dependencies:
+
+    ```shell
+    pip install pythainlp
+    pip install --group test-noauto-tensorflow
+    ```
+
 - Tests requiring TensorFlow:
   - deepcut tokenizer
 - Dependencies: ~1-2 GB
@@ -139,7 +168,13 @@ By separating tests by dependency group, we can:
 ##### ONNX Runtime-based: tests.noauto_onnx
 
 - Run `python -m unittest tests.noauto_onnx`
-  - Need dependencies from `pip install "pythainlp[noauto-onnx]"`
+  - Need dependencies:
+
+    ```shell
+    pip install pythainlp
+    pip install --group test-noauto-onnx
+    ```
+
 - Tests requiring ONNX Runtime:
   - oskut, sefr_cut tokenizers
 - Dependencies: ~200-500 MB
@@ -148,7 +183,13 @@ By separating tests by dependency group, we can:
 ##### Cython-compiled: tests.noauto_cython
 
 - Run `python -m unittest tests.noauto_cython`
-  - Need dependencies from `pip install "pythainlp[noauto-cython]"`
+  - Need dependencies:
+
+    ```shell
+    pip install pythainlp
+    pip install --group test-noauto-cython
+    ```
+
 - Tests requiring Cython-compiled packages:
   - phunspell spell checker
 - Requires: Cython, C compiler, system libraries (hunspell)
@@ -158,7 +199,13 @@ By separating tests by dependency group, we can:
 ##### Network-dependent: tests.noauto_network
 
 - Run `python -m unittest tests.noauto_network`
-  - Need dependencies from `pip install "pythainlp[noauto-network]"`
+  - Need dependencies:
+
+    ```shell
+    pip install pythainlp
+    pip install --group test-noauto-network
+    ```
+
 - Tests requiring network access:
   - Hugging Face Hub model downloads
   - External API calls