mirror of
https://github.com/logos-co/logos-protocol.git
synced 2026-08-31 05:51:08 +00:00
Dario Gabriel LipicarandClaude Opus 5
b3675bc9af
Groundwork for making logos_protocol and logos_qt_host proper shared libraries
for the in-process image set, replacing the whole-archive + generated .def +
empty-archive-shim scheme. On its own this changes no behaviour: every current
consumer still resolves the macro to nothing.
WHAT WAS MISSING. The macro had only two states -- __declspec(dllimport) under
_WIN32 && LOGOS_SHARED_USE_DLL, and empty otherwise. There was no export half at
all, because the export side was delegated to a .def generated at
liblogos_core's link (logos-liblogos/cmake/gen-shared-exports.sh). A shared
library cannot use a .def belonging to a different library, so the types need to
be able to mark themselves.
Now three-state, mirroring LP_API's existing shape in logos_protocol.h:
EXPORT while building the shared library that owns the symbol, IMPORT while
consuming it, EMPTY for the static archive.
TWO MACROS, NOT ONE. LOGOS_QT_HOST_API is added alongside for logos-plugin-qt's
LogosAPI, which lives in a DIFFERENT library. It cannot reuse LOGOS_SHARED_API,
because the two are not the same choice in the same translation unit: while
building the Qt host runtime shared library, LogosAPI must be EXPORTED while
TokenManager -- owned by logos-protocol -- must be IMPORTED. One macro cannot
say both, and on PE getting it wrong defines the type twice in the process.
Nothing uses LOGOS_QT_HOST_API yet; it is defined here so the split is stated
where the mechanism lives rather than discovered later, and so the Qt host
change is a mechanical switch. Off Windows the distinction is moot -- both
resolve to default visibility -- which is exactly why it would go unnoticed
until a Windows build.
Also corrects this file's own premise, which is where the false claim
originated: "ELF and Mach-O give this for free. Both formats interpose symbols
across the whole process image set." True of ELF, FALSE of Mach-O, whose
two-level namespace gives no interposition -- measured in logos-basecamp, where
one reference to LogosAPI::forIdentity dragged logos_api.cpp.o into the
executable and produced 31 refused calls against a baseline of 0.
VERIFIED, aarch64-darwin.
static archive symbol tables IDENTICAL (14257 lines each), same size, and
EXACTLY ONE byte differs in the whole 5.4MB file -- 351 -> 352
in __.SYMDEF's ar header, i.e. build-environment metadata, not
content. This is the no-behaviour-change claim, measured.
shared library 748 exported / 78 runtime symbols, unchanged. Expected: no
-fvisibility=hidden anywhere in this build, so a Mach-O shared
library already exported them by default.
checks.tests PASS
NOT VERIFIED LOCALLY: Windows, which is the only platform where this changes
anything -- without dllexport the symbols are simply absent from the PE export
table. packages.x86_64-windows requires an x86_64-linux builder, so it rests on
CI here.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>