Fix crash on iOS due to symbol resolution. (#4162)

Fixes #2588

cmake/asio.cmake

@@ -1,8 +1,9 @@
+include_guard(GLOBAL)
+
 function(download_asio)
   include(FetchContent)
 
   set(asio_URL  "https://github.com/chriskohlhoff/asio/archive/refs/tags/asio-1-24-0.tar.gz")
-  set(asio_URL2  "https://hf-mirror.com/csukuangfj/sherpa-onnx-cmake-deps/resolve/main/asio-asio-1-24-0.tar.gz")
   set(asio_HASH "SHA256=cbcaaba0f66722787b1a7c33afe1befb3a012b5af3ad7da7ff0f6b8c9b7a8a5b")
 
   # If you don't have access to the Internet,
@@ -20,7 +21,6 @@ function(download_asio)
       set(asio_URL  "${f}")
       file(TO_CMAKE_PATH "${asio_URL}" asio_URL)
       message(STATUS "Found local downloaded asio: ${asio_URL}")
-      set(asio_URL2)
       break()
     endif()
   endforeach()
@@ -28,7 +28,6 @@ function(download_asio)
   FetchContent_Declare(asio
     URL
       ${asio_URL}
-      ${asio_URL2}
     URL_HASH          ${asio_HASH}
   )
 

cmake/download-all-deps.py

@@ -71,13 +71,12 @@ def get_cmake_dir():
 def get_hardcoded_deps():
     """Return deps that don't have a cmake file (e.g., kaldifst).
 
-    Returns list of (name, url, url2, sha256).
+    Returns list of (name, url, sha256).
     """
     return [
         (
             "kaldifst",
             "https://github.com/k2-fsa/kaldifst/archive/refs/tags/v1.8.0.tar.gz",
-            "",
             "3f247b7e5a2409071202f5e2bc6200060f66728c0a3443c03923ad2723e040b3",
         ),
     ]
@@ -86,22 +85,18 @@ def get_hardcoded_deps():
 def parse_url_and_hash(filepath):
     """Parse a cmake file for URL and SHA256 hash pairs.
 
-    Returns a list of (name, url, url2, sha256) tuples.
+    Returns a list of (name, url, sha256) tuples.
     """
     text = filepath.read_text()
     deps = []
 
     # Pattern: set(name_URL  "https://...")
-    #          set(name_URL2 "https://...")  (optional)
     #          set(name_HASH "SHA256=...")
     # Use [\w-]+ to match names with hyphens (e.g., simple-sentencepiece)
     name_re = r'([\w-]+?)'
     url_pattern = re.compile(
         rf'set\({name_re}_URL\s+"([^"]+)"\)', re.MULTILINE
     )
-    url2_pattern = re.compile(
-        rf'set\({name_re}_URL2\s+"([^"]+)"\)', re.MULTILINE
-    )
     hash_pattern = re.compile(
         rf'set\({name_re}_HASH\s+"SHA256=([^"]+)"\)', re.MULTILINE
     )
@@ -113,14 +108,12 @@ def parse_url_and_hash(filepath):
         if "${" not in url and name not in urls:
             urls[name] = url
 
-    url2s = {m.group(1): m.group(2) for m in url2_pattern.finditer(text)}
     hashes = {m.group(1): m.group(2) for m in hash_pattern.finditer(text)}
 
     for name, url in urls.items():
         sha256 = hashes.get(name, "")
-        url2 = url2s.get(name, "")
         if sha256:
-            deps.append((name, url, url2, sha256))
+            deps.append((name, url, sha256))
 
     return deps
 
@@ -170,7 +163,7 @@ def parse_windows_onnxruntime(filepath, crt, build_type):
 def discover_common_deps(cmake_dir):
     """Discover platform-independent deps from cmake files.
 
-    Returns list of (name, url, url2, sha256).
+    Returns list of (name, url, sha256).
     """
     # These cmake files contain platform-independent deps
     dep_files = [
@@ -230,7 +223,7 @@ def ask_yes_no(question, default="y"):
 def select_onnxruntime_deps(cmake_dir):
     """Interactively select onnxruntime deps based on platform.
 
-    Returns list of (name, url, url2, sha256).
+    Returns list of (name, url, sha256).
     """
     os_choice = select_menu(
         "Select target OS:",
@@ -389,17 +382,14 @@ def parse_aarch64_gpu_deps(path):
 
         # Check for URL override in this block
         url_m = re.search(r'set\(onnxruntime_URL\s+"(https://[^"]+)"\)', block)
-        url2_m = re.search(r'set\(onnxruntime_URL2\s+"(https://[^"]+)"\)', block)
 
         if url_m:
             url = url_m.group(1)
-            url2 = url2_m.group(1) if url2_m else ""
         else:
             # Use template
             url = url_template.format(v=version)
-            url2 = ""
 
-        deps.append(("onnxruntime", url, url2, sha256))
+        deps.append(("onnxruntime", url, sha256))
 
     return deps
 
@@ -524,7 +514,7 @@ def main():
     # Discover common deps
     common_deps = get_hardcoded_deps() + discover_common_deps(cmake_dir)
     print(f"\nFound {len(common_deps)} common (platform-independent) dependencies:")
-    for name, url, url2, sha256 in common_deps:
+    for name, url, sha256 in common_deps:
         print(f"  {name}: {filename_from_url(url, name, cmake_dir)}")
 
     # Ask about onnxruntime
@@ -533,7 +523,7 @@ def main():
         ort_deps = select_onnxruntime_deps(cmake_dir)
         if ort_deps:
             print(f"\nSelected {len(ort_deps)} onnxruntime file(s):")
-            for name, url, url2, sha256 in ort_deps:
+            for name, url, sha256 in ort_deps:
                 print(f"  {name}: {filename_from_url(url, name, cmake_dir)}")
     else:
         print("\nSkipping onnxruntime download.")
@@ -560,18 +550,12 @@ def main():
     print(f"\nDownloading {len(all_deps)} file(s) to {dl_path}...")
     downloaded = []
     failed = []
-    for name, url, url2, sha256 in all_deps:
+    for name, url, sha256 in all_deps:
         fname = filename_from_url(url, name, cmake_dir)
         filepath = dl_path / fname
         if download_file(url, filepath, sha256):
             downloaded.append(fname)
         else:
-            # Try mirror URL
-            if url2:
-                print(f"  Trying mirror: {url2}")
-                if download_file(url2, filepath, sha256):
-                    downloaded.append(fname)
-                    continue
             failed.append(fname)
 
     # Summary

cmake/eigen.cmake

@@ -1,3 +1,5 @@
+include_guard(GLOBAL)
+
 function(download_eigen)
   include(FetchContent)
 

cmake/googletest.cmake

@@ -1,8 +1,9 @@
+include_guard(GLOBAL)
+
 function(download_googltest)
   include(FetchContent)
 
   set(googletest_URL  "https://github.com/google/googletest/archive/refs/tags/v1.13.0.tar.gz")
-  set(googletest_URL2 "https://hf-mirror.com/csukuangfj/sherpa-onnx-cmake-deps/resolve/main/googletest-1.13.0.tar.gz")
   set(googletest_HASH "SHA256=ad7fdba11ea011c1d925b3289cf4af2c66a352e18d4c7264392fead75e919363")
 
   # If you don't have access to the Internet,
@@ -20,7 +21,6 @@ function(download_googltest)
       set(googletest_URL  "${f}")
       file(TO_CMAKE_PATH "${googletest_URL}" googletest_URL)
       message(STATUS "Found local downloaded googletest: ${googletest_URL}")
-      set(googletest_URL2)
       break()
     endif()
   endforeach()
@@ -33,7 +33,6 @@ function(download_googltest)
   FetchContent_Declare(googletest
     URL
       ${googletest_URL}
-      ${googletest_URL2}
     URL_HASH          ${googletest_HASH}
   )
 

cmake/hclust-cpp.cmake

@@ -1,9 +1,10 @@
+include_guard(GLOBAL)
+
 function(download_hclust_cpp)
   include(FetchContent)
 
   # The latest release as of 2026.02.25
   set(hclust_cpp_URL  "https://github.com/csukuangfj/hclust-cpp/archive/refs/tags/2026-02-25.tar.gz")
-  set(hclust_cpp_URL2 "https://hf-mirror.com/csukuangfj/sherpa-onnx-cmake-deps/resolve/main/hclust-cpp-2026-02-25.tar.gz")
   set(hclust_cpp_HASH "SHA256=8f14e024c709d73afb40ae69cb22de4b73dba67cbce40f2e518813da8139ab56")
 
   # If you don't have access to the Internet,
@@ -21,7 +22,6 @@ function(download_hclust_cpp)
       set(hclust_cpp_URL  "${f}")
       file(TO_CMAKE_PATH "${hclust_cpp_URL}" hclust_cpp_URL)
       message(STATUS "Found local downloaded hclust_cpp: ${hclust_cpp_URL}")
-      set(hclust_cpp_URL2)
       break()
     endif()
   endforeach()
@@ -29,7 +29,6 @@ function(download_hclust_cpp)
   FetchContent_Declare(hclust_cpp
     URL
       ${hclust_cpp_URL}
-      ${hclust_cpp_URL2}
     URL_HASH          ${hclust_cpp_HASH}
   )
 

cmake/json.cmake

@@ -1,8 +1,9 @@
+include_guard(GLOBAL)
+
 function(download_json)
   include(FetchContent)
 
   set(json_URL  "https://github.com/nlohmann/json/archive/refs/tags/v3.12.0.tar.gz")
-  set(json_URL2 "https://hf-mirror.com/csukuangfj/sherpa-onnx-cmake-deps/resolve/main/json-3.12.0.tar.gz")
   set(json_HASH "SHA256=4b92eb0c06d10683f7447ce9406cb97cd4b453be18d7279320f7b2f025c10187")
 
   # If you don't have access to the Internet,
@@ -20,13 +21,12 @@ function(download_json)
       set(json_URL  "${f}")
       file(TO_CMAKE_PATH "${json_URL}" json_URL)
       message(STATUS "Found local downloaded json: ${json_URL}")
-      set(json_URL2)
       break()
     endif()
   endforeach()
 
   FetchContent_Declare(json
-    URL               ${json_URL} ${json_URL2}
+    URL               ${json_URL}
     URL_HASH          ${json_HASH}
   )
 

cmake/kaldi-decoder.cmake

@@ -1,3 +1,5 @@
+include_guard(GLOBAL)
+
 function(download_kaldi_decoder)
   include(FetchContent)
 

cmake/kaldi-native-fbank.cmake

@@ -1,8 +1,10 @@
+include_guard(GLOBAL)
+
 function(download_kaldi_native_fbank)
   include(FetchContent)
 
-  set(kaldi_native_fbank_URL   "https://github.com/csukuangfj/kaldi-native-fbank/archive/refs/tags/v1.22.4.tar.gz")
+  set(kaldi_native_fbank_URL   "https://github.com/csukuangfj/kaldi-native-fbank/archive/refs/tags/v1.23.0.tar.gz")
-  set(kaldi_native_fbank_HASH "SHA256=257fea639585ca6ee00860493bb9d7ab32488b8a5407ffe434c1f361ea545fae")
+  set(kaldi_native_fbank_HASH "SHA256=97e75ad01f486c45d6b0169a305e1e057f660fe84eb65db1aab773b732335062")
 
   set(KALDI_NATIVE_FBANK_BUILD_TESTS OFF CACHE BOOL "" FORCE)
   set(KALDI_NATIVE_FBANK_BUILD_PYTHON OFF CACHE BOOL "" FORCE)
@@ -11,11 +13,11 @@ function(download_kaldi_native_fbank)
   # If you don't have access to the Internet,
   # please pre-download kaldi-native-fbank
   set(possible_file_locations
-    $ENV{HOME}/Downloads/kaldi-native-fbank-1.22.4.tar.gz
+    $ENV{HOME}/Downloads/kaldi-native-fbank-1.23.0.tar.gz
-    ${CMAKE_SOURCE_DIR}/kaldi-native-fbank-1.22.4.tar.gz
+    ${CMAKE_SOURCE_DIR}/kaldi-native-fbank-1.23.0.tar.gz
-    ${CMAKE_BINARY_DIR}/kaldi-native-fbank-1.22.4.tar.gz
+    ${CMAKE_BINARY_DIR}/kaldi-native-fbank-1.23.0.tar.gz
-    /tmp/kaldi-native-fbank-1.22.4.tar.gz
+    /tmp/kaldi-native-fbank-1.23.0.tar.gz
-    /star-fj/fangjun/download/github/kaldi-native-fbank-1.22.4.tar.gz
+    /star-fj/fangjun/download/github/kaldi-native-fbank-1.23.0.tar.gz
   )
 
   foreach(f IN LISTS possible_file_locations)

cmake/onnxruntime.cmake

@@ -1,3 +1,5 @@
+include_guard(GLOBAL)
+
 # Copyright (c)  2022-2023  Xiaomi Corporation
 function(download_onnxruntime)
   include(FetchContent)

cmake/openfst.cmake

@@ -1,3 +1,5 @@
+include_guard(GLOBAL)
+
 # Copyright (c)  2020  Xiaomi Corporation (author: Fangjun Kuang)
 
 function(download_openfst)

cmake/portaudio.cmake

@@ -1,8 +1,9 @@
+include_guard(GLOBAL)
+
 function(download_portaudio)
   include(FetchContent)
 
   set(portaudio_URL  "http://files.portaudio.com/archives/pa_stable_v190700_20210406.tgz")
-  set(portaudio_URL2 "https://hf-mirror.com/csukuangfj/sherpa-onnx-cmake-deps/resolve/main/pa_stable_v190700_20210406.tgz")
   set(portaudio_HASH "SHA256=47efbf42c77c19a05d22e627d42873e991ec0c1357219c0d74ce6a2948cb2def")
 
   # If you don't have access to the Internet, please download it to your
@@ -21,7 +22,6 @@ function(download_portaudio)
       set(portaudio_URL  "${f}")
       file(TO_CMAKE_PATH "${portaudio_URL}" portaudio_URL)
       message(STATUS "Found local downloaded portaudio: ${portaudio_URL}")
-      set(portaudio_URL2)
       break()
     endif()
   endforeach()
@@ -35,7 +35,6 @@ function(download_portaudio)
   FetchContent_Declare(portaudio
     URL
       ${portaudio_URL}
-      ${portaudio_URL2}
     URL_HASH          ${portaudio_HASH}
   )
 

cmake/pybind11.cmake

@@ -1,8 +1,9 @@
+include_guard(GLOBAL)
+
 function(download_pybind11)
   include(FetchContent)
 
   set(pybind11_URL  "https://github.com/pybind/pybind11/archive/refs/tags/v3.0.0.tar.gz")
-  set(pybind11_URL2 "https://hf-mirror.com/csukuangfj/sherpa-onnx-cmake-deps/resolve/main/pybind11-3.0.0.tar.gz")
   set(pybind11_HASH "SHA256=453b1a3e2b266c3ae9da872411cadb6d693ac18063bd73226d96cfb7015a200c")
 
   # If you don't have access to the Internet,
@@ -20,7 +21,6 @@ function(download_pybind11)
       set(pybind11_URL  "${f}")
       file(TO_CMAKE_PATH "${pybind11_URL}" pybind11_URL)
       message(STATUS "Found local downloaded pybind11: ${pybind11_URL}")
-      set(pybind11_URL2)
       break()
     endif()
   endforeach()
@@ -28,7 +28,6 @@ function(download_pybind11)
   FetchContent_Declare(pybind11
     URL
       ${pybind11_URL}
-      ${pybind11_URL2}
     URL_HASH          ${pybind11_HASH}
   )
 

cmake/simple-sentencepiece.cmake

@@ -1,3 +1,5 @@
+include_guard(GLOBAL)
+
 # Write ${content} to ${path} only when it differs from what is on disk.
 # The patches below run on every configure; a plain file(WRITE) would update
 # the header's mtime each time and make/ssentencepiece (plus every dependent
@@ -18,7 +20,6 @@ function(download_simple_sentencepiece)
   include(FetchContent)
 
   set(simple-sentencepiece_URL  "https://github.com/pkufool/simple-sentencepiece/archive/refs/tags/v0.7.tar.gz")
-  set(simple-sentencepiece_URL2 "https://hf-mirror.com/csukuangfj/sherpa-onnx-cmake-deps/resolve/main/simple-sentencepiece-0.7.tar.gz")
   set(simple-sentencepiece_HASH "SHA256=1748a822060a35baa9f6609f84efc8eb54dc0e74b9ece3d82367b7119fdc75af")
 
   # If you don't have access to the Internet,
@@ -36,7 +37,6 @@ function(download_simple_sentencepiece)
       set(simple-sentencepiece_URL  "${f}")
       file(TO_CMAKE_PATH "${simple-sentencepiece_URL}" simple-sentencepiece_URL)
       message(STATUS "Found local downloaded simple-sentencepiece: ${simple-sentencepiece_URL}")
-      set(simple-sentencepiece_URL2)
       break()
     endif()
   endforeach()
@@ -47,7 +47,6 @@ function(download_simple_sentencepiece)
   FetchContent_Declare(simple-sentencepiece
     URL
       ${simple-sentencepiece_URL}
-      ${simple-sentencepiece_URL2}
     URL_HASH
       ${simple-sentencepiece_HASH}
   )

cmake/utf8proc.cmake

@@ -1,3 +1,5 @@
+include_guard(GLOBAL)
+
 function(download_utf8proc)
   include(FetchContent)
 
@@ -49,6 +51,48 @@ function(download_utf8proc)
 
   add_subdirectory(${utf8proc_SOURCE_DIR} ${utf8proc_BINARY_DIR} EXCLUDE_FROM_ALL)
 
+  # utf8proc's C symbols (utf8proc_map, utf8proc_free, ...) are generic enough
+  # that the host application may link its own copy of utf8proc (text libraries
+  # commonly bundle it). Merging two copies makes the linker bind our calls to
+  # the foreign one; if the two versions disagree (e.g. utf8proc_property_t
+  # layout), that corrupts memory at runtime — the same failure mode as the
+  # kissfft clash in https://github.com/k2-fsa/sherpa-onnx/issues/2588.
+  # Rename every public C symbol of utf8proc with a sherpa_onnx_ prefix. PUBLIC
+  # is required so our own sources see the renamed declarations.
+  target_compile_definitions(utf8proc PUBLIC
+    utf8proc_category=sherpa_onnx_utf8proc_category
+    utf8proc_category_string=sherpa_onnx_utf8proc_category_string
+    utf8proc_charwidth=sherpa_onnx_utf8proc_charwidth
+    utf8proc_charwidth_ambiguous=sherpa_onnx_utf8proc_charwidth_ambiguous
+    utf8proc_codepoint_valid=sherpa_onnx_utf8proc_codepoint_valid
+    utf8proc_decompose=sherpa_onnx_utf8proc_decompose
+    utf8proc_decompose_char=sherpa_onnx_utf8proc_decompose_char
+    utf8proc_decompose_custom=sherpa_onnx_utf8proc_decompose_custom
+    utf8proc_encode_char=sherpa_onnx_utf8proc_encode_char
+    utf8proc_errmsg=sherpa_onnx_utf8proc_errmsg
+    utf8proc_free=sherpa_onnx_utf8proc_free
+    utf8proc_get_property=sherpa_onnx_utf8proc_get_property
+    utf8proc_grapheme_break=sherpa_onnx_utf8proc_grapheme_break
+    utf8proc_grapheme_break_stateful=sherpa_onnx_utf8proc_grapheme_break_stateful
+    utf8proc_islower=sherpa_onnx_utf8proc_islower
+    utf8proc_isupper=sherpa_onnx_utf8proc_isupper
+    utf8proc_iterate=sherpa_onnx_utf8proc_iterate
+    utf8proc_map=sherpa_onnx_utf8proc_map
+    utf8proc_map_custom=sherpa_onnx_utf8proc_map_custom
+    utf8proc_NFC=sherpa_onnx_utf8proc_NFC
+    utf8proc_NFD=sherpa_onnx_utf8proc_NFD
+    utf8proc_NFKC=sherpa_onnx_utf8proc_NFKC
+    utf8proc_NFKC_Casefold=sherpa_onnx_utf8proc_NFKC_Casefold
+    utf8proc_NFKD=sherpa_onnx_utf8proc_NFKD
+    utf8proc_normalize_utf32=sherpa_onnx_utf8proc_normalize_utf32
+    utf8proc_reencode=sherpa_onnx_utf8proc_reencode
+    utf8proc_tolower=sherpa_onnx_utf8proc_tolower
+    utf8proc_totitle=sherpa_onnx_utf8proc_totitle
+    utf8proc_toupper=sherpa_onnx_utf8proc_toupper
+    utf8proc_unicode_version=sherpa_onnx_utf8proc_unicode_version
+    utf8proc_version=sherpa_onnx_utf8proc_version
+  )
+
   # utf8proc renames the static library to utf8proc_static under MSVC, which
   # would make the artifact name platform-dependent. Force it back to
   # "utf8proc" everywhere so -lutf8proc / utf8proc.lib / {$linklib utf8proc}

cmake/websocketpp.cmake

@@ -1,9 +1,10 @@
+include_guard(GLOBAL)
+
 function(download_websocketpp)
   include(FetchContent)
 
   # The latest commit on the develop branch os as 2022-10-22
   set(websocketpp_URL  "https://github.com/zaphoyd/websocketpp/archive/b9aeec6eaf3d5610503439b4fae3581d9aff08e8.zip")
-  set(websocketpp_URL2  "https://hf-mirror.com/csukuangfj/sherpa-onnx-cmake-deps/resolve/main/websocketpp-b9aeec6eaf3d5610503439b4fae3581d9aff08e8.zip")
   set(websocketpp_HASH "SHA256=1385135ede8191a7fbef9ec8099e3c5a673d48df0c143958216cd1690567f583")
 
   # If you don't have access to the Internet,
@@ -21,7 +22,6 @@ function(download_websocketpp)
       set(websocketpp_URL  "${f}")
       file(TO_CMAKE_PATH "${websocketpp_URL}" websocketpp_URL)
       message(STATUS "Found local downloaded websocketpp: ${websocketpp_URL}")
-      set(websocketpp_URL2)
       break()
     endif()
   endforeach()
@@ -29,7 +29,6 @@ function(download_websocketpp)
   FetchContent_Declare(websocketpp
     URL
       ${websocketpp_URL}
-      ${websocketpp_URL2}
     URL_HASH          ${websocketpp_HASH}
   )
 

scripts/check_style_cpplint.sh

@@ -75,7 +75,8 @@ function check_style() {
   if [[ $1 == mfc-example* ]]; then
     return
   fi
-  python3 $cpplint_src $1 || abort $1
+  # Silence vendored-cpplint DeprecationWarnings (sre_compile, codecs.open ...)
+  python3 -W ignore::DeprecationWarning $cpplint_src $1 || abort $1
 }
 
 function check_last_commit() {

sherpa-onnx/c-api/sherpa-onnx-symbols-c.exp

@@ -113,6 +113,7 @@ _SherpaOnnxOfflineSpeakerDiarizationResultSortByStartTime
 _SherpaOnnxOfflineSpeakerDiarizationSetConfig
 _SherpaOnnxOfflineSpeechDenoiserGetSampleRate
 _SherpaOnnxOfflineSpeechDenoiserRun
+_SherpaOnnxOfflineSpeechDenoiserSetVoiceSample
 _SherpaOnnxOfflineStreamGetOption
 _SherpaOnnxOfflineStreamHasOption
 _SherpaOnnxOfflineStreamSetOption
@@ -133,6 +134,7 @@ _SherpaOnnxOnlineSpeechDenoiserGetFrameShiftInSamples
 _SherpaOnnxOnlineSpeechDenoiserGetSampleRate
 _SherpaOnnxOnlineSpeechDenoiserReset
 _SherpaOnnxOnlineSpeechDenoiserRun
+_SherpaOnnxOnlineSpeechDenoiserSetVoiceSample
 _SherpaOnnxOnlineStreamAcceptWaveform
 _SherpaOnnxOnlineStreamGetOption
 _SherpaOnnxOnlineStreamHasOption