The directory restructuring looks good — centralizing thirdparty build logic into thirdparty/*/CMakeLists.txt is cleaner than having it all in the root CMakeLists.txt.

A few things this PR will need before it can close #366:

dxflib symbol embedding (#366): add_library(dxflib_bundled STATIC) still produces a separate archive — its symbols won't end up in libgerbv.a, which is the original bug. Downstream consumers like pcb2gcode get undefined references. Two options:

• Compile bundled sources directly into the gerbv target via target_sources() (what fix: embed bundled dxflib directly into libgerbv #369 did)
• Use an OBJECT library and consume via target_link_libraries(gerbv PRIVATE dxflib_bundled) — CMake 3.12+ propagates object files from OBJECT libraries through normal linking

Warning suppression on bundled dxflib (#372): The bundled sources trigger a GCC false positive (-Wuninitialized on DL_HatchEdgeData default construction). With CMAKE_COMPILE_WARNING_AS_ERROR ON in the CI preset, this becomes a build error. Needs set_source_files_properties or SYSTEM includes to suppress.

tinyscheme _GNU_SOURCE (#370): Building with GCC 15 / -std=c23 on Linux fails because isascii() isn't exposed without _GNU_SOURCE. The tinyscheme target needs:

if(CMAKE_SYSTEM_NAME STREQUAL "Linux")
    target_compile_definitions(tinyscheme PRIVATE _GNU_SOURCE)
endif()

MSVC _USE_MATH_DEFINES: M_PI and M_1_PI (used by DEG2RAD/RAD2DEG in gerbv.h) aren't defined under MSVC strict ANSI. This needs a compile definition on compilation-flags:

if(MSVC)
    target_compile_definitions(compilation-flags INTERFACE _USE_MATH_DEFINES)
endif()

I'm closing #369 and #371 in favor of this restructuring — the fixes above cover what those PRs addressed. Happy to help if needed.

_GNU_SOURCE: If you configure with --preset linux-gnu-gcc, _GNU_SOURCE is included in all compilations. Linux specific settings and Apple specific settings should preferably go into the respective toolchains files. Also valid for MSVC above.

I quickly looked up the isascii() and there seems to be more generic ways to fix it.

I have added the OBJECT flag now as well.

Makes sense — replacing isascii() with a portable equivalent like ((unsigned)(c) <= 0x7f) in scheme.c would be cleaner than relying on _GNU_SOURCE. That patch would go into GERBV_PATCHES alongside the existing vendored dxflib patches.

One small thing: thirdparty/CMakeLists.txt is missing a trailing newline at EOF — some tools (GCC, Git, various linters) will warn about that.

Everything else looks good after the squash. The OBJECT library for dxflib is the right call. Will watch for the HAVE_LIBDXFLIB guard cleanup whenever that lands.

Thanks for pushing this through — the thirdparty restructuring is a big improvement.