Run the engine's offline test suite on Windows

.github/workflows/windows-build.yml

@@ -81,6 +81,13 @@ jobs:
         platform: [x64, Win32]
         configuration: [Release]
     steps:
+      # The engine's test scripts are #!/bin/bash and the engine has no .gitattributes,
+      # so an autocrlf checkout would rewrite them and bash would die on $'\r'. Files
+      # land exactly as stored either way, which is what the Windows tooling wants too.
+      - name: Check out with the line endings the repos actually store
+        shell: pwsh
+        run: git config --global core.autocrlf false
+
       - name: Checkout httrack-windows
         uses: actions/checkout@v4
         with:
@@ -366,6 +373,97 @@ jobs:
           }
           if ($bad) { throw ($bad -join "`n") }
           Write-Host "::notice::all three binaries report $name $want"
+      # The engine ships ~90 tests and not one of them has ever run on Windows -- they
+      # only run under `make check` on Linux/macOS. These are the offline ones: no
+      # network, no server, just httrack.exe driven from PATH. They are the engine's
+      # tests, but ours is the binary that ships, and this code has never been executed
+      # on this platform.
+      - name: Run the engine test suite (offline tests)
+        shell: bash
+        working-directory: httrack/tests
+        run: |
+          set -u
+          bin="$(cygpath -u "$GITHUB_WORKSPACE")/httrack-windows/WinHTTrack/bin/${{ matrix.platform }}/${{ matrix.configuration }}"
+          export PATH="$bin:$PATH"
+
+          # httrack.exe is a native binary, so MSYS rewrites any argument shaped like a
+          # POSIX path -- and a URL path is shaped exactly like one. The mime self-test was
+          # handed "/a/b.html" and the engine received "C:/Program Files/Git/a/b.html".
+          # Switch the rewriting off, and hand the tests a temp dir that is already a Windows
+          # path, so nothing needs translating in either direction.
+          export MSYS_NO_PATHCONV=1
+          export MSYS2_ARG_CONV_EXCL='*'
+          export TMPDIR="$(cygpath -m "${RUNNER_TEMP}")"
+          echo "TMPDIR=$TMPDIR"
+          command -v httrack >/dev/null || { echo "::error::httrack.exe is not on PATH ($bin)"; exit 1; }
+          echo "httrack: $(command -v httrack)"
+
+          pass=0; fail=0; skip=0; failed=""
+          for t in 00_runnable.test 01_engine-*.test 01_zlib-*.test; do
+            rc=0
+            bash "$t" >"$t.log" 2>&1 || rc=$?
+            case "$rc" in
+              0)  pass=$((pass+1)); echo "PASS $t" ;;
+              77) skip=$((skip+1)); echo "SKIP $t -- $(tail -n 1 "$t.log")" ;;
+              *)  fail=$((fail+1)); failed="$failed $t"
+                  echo "FAIL $t (exit $rc)"
+                  # Several of these assert with `test "$(httrack ...)" == "..." || exit 1`,
+                  # which says nothing at all on failure. Re-run under a trace so the log
+                  # holds the call that failed and what it actually printed.
+                  bash -x "$t" >>"$t.log" 2>&1 || true
+                  sed 's/^/      /' "$t.log" | tail -n 25 ;;
+            esac
+          done
+          total=$((pass + fail + skip))
+          echo "--- ran=$total pass=$pass fail=$fail skip=$skip ---"
+
+          # A skipped test is not a passing test. Every gate in these scripts exits 77 --
+          # no python3, no HTTPS, no codec -- so a suite that quietly degrades to
+          # "everything skipped" would otherwise report success having tested nothing.
+          [ "$total" -ge 45 ] || { echo "::error::only $total tests were discovered: the glob found nothing"; exit 1; }
+          [ "$pass" -ge 40 ] || { echo "::error::only $pass tests actually ran and passed ($skip skipped)"; exit 1; }
+          [ "$fail" -eq 0 ]  || { echo "::error::failing tests:$failed"; exit 1; }
+
+      # The cache tests pipe stderr to /dev/null and print one word, so a failure says
+      # "FAIL" and nothing else. Run the two self-tests verbatim and keep everything they
+      # say: the engine side needs the assertion, not the verdict.
+      - name: Cache self-tests, verbatim
+        if: always()
+        shell: bash
+        working-directory: httrack/tests
+        run: |
+          bin="$(cygpath -u "$GITHUB_WORKSPACE")/httrack-windows/WinHTTrack/bin/${{ matrix.platform }}/${{ matrix.configuration }}"
+          export PATH="$bin:$PATH"
+          export MSYS_NO_PATHCONV=1
+          export MSYS2_ARG_CONV_EXCL='*'
+          export TMPDIR="$(cygpath -m "${RUNNER_TEMP}")"
+          command -v httrack >/dev/null || exit 0
+          for t in cache cache-corrupt; do
+            d="$TMPDIR/selftest-$t"
+            mkdir -p "$d"
+            echo "===== httrack -#test=$t $d"
+            httrack "-#test=$t" "$d" 2>&1 || echo "(exit $?)"
+            echo
+          done
+          # filterbounds prints nothing at all and the test just sees an empty string.
+          # An exit code tells us whether it died: 0xC00000FD is a stack overflow, and this
+          # self-test drives the matcher deep on purpose.
+          echo "===== httrack -#test=filterbounds"
+          httrack "-#test=filterbounds"; echo "exit=$? (139/0xC00000FD would be a stack overflow)"
+
+          echo "===== httrack -#test=sniff 'image/svg+xml' with a BOM"
+          printf '\xef\xbb\xbf  <?xml version="1.0"?>' > "$TMPDIR/bom.txt"
+          httrack "-#test=sniff" 'image/svg+xml' "text:$(cat "$TMPDIR/bom.txt")" 2>&1 || echo "(exit $?)"
+
+      # A 25-line tail cannot tell a real Windows bug from a POSIX assumption in the
+      # test. Keep every test's whole output.
+      - name: Upload the test logs
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: engine-tests-${{ matrix.platform }}-${{ matrix.configuration }}
+          path: httrack/tests/*.log
+          if-no-files-found: warn
 
       # Code signing, 1 of 2: the payload. SignPath signs an Inno installer as an opaque
       # PE and cannot reach the files inside it, so sign these BEFORE ISCC packages them.