diff --git a/.github/scripts/package-arch-in-container.sh b/.github/scripts/package-arch-in-container.sh
new file mode 100755
index 00000000000..304b4474533
--- /dev/null
+++ b/.github/scripts/package-arch-in-container.sh
@@ -0,0 +1,35 @@
+#!/bin/sh
+
+set -eu #Needed so CI fails when anything is wrong
+set -x
+
+# Build and test the architecture-specific Debian packages inside a plain
+# Debian container started with "docker run", for architectures where the
+# GitHub "container:" job key cannot be used (the runner injects its own
+# 64-bit node into the container to run actions).
+# Same steps as the package-arch job, run from the repository root.
+
+echo "Kernel: $(uname -m), userland: $(dpkg --print-architecture), CPUs: $(nproc)"
+
+apt-get -q update
+apt-get --yes upgrade
+# Install stuff needed to turn the repo into a debian source package and run tests.
+apt-get --yes --no-install-recommends install curl git lsb-release python3 devscripts sudo adduser xauth
+
+git config --global --add safe.directory "${PWD}"
+
+DEBEMAIL=emc-developers@lists.sourceforge.net \
+DEBFULLNAME="LinuxCNC Github CI Robot" \
+    .github/scripts/build-package-arch.sh
+
+.github/scripts/verify-clean-repo.sh ':(exclude)VERSION' ':(exclude)debian/changelog'
+
+apt-get --yes --no-install-recommends install ../*.deb
+
+adduser --disabled-password --gecos "" testrunner
+passwd -d testrunner
+adduser testrunner sudo
+find tests/ -type d -exec chmod 0777 {} + # make test dirs world-writable for the testrunner
+su -c "./scripts/runtests -p -d ./tests" testrunner
+
+.github/scripts/verify-clean-repo.sh ':(exclude)VERSION' ':(exclude)debian/changelog'
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index f943a1152a9..597d17646bf 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -393,6 +393,58 @@ jobs:
         path: artifacts/${{ steps.meta.outputs.dist }}/${{ steps.meta.outputs.arch }}
         if-no-files-found: error
 
+  package-arch-32bit:
+    runs-on: ${{ matrix.runner }}
+    timeout-minutes: 120
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - arch: i386
+            platform: linux/386
+            runner: ubuntu-24.04
+          - arch: armhf
+            platform: linux/arm/v7
+            runner: ubuntu-24.04-arm
+    steps:
+    - name: Dump GitHub context
+      env:
+        GITHUB_CONTEXT: ${{ toJson(github) }}
+      run: |
+        echo Number of CPUs: $(nproc)
+        echo "$GITHUB_CONTEXT"
+    - name: Checkout repository
+      uses: actions/checkout@v6
+      with:
+        submodules: false
+        # "fetch-depth: 0" fetches all of history, this is needed by
+        # our build system to determine the version from tags
+        fetch-depth: 0
+    - name: Check for native 32-bit execution
+      id: native
+      run: |
+        set -x
+        if docker run --rm --platform ${{ matrix.platform }} debian:trixie true; then
+          echo "native=true" >> $GITHUB_OUTPUT
+        else
+          echo "native=false" >> $GITHUB_OUTPUT
+        fi
+    - name: Set up QEMU
+      if: steps.native.outputs.native != 'true'
+      uses: docker/setup-qemu-action@v3
+      with:
+        platforms: ${{ matrix.platform }}
+    - name: Build and test ${{ matrix.arch }} Debian packages
+      run: |
+        set -x
+        # IPC_OWNER is needed for shmget IPC_CREAT
+        # SYS_ADMIN is needed for shmctl IPC_SET
+        docker run --rm --platform ${{ matrix.platform }} \
+          --cap-add=IPC_OWNER --cap-add=SYS_ADMIN \
+          -e DEBIAN_FRONTEND \
+          -v "$PWD:/build/linuxcnc" -w /build/linuxcnc \
+          debian:trixie .github/scripts/package-arch-in-container.sh
+
   package-indep:
     runs-on: ubuntu-24.04
     timeout-minutes: 75
diff --git a/src/emc/rs274ngc/interp_convert.cc b/src/emc/rs274ngc/interp_convert.cc
index 6a4e77b8c71..76a4f7b90ab 100644
--- a/src/emc/rs274ngc/interp_convert.cc
+++ b/src/emc/rs274ngc/interp_convert.cc
@@ -6086,7 +6086,9 @@ int Interp::convert_threading_cycle(block_pointer block,
 
     depth = start_depth;
     zoff = (depth - full_dia_depth) * tan(compound_angle);
-    while (depth < end_depth) {
+    // TINY: with x87 math (i386) depth can be held in extended precision just
+    // below an end_depth it equals as a double, adding a duplicate full-depth pass
+    while (depth < end_depth - TINY) {
 	threading_pass(settings, block, boring, safe_x, depth, end_depth, start_y,
 		       start_z, zoff, taper_dist, entry_taper, exit_taper,
 		       taper_pitch, pitch, full_threadheight, target_z, angle_offset);
diff --git a/tests/inifile/ini_api_c/initest.c b/tests/inifile/ini_api_c/initest.c
index d71f5a3e213..f2996c30d8c 100644
--- a/tests/inifile/ini_api_c/initest.c
+++ b/tests/inifile/ini_api_c/initest.c
@@ -3,6 +3,7 @@
 #include <getopt.h>
 #include <ctype.h>
 #include <errno.h>
+#include <inttypes.h>
 
 #include <inifile.h>
 
@@ -94,7 +95,7 @@ int main(int argc, char *argv[])
 	case 'i':
 		rv = iniFindSInt(argv[optind], var, sec, &i);
 		if(!rv) {
-			printf("%ld\n", i);
+			printf("%" PRId64 "\n", i);
 		} else {
 			myerror("iniFindSInt", rv);
 			return -rv;
@@ -103,7 +104,7 @@ int main(int argc, char *argv[])
 	case 'u':
 		rv = iniFindUInt(argv[optind], var, sec, &u);
 		if(!rv) {
-			printf("%lu\n", u);
+			printf("%" PRIu64 "\n", u);
 		} else {
 			myerror("iniFindUInt", rv);
 			return -rv;