Strip zero-width or unprintable unicode characters

.github/actions/workflow-build/build-workflow.py

@@ -1267,7 +1267,7 @@ def process_job_array(group_name, array_name, parent_json):
         )
     }
 
-    runner_heading = f"🏃‍ Runner counts (total jobs: {total_jobs})"
+    runner_heading = f"🏃 Runner counts (total jobs: {total_jobs})"
 
     runner_counts_table = f"| {'#':^4} | Runner\n"
     runner_counts_table += "|------|------\n"

.pre-commit-config.yaml

@@ -94,6 +94,14 @@ repos:
                "python/cuda_cccl/cuda/compute/"]
         pass_filenames: false
 
+  - repo: https://github.com/sirosen/texthooks
+    rev: 0.7.1
+    hooks:
+      - id: fix-smartquotes
+      - id: fix-spaces
+      - id: forbid-bidi-controls
+      - id: fix-ligatures
+
   - repo: local
     hooks:
       - id: check-shebang
@@ -106,5 +114,15 @@ repos:
             ^.*libcudacxx/cmake/config\.guess$
           )
 
+      - id: unprintable-unicode
+        name: unprintable-unicode
+        entry: ci/util/pre-commit/strip_unprintable.bash
+        language: unsupported_script
+        types: [text]
+        exclude: |
+          (?x)^(
+            ^.*ci/util/pre-commit/strip_unprintable\.bash$
+          )
+
 default_language_version:
   python: python3

AGENTS.md

@@ -334,7 +334,7 @@ CCCL's CI is built on GitHub Actions and relies on a dynamically generated job m
   * Declares build and test jobs for `pull_request`, `nightly`, and `weekly` workflows.
   * Pull request (PR) runs typically spawn ~250 jobs.
   * To reduce overhead, you can add an override matrix in `workflows.override`. This limits the PR CI run to a targeted subset of jobs. Overrides are recommended when:
-    * Changes touch high-dependency areas (e.g. top-level CI/devcontainers, libcudacxx, thrust, CUB). See `ci/inspect_changes.py` for dependency information.
+    * Changes touch high-dependency areas (e.g. top-level CI/devcontainers, libcudacxx, thrust, CUB). See `ci/inspect_changes.py` for dependency information.
     * A smaller subset of jobs is enough to validate the change (e.g. infra changes, targeted fixes).
   * Important rules:
     * PR merges are blocked while an override matrix is active.

ci/util/pre-commit/strip_unprintable.bash

@@ -0,0 +1,100 @@
+#!/usr/bin/env bash
+#
+# strip_unprintable.bash - Remove invisible / unprintable characters from text files.
+
+set -euo pipefail
+
+PROG=$(basename "$0")
+
+# Canonical definition of what gets removed. One row per range group:
+#   <perl character-class fragment> <tab> <human description>
+# The Perl class and the --help listing are both derived from this table, so
+# adding or removing a range only needs to happen here.
+ranges() {
+  printf '%s\t%s\n' \
+    '\x{0000}-\x{0008}\x{000B}\x{000C}\x{000E}-\x{001F}\x{007F}' 'C0 controls / DEL (TAB, LF, CR preserved)' \
+    '\x{0080}-\x{009F}'                                          'C1 controls' \
+    '\x{00A0}'                                                   'no-break space' \
+    '\x{200B}-\x{200F}'                                          'zero-width space/joiners, bidi marks' \
+    '\x{202A}-\x{202E}'                                          'bidi embedding/override' \
+    '\x{2060}-\x{2064}'                                          'word joiner, invisible operators' \
+    '\x{FEFF}'                                                   'BOM / zero-width no-break space'
+}
+
+# Perl character class assembled from column 1 of the ranges table.
+BAD_CLASS="[$(ranges | cut -f1 | tr -d '\n')]"
+PERL="${PERL:-perl}"
+
+usage() {
+  cat <<EOF
+Usage: ${PROG} [--check] FILE [FILE...]
+
+Remove invisible / unprintable characters from text files, in place, while
+preserving ordinary whitespace (TAB U+0009, LF U+000A, CR U+000D).
+
+Options:
+  --check     Report offending files and their line/column locations; make no
+              edits. Exits non-zero if any unprintable characters are found.
+  -h, --help  Show this help and exit.
+
+Removed characters:
+$(ranges | sed 's/^/  /')
+EOF
+}
+
+check_only=0
+
+# Parse options. Stop at the first non-option; everything after is a file.
+while [[ $# -gt 0 ]]; do
+  case "${1}" in
+    -h | --help)
+      usage
+      exit 0
+      ;;
+    --check)
+      check_only=1
+      shift
+      ;;
+    --)
+      shift
+      break
+      ;;
+    -*)
+      echo "error: unknown option: ${1}" >&2
+      usage >&2
+      exit 2
+      ;;
+    *)
+      break
+      ;;
+  esac
+done
+
+if [[ $# -eq 0 ]]; then
+  echo "error: no files given" >&2
+  usage >&2
+  exit 2
+fi
+
+status=0
+# Both modes process every file in a single perl process to avoid per-file interpreter
+# startup overhead. In --check mode, `close ARGV if eof` resets the line counter ($.) at
+# each file boundary so reported line numbers are per-file.
+if [[ ${check_only} -eq 1 ]]; then
+  # $cls/$ARGV/$./$& are Perl variables and must stay literal inside the single quotes;
+  # only BAD_CLASS is interpolated, via the deliberate quote break-out.
+  #
+  # shellcheck disable=SC2016
+  ${PERL} -CSD -ne '
+    BEGIN { $cls = qr/'"${BAD_CLASS}"'/; }
+    while (/$cls/g) {
+      printf "%s:%d:%d: U+%04X\n", $ARGV, $., pos() - length($&) + 1, ord($&);
+      $found = 1;
+    }
+    close ARGV if eof;
+    END { exit($found ? 1 : 0); }
+  ' "$@" || status=1
+  exit "${status}"
+fi
+
+${PERL} -CSD -i -pe 's/'"${BAD_CLASS}"'//g' "$@"

ci/windows/build_common.psm1

@@ -1,4 +1,4 @@
-Param(
+Param(
     [Parameter(Mandatory = $false)]
     [Alias("std")]
     [ValidateNotNullOrEmpty()]

ci/windows/build_cuda_cccl_python.ps1

@@ -28,8 +28,8 @@
     When set, only that version is built and the *merge* step is skipped.
 
 .PARAMETER Cuda13Image
-    Optional. The Docker image name used for a nested build of the CUDA 13
-    wheel when the outer container defaults to CUDA 12.9.  The default value
+    Optional. The Docker image name used for a nested build of the CUDA 13
+    wheel when the outer container defaults to CUDA 12.9.  The default value
     matches the RAPIDS dev‑container image that contains the required
     toolchain: `rapidsai/devcontainers:26.06-cuda13.0-cl14.44-windows2022`.
 
@@ -39,7 +39,7 @@
     Action.
 
 .EXAMPLE
-    # Build a single cuda-cccl wheel for Python 3.13 (consisting of both CUDA
+    # Build a single cuda-cccl wheel for Python 3.13 (consisting of both CUDA
     # 12 and 13 versions), and, if in CI, upload the resulting wheel as an
     # artifact.
     .\build_cuda_cccl_python.ps1 -PyVersion 3.11
@@ -155,7 +155,7 @@ ${null} = New-Item -ItemType Directory -Path (Join-Path $RepoRoot 'wheelhouse_cu
 function Invoke-Cuda13NestedBuild {
     <#
     .SYNOPSIS
-        Run the nested Docker build for CUDA 13 when we are already inside a
+        Run the nested Docker build for CUDA 13 when we are already inside a
         CUDA 12 builder image.
 
     .DESCRIPTION

ci/windows/run_cpu_bisect.ps1

@@ -1,4 +1,4 @@
-Param(
+Param(
     [Parameter(ValueFromRemainingArguments = $true)]
     [string[]]$PassthroughArgs
 )

ci/windows/run_cpu_target.ps1

@@ -1,4 +1,4 @@
-Param(
+Param(
     [Parameter(ValueFromRemainingArguments = $true)]
     [string[]]$PassthroughArgs
 )

ci/windows/run_gpu_bisect.ps1

@@ -1,4 +1,4 @@
-Param(
+Param(
     [Parameter(ValueFromRemainingArguments = $true)]
     [string[]]$PassthroughArgs
 )

ci/windows/run_gpu_target.ps1

@@ -1,4 +1,4 @@
-Param(
+Param(
     [Parameter(ValueFromRemainingArguments = $true)]
     [string[]]$PassthroughArgs
 )

cudax/include/cuda/experimental/__stf/internal/acquire_release.cuh

@@ -180,7 +180,7 @@ inline event_list task::acquire(backend_ctx_untyped& ctx)
     result.merge(ctx.get_start_events());
   }
 
-  // @@@@ TODO@@@@ explain this algorithm, and why we need to go in reversed
+  // @@@@ TODO@@@@ explain this algorithm, and why we need to go in reversed
   // order because we skipped equivalent dependencies that were stored
   // contiguously after sorting.
   instance_id_t previous_instance_id = instance_id_t::invalid;
@@ -193,7 +193,7 @@ inline event_list task::acquire(backend_ctx_untyped& ctx)
     else
     {
       assert(previous_instance_id != instance_id_t::invalid);
-      // @@@@ TODO @@@@ make a unit test to make sure we have the same instance id for different acquired_data
+      // @@@@ TODO @@@@ make a unit test to make sure we have the same instance id for different acquired_data
       // ? fprintf(stderr, "SETTING SKIPPED INSTANCE ID ... %d\n", previous_instance_id);
       it->set_instance_id(previous_instance_id);
     }

cudax/test/stf/examples/05-stencil-no-copy.cu

@@ -14,7 +14,7 @@ using namespace cuda::experimental::stf;
 
 /*
  * DATA BLOCKS
- *   | GHOSTS | DATA | GHOSTS |
+ *   | GHOSTS | DATA | GHOSTS |
  */
 template <typename T>
 class data_block

cudax/test/stf/examples/05-stencil.cu

@@ -16,7 +16,7 @@ static stream_ctx ctx;
 
 /*
  * DATA BLOCKS
- *   | GHOSTS | DATA | GHOSTS |
+ *   | GHOSTS | DATA | GHOSTS |
  */
 template <typename T>
 class data_block

cudax/test/stf/stencil/stencil-1D.cu

@@ -18,7 +18,7 @@ static graph_ctx ctx;
 
 /*
  * DATA BLOCKS
- *   | GHOSTS | DATA | GHOSTS |
+ *   | GHOSTS | DATA | GHOSTS |
  */
 template <typename T>
 class data_block

docs/cub/tuning.rst

@@ -542,7 +542,7 @@ since the names of the reported parameters in the variant are derived from these
 The variant ``ipt_19.tpb_512``, which stands for 19 items per thread (``ipt``) and 512 threads per block (``tpb``),
 was thus compiled with ``-DTUNE_ITEMS_PER_THREAD=19 -DTUNE_THREADS_PER_BLOCK=512``.
 The meaning of these values is specific to the benchmark definition,
-and we have to check the benchmark’s source code for how they are applied.
+and we have to check the benchmark's source code for how they are applied.
 Equally named tuning parameters may not translate to different benchmarks (please double check).
 
 As a user of CUB, such a new set of tuning parameters (i.e. a variant) can then be used to define a policy selector,
@@ -561,7 +561,7 @@ as :ref:`sketched above <cub-tuning-authoring-benchmarks>`:
     }
   };
 
-The default tunings defined inside CUB’s source use the same infrastructure
+The default tunings defined inside CUB's source use the same infrastructure
 but should only be changed and extended by the CCCL maintainers.
 All default tunings are found in the :code:`cub/device/detail/tuning/tuning_*.cuh` headers, organized by algorithm.
 CUB's policy selectors are highly parameterized on type information and traits of the input arguments to CUB algorithms