Splice the softfp shims before the sysroot leaves for stage 2

[frangarcj]

softfp-shim was a dependency of finalize-sdk only. A producer does not run finalize-sdk — it builds the sysroot target and hands the target half to stage 2 — so the stub archives it exported were never patched. The first softfp build in CI failed all eight stage 2 and stage 3 jobs on:

checking the softfp ABI shims (23 functions) resolve against the shim, not the raw stub
missing sceGxmSetViewport resolved against the softfp shim wrapper

The container builds this was verified in are single-stage, where finalize-sdk runs.

The fix names the target half once, in CMakeLists.txt, with the splice in it:

set(target_half_dependencies
    ${gcc_final_barrier} vita-headers newlib pthread-embedded samples
    ${version_info_file})
if(VITASDK_FLOAT_ABI STREQUAL "softfp" AND NOT VITASDK_STAGE1_DIR)
    list(APPEND target_half_dependencies softfp-shim)
endif()

ExportSysroot.cmake and FinalizeSdk.cmake both wait for that list instead of spelling the target half out separately. Stage 2 is excluded because it imports the archives already spliced and builds no components, so there is no softfp-shim target there.

The test checks both recipes take the list and that neither names softfp-shim itself, which is what let them drift.

🤖 Generated with Claude Code