[build] E: Build libxml2.so alongside libxml2.a

[esaurez]

Summary

Produce a position-independent libxml2.so in addition to the existing static libxml2.a. The shared library embeds zlib via --whole-archive so it is self-contained, and leaves libposix / libc / libm symbols unresolved so they bind to the host executable's .dynsym at dlopen time -- matching the extension-module model used elsewhere on Nanvix.

What changes

File	Change
.nanvix/Makefile.nanvix	Adds a $(SHAREDLIB) target (.libs/libxml2.so) that links manually from libxml2.a + libz.a via -shared -fPIC -nostdlib, sets DT_SONAME=libxml2.so so downstream consumers emit the right DT_NEEDED entry. test-functional now sanity-checks the .so: presence, minimum size, DT_SONAME, and that the public libxml2 entry points (xmlInitParser, xmlParseMemory) appear in .dynsym. package / verify-package ship both libxml2.a and libxml2.so.
.nanvix/z.py	Adds .libs/libxml2.so to _BUILD_OUTPUTS (so the Docker wrapper copies the .so back to the host) and to _staged_output_files() (so release packages it).

Why bare SONAME (libxml2.so, not libxml2.so.2)

Upstream libxml2 ships SONAME libxml2.so.2. The Nanvix port uses the bare libxml2.so because Nanvix is currently single-vendor: every consumer of this library is also built in this monorepo against this exact version, and Nanvix has no LD_LIBRARY_PATH-style runtime version search. If a third-party consumer ever needs the versioned name, the package target can install a libxml2.so.2 -> libxml2.so symlink as a follow-up; that change is non-load-bearing for current consumers and intentionally out of scope here.

Configure invariants kept

--enable-static --disable-shared stays in configure (libtool's shared-library detection has no rules for i686-nanvix); -fPIC is added to CFLAGS so the same .o files are usable for both archives.

Runtime dependency

The .so only becomes useful once the loader changes in nanvix/nanvix#2473 ([syscall] E: Run dlopen ctors/dtors and DT_RUNPATH) ship, because most C/C++ libraries -- including libxml2's downstream consumers libxslt and lxml -- rely on .init_array constructors that the current loader silently skips. The libxml2 unit-test in this PR does not exercise that path; downstream .so consumers will.

Validation

./z build produces .libs/libxml2.a (~5.7 MB) and .libs/libxml2.so (~1.5 MB). test-functional passes the structural checks (SONAME present, public entry points exported).

[Copilot]

Pull request overview

This PR updates the Nanvix port build to produce a shared libxml2.so alongside the existing static libxml2.a, and to include the .so in build outputs and release staging.

Changes:

• Add a libxml2.so build target that links from libxml2.a plus libz.a with PIC and a fixed SONAME.
• Extend functional tests to sanity-check the produced .so (SONAME and exported symbols).
• Update the Nanvix z.py build script to treat libxml2.so as a build/release artifact.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
.nanvix/z.py	Adds libxml2.so to Docker-copied build outputs and to the staged outputs list for release.
.nanvix/Makefile.nanvix	Adds a shared-library link step and .so checks in tests; adds package/verify targets; changes build/staging flow.

Comments suppressed due to low confidence (1)

.nanvix/z.py:79

• Libxml2Build.docker_config() will always try to copy back the install-staged outputs returned by _staged_output_files() (e.g., lib_out()/libxml2.a, bin_out()/xml2-config, etc.). However, the updated Makefile no longer has an install/staging step in the all target, so these paths will not be created during ./z build and the Docker output-copy step is likely to fail (or ./z release will miss artifacts). Either restore a staging step in .nanvix/Makefile.nanvix (preferred, since the comments here assume it), or update this script to copy/package artifacts from their new locations (e.g., the package target’s tarball).

    def _staged_output_files(self) -> list[str]:
        """Return install-staged artifact paths (relative to repo_root())
        so Windows tar-copy mode also copies them back to the host.
        """
        root = repo_root()
        return [
            str((lib_out() / "libxml2.a").relative_to(root)),
            str((lib_out() / "libxml2.so").relative_to(root)),
            str(
                (include_out() / "libxml2" / "libxml" / "xmlversion.h").relative_to(
                    root
                )
            ),
            str((bin_out() / "xml2-config").relative_to(root)),
            str((test_out() / "test_libxml2.elf").relative_to(root)),

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.