Sanitize XML 1.0 illegal control bytes in printer output

[johnathan-arsenault]

Summary

• Replace XML-1.0-illegal C0 control bytes (e.g. \x16 SYN) with spaces in element values, attributes, and raw fragments before handing off to libxml2. This was the root cause of lxml/etree refusing to parse agent output (PCDATA invalid Char value 22). libxml2 escapes &<>"' but does not validate against the XML Char production, so callers must.
• Switch the Ubuntu build matrix to shared=False and exclude boost::stacktrace, producing a fully static agent binary that runs on stock Ubuntu 22.04 without needing the gcc-13 libstdc++ ABI (GLIBCXX_3.4.32).
• Add -o development=True to both Ubuntu workflows so CI now runs the test suite — it wasn't before.

What changed

XML escape fix (the actual bug):

• src/mtconnect/printer/xml_printer_helper.hpp — sanitizeXmlChars (returns std::optional<std::string>; clean inputs avoid allocation) and writeXmlString / writeXmlAttribute / writeXmlRaw wrappers. Replacement-with-space (not deletion) preserves token boundaries: "1.25 \x16 3" → "1.25 3" rather than "1.253". UTF-8 continuation bytes (≥0x80) pass through untouched.
• src/mtconnect/entity/xml_printer.cpp, src/mtconnect/printer/xml_printer.cpp — every direct xmlTextWriterWrite{String,Attribute,Raw} callsite routed through the wrappers.
• printer/xml_printer.cpp cleanup: removed file-local duplicates of XmlWriter, AutoElement, openElement, closeElement, and the THROW_IF_XML2_{ERROR,NULL} macros — they were byte-identical to the canonical versions in xml_printer_helper.hpp.

Build / packaging (needed for the fix to actually ship to the customer's device):

• conanfile.py — boost*:without_stacktrace=True. Under static linking, libboost_stacktrace_from_exception.a collides with libstdc++.a over __cxa_allocate_exception. The agent does not use boost::stacktrace.
• .github/workflows/build-deb.yml, .github/workflows/build-ubuntu-22.04.yml — matrix shared: ["False"] and -o development=True.

Local dev tooling (new):

• docker/Dockerfile.ubuntu-22.04 + docker/build.sh — mirror the GHA workflow inside a container so devs can reproduce CI failures locally. docker/build.sh deb [version] produces a .deb matching jiro4989/build-deb-action@v3 output. ~/.cache/cppagent-conan persists the conan cache between runs.
• .gitignore updated to exclude the resulting build artifacts (*.deb, _CPack_Packages/, pkgroot/{home,DEBIAN/control}).

Verification

• 753/753 tests pass, including new XmlPrinterTest.IllegalControlCharactersAreSanitized regression that uses the exact <Block>-with-\x16 payload from the lxml traceback.
• Resulting agent binary in the static build:
  • ldd agent → only libm.so.6, libc.so.6, ld-linux. No libstdc++.so.6, no libboost_*.
  • strings | grep GLIBCXX → only GLIBCXX_TUNABLES (an unversioned token, not a symbol requirement).
  • max glibc symbol used: GLIBC_2.35 (Ubuntu 22.04 ships 2.35).
• Built mtconnect-agent_2.6.0.2.2-dnx_all.deb locally, installed on the failing customer device (rolarproducts-edge1, Ubuntu 22.04.5 LTS) — the agent now starts cleanly.

Test plan

• docker/build.sh runs the full ctest suite locally (753/753 pass).
• Agent service starts successfully on the affected customer device after dpkg -i.
• CI (this PR) passes both build-deb.yml and build-ubuntu-22.04.yml.

Notes / follow-ups (not in this PR)

• conan/profiles/gcc still uses [system_tools], deprecated in Conan 2.x → [platform_tool_requires]. docker/build.sh rewrites the section name on the fly with a TODO pointer so the upstream rename is findable.
• make-agent-deb.sh is stale (still expects the pkgroot/DEBIAN/control that was removed in da3e934e); the deb path is now the GHA action / docker/build.sh deb. Cleanup is out of scope here.

🤖 Generated with Claude Code