From d32195c582ff966487c12ff2b5171aa3613321de Mon Sep 17 00:00:00 2001
From: mattip <matti.picus@gmail.com>
Date: Fri, 16 Oct 2026 12:07:53 +0300
Subject: [PATCH 1/6] use a per-thread queue on windows

---
 tools/build_steps_windows.sh | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/tools/build_steps_windows.sh b/tools/build_steps_windows.sh
index bfa1b8e8..2003e85e 100644
--- a/tools/build_steps_windows.sh
+++ b/tools/build_steps_windows.sh
@@ -90,9 +90,10 @@ interface_flags="$interface_flags SYMBOLPREFIX=scipy_ LIBNAMEPREFIX=scipy_ FIXED
 GCC_TAG="gcc_$(gcc -dumpversion | tr .- _)"
 OPENBLAS_VERSION=$(git describe --tags --abbrev=8)
 
-# Patch OpenBLAS build to resolve all symbols and avoid linking
-# with libquadmath
-patch -p1 < ../patches-windows/openblas-make-libs.patch
+for p in ../patches-windows/*.patch; do
+    [ -e "$p" ] || continue
+    patch -p1 < "$p"
+done
 
 # Patch VERSION
 version=$(grep "^version =" ../pyproject.toml | sed 's/version = "//;s/"//')

From f058496d84f141743822a59b48ecb9fc0e2b3834 Mon Sep 17 00:00:00 2001
From: mattip <matti.picus@gmail.com>
Date: Fri, 16 Oct 2026 12:08:37 +0300
Subject: [PATCH 2/6] increment build number

---
 pyproject.toml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/pyproject.toml b/pyproject.toml
index 13381375..69ed4ff2 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -9,7 +9,7 @@ build-backend = "setuptools.build_meta"
 [project]
 name = "scipy-openblas64"
 # v0.3.34-237-g1ad90fb1
-version = "0.3.34.237.1"
+version = "0.3.34.237.2"
 requires-python = ">=3.7"
 description = "Provides OpenBLAS for python packaging"
 readme = "README.md"

From ca11ec4b67c77d6760916b5fe40508c2a3e94062 Mon Sep 17 00:00:00 2001
From: mattip <matti.picus@gmail.com>
Date: Fri, 16 Oct 2026 14:31:51 +0300
Subject: [PATCH 3/6] also patch on win-arm64

---
 patches-windows/openblas-make-libs.patch      |   5 +-
 .../openblas-win32-pool-queue-link.patch      | 101 ++++++++++++++++++
 tools/build_steps_win_arm64.bat               |   8 +-
 3 files changed, 109 insertions(+), 5 deletions(-)
 create mode 100644 patches-windows/openblas-win32-pool-queue-link.patch

diff --git a/patches-windows/openblas-make-libs.patch b/patches-windows/openblas-make-libs.patch
index 1e4d4aeb..9fd3ba99 100644
--- a/patches-windows/openblas-make-libs.patch
+++ b/patches-windows/openblas-make-libs.patch
@@ -1,8 +1,7 @@
 diff --git a/exports/Makefile b/exports/Makefile
-index 668a4866e..956b51bf4 100644
 --- a/exports/Makefile
 +++ b/exports/Makefile
-@@ -126,8 +126,9 @@ dll  : ../$(LIBDLLNAME)
+@@ -129,8 +129,9 @@ dll  : ../$(LIBDLLNAME)
  ../$(LIBDLLNAME) : ../$(LIBNAME) $(LIBPREFIX).def dllinit.$(SUFFIX)
  	$(RANLIB) ../$(LIBNAME)
  	$(CC) $(CFLAGS) $(LDFLAGS) $(LIBPREFIX).def dllinit.$(SUFFIX) \
@@ -12,4 +11,4 @@ index 668a4866e..956b51bf4 100644
 +	dlltool -D $(LIBDLLNAME) -d $(LIBPREFIX).def -l ../$(LIBDLLNAME).a
  
  $(LIBPREFIX).def : $(GENSYM)
- 	./$(GENSYM) win2k    $(ARCH) dummy $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) $(BUILD_BFLOAT16) $(BUILD_SINGLE) $(BUILD_DOUBLE) $(BUILD_COMPLEX) $(BUILD_COMPLEX16) > $(@F)
+ 	./$(GENSYM) win2k    $(ARCH) dummy $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) $(BUILD_BFLOAT16) $(BUILD_HFLOAT16) $(BUILD_SINGLE) $(BUILD_DOUBLE) $(BUILD_COMPLEX) $(BUILD_COMPLEX16) > $(@F)
diff --git a/patches-windows/openblas-win32-pool-queue-link.patch b/patches-windows/openblas-win32-pool-queue-link.patch
new file mode 100644
index 00000000..80251ba1
--- /dev/null
+++ b/patches-windows/openblas-win32-pool-queue-link.patch
@@ -0,0 +1,101 @@
+Fix memory corruption from concurrent BLAS calls on Windows
+
+The Windows thread server keeps one global list of jobs waiting to be picked
+up, and it links that list through blas_queue_t::next -- the same field the
+*caller* owns and rebuilds on every call.
+
+diff --git a/common_thread.h b/common_thread.h
+index 7f5d7b2..cd7ca77 100644
+--- a/common_thread.h
++++ b/common_thread.h
+@@ -119,6 +119,14 @@ typedef struct blas_queue {
+   struct blas_queue *next;
+ 
+ #if defined( __WIN32__) || defined(__CYGWIN32__) || defined(_WIN32) || defined(__CYGWIN__)
++  /*
++   * Link used exclusively by the Windows thread pool to chain the jobs that
++   * are waiting to be picked up.  It is deliberately kept separate from
++   * "next" above, which is owned (and rebuilt on every call) by whoever
++   * submitted the job: appending to the pool must never write into the
++   * queue nodes of another caller.
++   */
++  struct blas_queue *pool_next;
+    CRITICAL_SECTION lock;
+    HANDLE finish;
+   volatile int finished;
+diff --git a/driver/others/blas_server_win32.c b/driver/others/blas_server_win32.c
+index e827cee..4f24823 100644
+--- a/driver/others/blas_server_win32.c
++++ b/driver/others/blas_server_win32.c
+@@ -57,6 +57,7 @@ typedef struct{
+   HANDLE killed;
+ 
+   blas_queue_t	*queue;    /* Parameter Pointer */
++  blas_queue_t	*tail;     /* Last job waiting in "queue", for O(1) append */
+   int		shutdown;  /* server shutdown flag */
+ 
+ } blas_pool_t;
+@@ -239,7 +240,10 @@ static DWORD WINAPI blas_thread_server(void *arg){
+     EnterCriticalSection(&pool.lock);
+ 
+     queue = pool.queue;
+-    if (queue) pool.queue = queue->next;
++    if (queue) {
++      pool.queue = queue->pool_next;
++      if (pool.queue == NULL) pool.tail = NULL;
++    }
+ 
+     LeaveCriticalSection(&pool.lock);
+ 
+@@ -374,6 +378,7 @@ int blas_thread_init(void){
+ 
+     pool.shutdown = 0;
+     pool.queue    = NULL;
++    pool.tail     = NULL;
+ 
+     for(i = 0; i < blas_cpu_number - 1; i++){
+       blas_threads[i] = CreateThread(NULL, 0,
+@@ -405,7 +410,7 @@ int exec_blas_async(BLASLONG pos, blas_queue_t *queue){
+   if (unlikely(blas_server_avail == 0)) blas_thread_init();
+ #endif
+ 
+-  blas_queue_t *current;
++  blas_queue_t *current, *last = NULL;
+ 
+   current = queue;
+ 
+@@ -419,19 +424,22 @@ int exec_blas_async(BLASLONG pos, blas_queue_t *queue){
+     __asm__ __volatile__ ("stmxcsr %0" : "=m" (current -> sse_mode));
+ #endif
+ 
++    /* Build the pool's own chain over this caller's jobs */
++    current -> pool_next = current -> next;
++    last = current;
++
+     current = current -> next;
+     pos ++;
+   }
+ 
+   EnterCriticalSection(&pool.lock);
+ 
+-  if (pool.queue) {
+-    current = pool.queue;
+-    while (current -> next) current = current -> next;
+-    current -> next = queue;
++  if (pool.tail) {
++    pool.tail -> pool_next = queue;
+   } else {
+     pool.queue = queue;
+   }
++  pool.tail = last;
+ 
+   LeaveCriticalSection(&pool.lock);
+ 
+@@ -568,6 +576,7 @@ void goto_set_num_threads(int num_threads)
+ 
+ 			pool.shutdown = 0;
+ 			pool.queue    = NULL;
++			pool.tail     = NULL;
+ 			blas_server_avail = 1;
+ 		}
+ 
diff --git a/tools/build_steps_win_arm64.bat b/tools/build_steps_win_arm64.bat
index 9edc1260..e2332b00 100755
--- a/tools/build_steps_win_arm64.bat
+++ b/tools/build_steps_win_arm64.bat
@@ -71,8 +71,12 @@ if errorlevel 1 (
 )
  
 REM Patch
-for /r %%f in (..\patches\*) do git apply %%f
-if errorlevel 1 exit /b 1
+set patch_rc=0
+for %%f in (..\patches\*.patch ..\patches-windows\*.patch) do (
+    git apply %%f
+    if errorlevel 1 set patch_rc=1
+)
+if not "%patch_rc%"=="0" exit /b 1
 
 REM Patch VERSION
 REM version=$(grep "^version =" ../pyproject.toml | sed 's/version = "//;s/"//')

From fa30e5157e5c2e3f3a633b6c5b8d328c03bd37d5 Mon Sep 17 00:00:00 2001
From: mattip <matti.picus@gmail.com>
Date: Fri, 16 Oct 2026 16:25:01 +0300
Subject: [PATCH 4/6] fix patching lineending handling

---
 .gitattributes                                       | 3 +++
 patches-windows/openblas-win32-pool-queue-link.patch | 8 +-------
 tools/build_steps_win_arm64.bat                      | 2 +-
 tools/build_steps_windows.sh                         | 5 ++++-
 4 files changed, 9 insertions(+), 9 deletions(-)
 create mode 100644 .gitattributes

diff --git a/.gitattributes b/.gitattributes
new file mode 100644
index 00000000..4bce0d61
--- /dev/null
+++ b/.gitattributes
@@ -0,0 +1,3 @@
+# Patches must reach the build byte-identical: their line endings have to match
+# the files they patch, and checkout conversion would break that.
+*.patch -text
diff --git a/patches-windows/openblas-win32-pool-queue-link.patch b/patches-windows/openblas-win32-pool-queue-link.patch
index 80251ba1..98f78387 100644
--- a/patches-windows/openblas-win32-pool-queue-link.patch
+++ b/patches-windows/openblas-win32-pool-queue-link.patch
@@ -1,11 +1,6 @@
-Fix memory corruption from concurrent BLAS calls on Windows
-
-The Windows thread server keeps one global list of jobs waiting to be picked
-up, and it links that list through blas_queue_t::next -- the same field the
-*caller* owns and rebuilds on every call.
+Give the Windows thread pool its own queue link, see MacPython/openblas-libs#283
 
 diff --git a/common_thread.h b/common_thread.h
-index 7f5d7b2..cd7ca77 100644
 --- a/common_thread.h
 +++ b/common_thread.h
 @@ -119,6 +119,14 @@ typedef struct blas_queue {
@@ -24,7 +19,6 @@ index 7f5d7b2..cd7ca77 100644
     HANDLE finish;
    volatile int finished;
 diff --git a/driver/others/blas_server_win32.c b/driver/others/blas_server_win32.c
-index e827cee..4f24823 100644
 --- a/driver/others/blas_server_win32.c
 +++ b/driver/others/blas_server_win32.c
 @@ -57,6 +57,7 @@ typedef struct{
diff --git a/tools/build_steps_win_arm64.bat b/tools/build_steps_win_arm64.bat
index e2332b00..d585a71c 100755
--- a/tools/build_steps_win_arm64.bat
+++ b/tools/build_steps_win_arm64.bat
@@ -73,7 +73,7 @@ if errorlevel 1 (
 REM Patch
 set patch_rc=0
 for %%f in (..\patches\*.patch ..\patches-windows\*.patch) do (
-    git apply %%f
+    git apply --ignore-whitespace %%f
     if errorlevel 1 set patch_rc=1
 )
 if not "%patch_rc%"=="0" exit /b 1
diff --git a/tools/build_steps_windows.sh b/tools/build_steps_windows.sh
index 2003e85e..7fbb9ae8 100644
--- a/tools/build_steps_windows.sh
+++ b/tools/build_steps_windows.sh
@@ -90,9 +90,12 @@ interface_flags="$interface_flags SYMBOLPREFIX=scipy_ LIBNAMEPREFIX=scipy_ FIXED
 GCC_TAG="gcc_$(gcc -dumpversion | tr .- _)"
 OPENBLAS_VERSION=$(git describe --tags --abbrev=8)
 
+# --ignore-whitespace because the files patched here do not agree on line
+# endings (blas_server_win32.c is CRLF, common_thread.h is not) and neither
+# does checkout across platforms.
 for p in ../patches-windows/*.patch; do
     [ -e "$p" ] || continue
-    patch -p1 < "$p"
+    git apply --ignore-whitespace "$p"
 done
 
 # Patch VERSION

From 1ef66087d338e92d347392bf31c43f3f17433306 Mon Sep 17 00:00:00 2001
From: mattip <matti.picus@gmail.com>
Date: Fri, 16 Oct 2026 16:31:13 +0300
Subject: [PATCH 5/6] use pre-built conda for anaconda upload

---
 .github/workflows/windows-arm.yml | 16 +++++++---------
 1 file changed, 7 insertions(+), 9 deletions(-)

diff --git a/.github/workflows/windows-arm.yml b/.github/workflows/windows-arm.yml
index d2f2a512..16c7cf45 100644
--- a/.github/workflows/windows-arm.yml
+++ b/.github/workflows/windows-arm.yml
@@ -81,15 +81,12 @@ jobs:
         name: wheels-${{ env.PLAT }}-${{ env.INTERFACE64 }}
         path: dist/scipy_openblas*.whl
 
-    - name: Install Anaconda client
-      run: |
-        # Rust installation needed for rpds-py.
-        Invoke-WebRequest https://static.rust-lang.org/rustup/dist/aarch64-pc-windows-msvc/rustup-init.exe -UseBasicParsing -Outfile rustup-init.exe
-        .\rustup-init.exe -y
-        $env:PATH="$env:PATH;$env:USERPROFILE\.cargo\bin"
-        # cryptography dropped win-arm64 for 46.0.4, make sure we don't try to compile from source
-        pip install --only-binary :all: cryptography
-        pip install anaconda-client
+    - uses: conda-incubator/setup-miniconda@v4.0.1
+      with:
+        channels: conda-forge
+        channel-priority: true
+        activate-environment: upload
+        miniforge-version: latest
 
     - name: Upload
       # see https://github.com/marketplace/actions/setup-miniconda for why
@@ -98,5 +95,6 @@ jobs:
       env:
        ANACONDA_SCIENTIFIC_PYTHON_UPLOAD: ${{ secrets.ANACONDA_SCIENTIFIC_PYTHON_UPLOAD }}
       run: |
+        conda install -y anaconda-client
         source tools/upload_to_anaconda_staging.sh
         upload_wheels

From f21b340ee885cc6f220c8be392a234a4703f531d Mon Sep 17 00:00:00 2001
From: mattip <matti.picus@gmail.com>
Date: Sat, 17 Oct 2026 11:21:16 +0300
Subject: [PATCH 6/6] use a single anaconda upload per job matrix

---
 .github/workflows/posix.yml         | 13 ++++++++++++-
 .github/workflows/windows-arm.yml   | 25 ++++++++++++++++---------
 .github/workflows/windows.yml       | 25 ++++++++++++++++---------
 CHANGELOG.md                        |  5 +++++
 tools/upload_to_anaconda_staging.sh | 10 +++++++---
 5 files changed, 56 insertions(+), 22 deletions(-)

diff --git a/.github/workflows/posix.yml b/.github/workflows/posix.yml
index 2e77ec60..1e1b8b3b 100644
--- a/.github/workflows/posix.yml
+++ b/.github/workflows/posix.yml
@@ -119,8 +119,19 @@ jobs:
         name: wheels-${{ matrix.os }}-${{ matrix.PLAT }}-${{ matrix.INTERFACE64 }}-${{ matrix.MB_ML_LIBC }}-${{ matrix.MB_ML_VER }}
         path: dist/scipy_openblas*.whl
 
+  upload:
+    name: Upload to anaconda.org
+    needs: build
+    if: ${{ inputs.publish == false && (github.event_name == 'schedule' || github.event_name == 'workflow_dispatch') }}
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/download-artifact@37930b1c2abaa49bbe596cd826c3c89aef350131 # v7.0.0
+      with:
+        pattern: wheels-*
+        merge-multiple: true
+        path: dist
+
     - name: Upload to ananconda.org
-      if: ${{ inputs.publish == false && (github.event_name == 'schedule' || github.event_name == 'workflow_dispatch') }}
       uses: scientific-python/upload-nightly-action@0.6.4
       with:
         artifacts_path: dist
diff --git a/.github/workflows/windows-arm.yml b/.github/workflows/windows-arm.yml
index 16c7cf45..060ff817 100644
--- a/.github/workflows/windows-arm.yml
+++ b/.github/workflows/windows-arm.yml
@@ -81,20 +81,27 @@ jobs:
         name: wheels-${{ env.PLAT }}-${{ env.INTERFACE64 }}
         path: dist/scipy_openblas*.whl
 
-    - uses: conda-incubator/setup-miniconda@v4.0.1
+  upload:
+    name: Upload to anaconda.org
+    needs: build
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v6.0.1
+
+    - uses: actions/setup-python@v6.2.0
+      with:
+        python-version: 3.11
+
+    - uses: actions/download-artifact@37930b1c2abaa49bbe596cd826c3c89aef350131 # v7.0.0
       with:
-        channels: conda-forge
-        channel-priority: true
-        activate-environment: upload
-        miniforge-version: latest
+        pattern: wheels-*
+        merge-multiple: true
+        path: dist
 
     - name: Upload
-      # see https://github.com/marketplace/actions/setup-miniconda for why
-      # `-el {0}` is required.
-      shell: bash -el {0}
       env:
        ANACONDA_SCIENTIFIC_PYTHON_UPLOAD: ${{ secrets.ANACONDA_SCIENTIFIC_PYTHON_UPLOAD }}
       run: |
-        conda install -y anaconda-client
+        python -m pip install anaconda-client
         source tools/upload_to_anaconda_staging.sh
         upload_wheels
diff --git a/.github/workflows/windows.yml b/.github/workflows/windows.yml
index 9601ac82..303c422f 100644
--- a/.github/workflows/windows.yml
+++ b/.github/workflows/windows.yml
@@ -168,20 +168,27 @@ jobs:
         python -m pip install pkgconf
         python -m pkgconf scipy-openblas --cflags
 
-    - uses: conda-incubator/setup-miniconda@v4.0.1
+  upload:
+    name: Upload to anaconda.org
+    needs: build
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v6.0.1
+
+    - uses: actions/setup-python@v6.2.0
+      with:
+        python-version: 3.11
+
+    - uses: actions/download-artifact@37930b1c2abaa49bbe596cd826c3c89aef350131 # v7.0.0
       with:
-        channels: conda-forge
-        channel-priority: true
-        activate-environment: upload
-        miniforge-version: latest
+        pattern: wheels-*
+        merge-multiple: true
+        path: dist
 
     - name: Upload
-      # see https://github.com/marketplace/actions/setup-miniconda for why
-      # `-el {0}` is required.
-      shell: bash -el {0}
       env:
        ANACONDA_SCIENTIFIC_PYTHON_UPLOAD: ${{ secrets.ANACONDA_SCIENTIFIC_PYTHON_UPLOAD }}
       run: |
-        conda install -y anaconda-client
+        python -m pip install anaconda-client
         source tools/upload_to_anaconda_staging.sh
         upload_wheels
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 9570f875..da5c7414 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,10 @@
 ## OpenBLAS v0.3.34 (v0.3.34)
 
+### 0.3.34.237.2 (2026-10-16)
+
+- Patch windows to use a per-thread queue instead of mistakenly sharing the
+  thread queue. Also simplify windows-arm64 upload to anaconda
+
 ### 0.3.34.237.1 (2026-10-09)
 
 - Build every manylinux, musllinux and macOS wheel without a runtime
diff --git a/tools/upload_to_anaconda_staging.sh b/tools/upload_to_anaconda_staging.sh
index 31435a69..bf6da2ed 100644
--- a/tools/upload_to_anaconda_staging.sh
+++ b/tools/upload_to_anaconda_staging.sh
@@ -9,9 +9,13 @@ upload_wheels() {
         cd $our_wd
     fi
 
-    pushd OpenBLAS
-    VERSION=$(git describe --tags --abbrev=8)
-    popd
+    # The upload job only checks out openblas-libs, not the submodule;
+    # openblas_commit.txt holds the same string git describe would print.
+    if [ -d OpenBLAS/.git ]; then
+        VERSION=$(git -C OpenBLAS describe --tags --abbrev=8)
+    else
+        VERSION=$(cat openblas_commit.txt)
+    fi
 
     if [ "$ANACONDA_SCIENTIFIC_PYTHON_UPLOAD" == "" ]; then
         echo "ANACONDA_SCIENTIFIC_PYTHON_UPLOAD is not defined: skipping."