Skip to content

tesserocr: add build-tesserocr.yml for riscv64 wheels #1821

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
luhenry merged 2 commits into from
Sep 14, 2026
Merged

tesserocr: add build-tesserocr.yml for riscv64 wheels #1821

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9f3e1ba
tesserocr: add build-tesserocr.yml for riscv64 wheels
luhenry Sep 12, 2026
152074b
tesserocr: allow Pillow to build from source on musllinux_riscv64
luhenry Sep 12, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
195 changes: 195 additions & 0 deletions .github/workflows/build-tesserocr.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,195 @@
# SPDX-FileCopyrightText: 2026 The RISE Project
# SPDX-License-Identifier: MIT
---
# This workflow is based on upstream's own
# https://github.com/sirfz/tesserocr/blob/v2.11.0/.github/workflows/build.yml
name: Build tesserocr wheels (riscv64)

on:
workflow_dispatch:
inputs:
version:
description: 'tesserocr version to build (git tag without the leading v, e.g. 2.11.0)'
required: true
default: '2.11.0'
pull_request:
paths:
- '.github/workflows/build-tesserocr.yml'
- 'patches/tesserocr/**'

concurrency:
group: ${{ github.workflow }}-${{ inputs.version || '2.11.0' }}-${{ github.head_ref || github.run_id }}
cancel-in-progress: true

permissions:
contents: read # to fetch code (actions/checkout)

env:
# `inputs.version` is empty on pull_request events; default to 2.11.0 there.
TESSEROCR_VERSION: ${{ inputs.version || '2.11.0' }}
MANYLINUX_RISCV64_IMAGE: quay.io/pypa/manylinux_2_39_riscv64
MUSLLINUX_RISCV64_IMAGE: quay.io/pypa/musllinux_1_2_riscv64
# Matches the pins in upstream's own .github/build-scripts/common-install-tesseract.sh.
LEPTONICA_VERSION: '1.85.0'
TESSERACT_VERSION: '5.5.1'

jobs:
setup:
uses: $/.github/workflows/_setup.yml

build_wheels:
needs: [setup]
name: Build tesserocr ${{ inputs.version || '2.11.0' }} ${{ matrix.python }}-${{ matrix.libc }}_riscv64
runs-on: ubuntu-24.04-riscv
timeout-minutes: 90
strategy:
fail-fast: false
matrix:
python: ["cp312", "cp313", "cp314", "cp314t"]
libc: [manylinux, musllinux]

steps:
- name: Checkout tesserocr v${{ env.TESSEROCR_VERSION }}
uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
with:
repository: sirfz/tesserocr
ref: v${{ env.TESSEROCR_VERSION }}
persist-credentials: false

- name: Checkout python-wheels
uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
with:
path: python-wheels
persist-credentials: false

- name: Patch tesserocr source
# 0001-*/0003-*: git apply against this checkout. 0002-* targets the
# vendored tesseract tarball (no .git dir) fetched later in
# CIBW_BEFORE_ALL_LINUX, applied there with `patch -p1` instead --
# excluded from this glob.
run: |
git apply python-wheels/patches/tesserocr/${{ env.TESSEROCR_VERSION }}/0001-*.patch
git apply python-wheels/patches/tesserocr/${{ env.TESSEROCR_VERSION }}/0003-*.patch

- uses: pypa/cibuildwheel@1828c10ab37f080699c7b81cea34097c684a7074 # v4.2.0
with:
output-dir: wheelhouse/
only: ${{ matrix.python }}-${{ matrix.libc }}_riscv64
env:
CIBW_MANYLINUX_RISCV64_IMAGE: ${{ env.MANYLINUX_RISCV64_IMAGE }}
CIBW_MUSLLINUX_RISCV64_IMAGE: ${{ env.MUSLLINUX_RISCV64_IMAGE }}
# Neither manylinux_riscv64 nor musllinux_riscv64 ships tesseract/leptonica
# devel packages (no riscv64 EPEL, and Rocky 10 carries neither at all); built
# from source at upstream's own pinned versions instead of yum-installing them,
# same shape as upstream's linux-install-tesseract.sh but via dnf/apk.
CIBW_BEFORE_ALL_LINUX: |
set -ex
if command -v apk > /dev/null; then
apk add --no-cache cmake pkgconf libpng-dev jpeg-dev tiff-dev libwebp-dev zlib-dev
else
dnf install -y cmake pkgconfig libpng-devel libjpeg-turbo-devel libtiff-devel libwebp-devel zlib-devel
fi
mkdir -p /tmp/leptonica /tmp/tesseract
curl -fsSL "https://github.com/DanBloomberg/leptonica/releases/download/$LEPTONICA_VERSION/leptonica-$LEPTONICA_VERSION.tar.gz" | tar xz --strip-components=1 -C /tmp/leptonica
curl -fsSL "https://github.com/tesseract-ocr/tesseract/archive/refs/tags/$TESSERACT_VERSION.tar.gz" | tar xz --strip-components=1 -C /tmp/tesseract
# Vendored from a tarball, not a git checkout, so `patch -p1`
# rather than `git apply` (gotcha precedent: build-rerun-sdk.yml's
# lance-core/lance-linalg crates).
patch -p1 -d /tmp/tesseract < python-wheels/patches/tesserocr/${{ env.TESSEROCR_VERSION }}/0002-tesseract-osdetect-guard-against-unset-script-id.patch
cmake -S /tmp/leptonica -B /tmp/leptonica/build \
-D CMAKE_BUILD_TYPE=RelWithDebInfo \
-D CMAKE_INSTALL_PREFIX=/usr/local \
-D CMAKE_INSTALL_LIBDIR=lib \
-D BUILD_SHARED_LIBS=ON \
-D ENABLE_GIF=OFF \
-D ENABLE_OPENJPEG=OFF
cmake --build /tmp/leptonica/build -j"$(nproc)"
cmake --install /tmp/leptonica/build
# Upstream's own workaround: leptonica's .pc file bakes the version into
# its name (lept_X.Y.Z.pc) instead of the lept.pc setup.py looks for.
mv /usr/local/lib/pkgconfig/lept_*.pc /usr/local/lib/pkgconfig/lept.pc
cmake -S /tmp/tesseract -B /tmp/tesseract/build \
-D CMAKE_BUILD_TYPE=RelWithDebInfo \
-D CMAKE_INSTALL_PREFIX=/usr/local \
-D CMAKE_INSTALL_LIBDIR=/usr/local/lib \
-D BUILD_SHARED_LIBS=ON \
-D OPENMP_BUILD=OFF \
-D BUILD_TRAINING_TOOLS=OFF
cmake --build /tmp/tesseract/build -j"$(nproc)"
cmake --install /tmp/tesseract/build
ldconfig || true
mkdir -p /usr/local/share/tessdata
curl -fsSL -o /usr/local/share/tessdata/eng.traineddata https://github.com/tesseract-ocr/tessdata_fast/raw/main/eng.traineddata
curl -fsSL -o /usr/local/share/tessdata/osd.traineddata https://github.com/tesseract-ocr/tessdata_fast/raw/main/osd.traineddata
# LEPTONICA_VERSION/TESSERACT_VERSION must be restated here: CIBW_BEFORE_ALL_LINUX
# runs inside the manylinux container via `sh -c`, which does not inherit the
# job's plain `env:` (only what CIBW_ENVIRONMENT passes through).
CIBW_ENVIRONMENT: >-
LEPTONICA_VERSION=${{ env.LEPTONICA_VERSION }}
TESSERACT_VERSION=${{ env.TESSERACT_VERSION }}
PKG_CONFIG_PATH=/usr/local/lib/pkgconfig
TESSDATA_PREFIX=/usr/local/share/tessdata
PIP_EXTRA_INDEX_URL=https://pypi.riseproject.dev/simple/
PIP_ONLY_BINARY=cysignals
# setup.py cythonizes inside build_ext (no [build-system] table to declare
# Cython as a build requirement), so build isolation is disabled and Cython
# + cysignals are preinstalled into the same environment pip builds in.
CIBW_BEFORE_BUILD: pip install "Cython>=3.0.0,<3.2.0" cysignals setuptools wheel
CIBW_BUILD_FRONTEND: "pip; args: --no-build-isolation"
# Pillow, not just cysignals: pypi.riseproject.dev has never published a
# musllinux_riscv64 Pillow wheel (manylinux_riscv64 only, plus a legacy
# untagged linux_riscv64 wheel for cp39-cp312 that pip happens to accept
# on musl too), so forcing binary-only here made cp313/cp314/cp314t-musllinux
# fail outright with "Could not find a version that satisfies the
# requirement Pillow (from versions: none)" -- confirmed via
# `curl -s https://pypi.riseproject.dev/simple/pillow/`. Left off
# PIP_ONLY_BINARY so pip falls back to Pillow's sdist there; the
# jpeg/png/tiff/webp/zlib -dev packages this step's apk/dnf install
# already puts in the same container (for leptonica/tesseract) are
# exactly what Pillow's setup.py auto-detects and needs for that build.
CIBW_TEST_REQUIRES: pytest Pillow
CIBW_TEST_SOURCES: tests
# A relative path, not {project}/tests: tesserocr has no src/ layout, so
# {project}/tests would put {project} (the full checkout) on pytest's
# rootdir-inserted sys.path and shadow the installed wheel with the
# uncompiled tesserocr/ source tree (no .so) instead.
# Upstream's own CI (build.yml) doesn't run this suite at all -- its
# "Test wheel" step is a bare `import tesserocr` smoke test -- so
# test_init's Init(oem=OEM.TESSERACT_ONLY) call is a gap upstream
# itself never had to close: it needs the legacy-engine model data
# that tessdata_fast's traineddata (LSTM-only) doesn't ship.
# test_detect_os's own brittle final assertion is patched out
# instead of deselecting the whole test (patches/tesserocr/2.11.0/
# 0003-*), since it's the exact test that was crashing before
# 0002's fix and is worth keeping as regression coverage.
CIBW_TEST_COMMAND: >-
python -m pytest tests -v
--deselect "tests/test_api.py::TestTessBaseApi::test_init"

- name: Check the wheel is compiled and carries the vendored licences
run: |
python3 - wheelhouse/*.whl <<'EOF'
import sys, zipfile
for whl in sys.argv[1:]:
names = zipfile.ZipFile(whl).namelist()
assert any(n.startswith("tesserocr/tesserocr") and n.endswith(".so") for n in names), names
for licence in ("LICENSE.leptonica", "LICENSE.tesseract-ocr"):
assert any(n.endswith(f".dist-info/licenses/{licence}") for n in names), (licence, names)
print(whl, "ok")
EOF

- uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
with:
name: tesserocr-${{ env.TESSEROCR_VERSION }}-${{ matrix.python }}-${{ matrix.libc }}_riscv64
path: wheelhouse/*.whl
if-no-files-found: error

publish:
name: Publish tesserocr ${{ inputs.version || '2.11.0' }}
needs: [setup, build_wheels]
permissions:
contents: write
pull-requests: write
uses: $/.github/workflows/_publish-wheel.yml
with:
artifact-pattern: tesserocr-${{ inputs.version || '2.11.0' }}-*riscv64
Loading