diff --git a/tests/fff_print/CMakeLists.txt b/tests/fff_print/CMakeLists.txt
index abddfebe5e..ac1b12d61f 100644
--- a/tests/fff_print/CMakeLists.txt
+++ b/tests/fff_print/CMakeLists.txt
@@ -8,6 +8,7 @@ add_executable(${_TEST_NAME}_tests
 	test_fill.cpp
 	test_flow.cpp
 	test_gcode_timing.cpp
+	test_gcodeprocessor.cpp
 	test_gcodewriter.cpp
 	test_model.cpp
 	test_multifilament.cpp
diff --git a/tests/fff_print/test_gcodeprocessor.cpp b/tests/fff_print/test_gcodeprocessor.cpp
new file mode 100644
index 0000000000..b8ce32e05c
--- /dev/null
+++ b/tests/fff_print/test_gcodeprocessor.cpp
@@ -0,0 +1,105 @@
+#include <catch2/catch_all.hpp>
+
+#include "libslic3r/GCode/GCodeProcessor.hpp"
+
+#include "test_utils.hpp"
+
+#include <boost/filesystem.hpp>
+#include <boost/nowide/fstream.hpp>
+
+#include <iterator>
+#include <string>
+#include <vector>
+
+using namespace Slic3r;
+
+namespace {
+
+// Writes the bytes verbatim: the G-code export is binary, so no newline translation must creep in
+// here either, otherwise the CRLF cases below would silently test LF.
+void write_gcode(const std::string &path, const std::string &content)
+{
+    boost::nowide::ofstream f(path, std::ios::binary);
+    f << content;
+}
+
+// Cuts line `id` (1-based) out of the file the way GCodeViewer's G-code window does: the text it
+// shows is bytes [lines_ends[id - 2], lines_ends[id - 1]). Reproducing that here is what makes these
+// assertions about the rendered result rather than about the offsets themselves.
+std::string line_at(const std::string &path, const std::vector<size_t> &lines_ends, size_t id)
+{
+    boost::nowide::ifstream f(path, std::ios::binary);
+    const std::string       bytes((std::istreambuf_iterator<char>(f)), std::istreambuf_iterator<char>());
+    const size_t            start = id == 1 ? 0 : lines_ends[id - 2];
+    return bytes.substr(start, lines_ends[id - 1] - start);
+}
+
+} // namespace
+
+TEST_CASE("rebuild_lines_ends realigns the line offsets after the G-code is rewritten in place", "[GCodeProcessor]")
+{
+    ScopedTemporaryFile  tmp(".gcode");
+    GCodeProcessorResult result;
+    result.filename = tmp.string();
+
+    write_gcode(tmp.string(), "G1 X1\nG1 X2\nG1 X3\n");
+    REQUIRE(result.rebuild_lines_ends());
+    REQUIRE(result.lines_ends.size() == 3);
+    CHECK(line_at(tmp.string(), result.lines_ends, 2) == "G1 X2\n");
+
+    // What a post-processing script that opens the file in Python text mode on Windows does: prepend a
+    // comment and translate every LF to CRLF. Every byte offset shifts, and cumulatively.
+    write_gcode(tmp.string(), ";EDITED\r\nG1 X1\r\nG1 X2\r\nG1 X3\r\n");
+    CHECK(line_at(tmp.string(), result.lines_ends, 2) != "G1 X1\r\n");
+
+    REQUIRE(result.rebuild_lines_ends());
+    REQUIRE(result.lines_ends.size() == 4);
+    CHECK(line_at(tmp.string(), result.lines_ends, 1) == ";EDITED\r\n");
+    CHECK(line_at(tmp.string(), result.lines_ends, 2) == "G1 X1\r\n");
+    CHECK(line_at(tmp.string(), result.lines_ends, 4) == "G1 X3\r\n");
+}
+
+TEST_CASE("rebuild_lines_ends maps a file longer than one read chunk", "[GCodeProcessor]")
+{
+    ScopedTemporaryFile  tmp(".gcode");
+    GCodeProcessorResult result;
+    result.filename = tmp.string();
+
+    const size_t line_count = 20000;
+    std::string  content;
+    for (size_t i = 0; i < line_count; ++i)
+        content += "G1 X" + std::to_string(i) + "\n";
+    // The scan reads the file in chunks, so it has to span at least one chunk boundary to be meaningful.
+    REQUIRE(content.size() > 65536);
+    write_gcode(tmp.string(), content);
+
+    REQUIRE(result.rebuild_lines_ends());
+    REQUIRE(result.lines_ends.size() == line_count);
+    CHECK(result.lines_ends.back() == content.size());
+    CHECK(line_at(tmp.string(), result.lines_ends, line_count) == "G1 X" + std::to_string(line_count - 1) + "\n");
+}
+
+TEST_CASE("rebuild_lines_ends ignores a trailing line that has no newline", "[GCodeProcessor]")
+{
+    ScopedTemporaryFile  tmp(".gcode");
+    GCodeProcessorResult result;
+    result.filename = tmp.string();
+
+    // Matches the export pass, which only ever records the offset one past a '\n'.
+    write_gcode(tmp.string(), "G1 X1\nG1 X2");
+    REQUIRE(result.rebuild_lines_ends());
+    CHECK(result.lines_ends.size() == 1);
+    CHECK(line_at(tmp.string(), result.lines_ends, 1) == "G1 X1\n");
+}
+
+TEST_CASE("rebuild_lines_ends empties the map when the G-code file cannot be read", "[GCodeProcessor]")
+{
+    // A partially rebuilt map would have the G-code window slicing lines at wrong offsets again; an
+    // empty one just hides the window.
+    GCodeProcessorResult result;
+    result.filename   = (boost::filesystem::temp_directory_path() / "orca-nonexistent-gcode-file.gcode").string();
+    result.lines_ends = {1, 2, 3};
+
+    CHECK_FALSE(result.rebuild_lines_ends());
+    CHECK(result.lines_ends.empty());
+}