Skip to content

support DLLs on Windows#75

Draft
jll63 wants to merge 90 commits into
boostorg:developfrom
jll63:feature/windll3
Draft

support DLLs on Windows#75
jll63 wants to merge 90 commits into
boostorg:developfrom
jll63:feature/windll3

Conversation

@jll63

@jll63 jll63 commented Jun 13, 2026

Copy link
Copy Markdown
Collaborator

No description provided.

jll63 and others added 27 commits May 24, 2026 09:50
Replace declspec(import/export) approach with method consolidation:
- Remove detail::static_fn infrastructure
- Revert fn to simple static method member (not via MAKE_STATICS)
- In augment_methods(), group methods by type_id and collect
  overriders from all module copies into canonical method
- Build dispatch tables once per unique method, not once per copy

This avoids intrusive list manipulation during initialization while
enabling correct dispatch table construction with shared method state
across DLL boundaries.

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
- Resolve deferred type IDs before consolidation (deferred_static_rtti)
- Update test_custom_rtti to assign unique IDs to non-polymorphic types
  via non_polymorphic_static_type<T>() function-local counter, so method
  tag classes don't collapse to a single sentinel and trigger spurious
  ambiguous dispatch
- Remove get_fn test from dynamic_loading: each module now has its own
  fn instance, so address equality no longer holds (the state is
  synchronized via consolidation at initialize time instead)

All 77 tests pass.

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
- Add `meet` multi-method to test_dynamic_loading so the test exercises
  multiple methods and non-zero slots/strides — previously the test
  accidentally passed because the lone `speak` method got slot 0
  (matching zero-initialized arrays in non-local module fns)
- Rename `canonical_methods` → `local_methods` in augment_methods: the
  selected method_info isn't canonical in any global sense, it's just
  the one that happens to live in the module performing initialize()
- After dispatch table build, copy slots_strides values from each
  local method_info to all other module copies. Each module's `fn`
  has its own slots_strides[] array and dispatch reads it directly,
  so every copy must hold the same values.

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Iterating methods directly and skipping the self pointer is clearer than
building a side map keyed by type_index.

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Revert from signed int back to std::size_t. To distinguish polymorphic
from non-polymorphic types, set the high bit (1 << (sizeof(size_t)*8-1))
in the non_polymorphic_static_type counter values. Polymorphic Animal/
Dog/Cat keep small positive values 1/2/3; method tag classes get values
like 0x800...001, 0x800...002 — comfortably outside the 1..3 range used
by type_name. Avoids the size mismatch warnings (C4312) that b2's W4/WX
flags treat as errors on the int->const-void* reinterpret_cast path.

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Policy state moved into registry_state, so stderr_output::os is now a deprecated alias and stream() returns the registry-backed stream. The library's own diagnostics still wrote to os, which broke warnings-as-errors builds (e.g. b2 with /WX). Route them through stream() and give the test's capture_output policy a stream() accessor.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
boost::dll resolves the relative library name against the working directory. Without WORKING_DIRECTORY, ctest ran the test from the binary tree where a stale boost_openmethod-shared.dll could be picked up, failing at load with ERROR_PROC_NOT_FOUND. Point it at $<TARGET_FILE_DIR:boost_openmethod-dynamic>, where the freshly-built DLL sits co-located with the exe.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Move the policy-independent dispatch-table-construction algorithm (~1300 lines)
from the doubly-nested template registry<Policies...>::compiler<Options...> into
the non-template detail::generic_compiler, so it compiles once per TU instead of
once per distinct registry type. This fixes the 32-bit MSVC (cl.exe, ~2GB VA space)
C1060 "compiler out of heap space" error when test_compiler.cpp etc. defined
multiple registries.

Key changes:
- Add abstract trace_sink with per-registry trace_sink_impl<Registry> virtual
  overrides (5 small virtuals per registry instead of hundreds of operator<<
  instantiations).
- Move all algorithm member functions to detail::generic_compiler (now non-
  template): build_class_lattice, assign_slots, build_dispatch_tables, etc.
  Replace if constexpr(has_trace) with if(tr.on()) and has_option<n2216> with
  use_n2216.
- Reduce registry<...>::compiler<> to a thin shim: resolve_classes,
  resolve_methods, ctor binding tr.sink to the concrete sink impl, and
  install_global_tables running policies then swapping dispatch_data into
  registry::state.
- All changes confined to initialize.hpp. Remove test/Jamfile's
  <toolset>msvc-14.5,<address-model>32:<build>no skip now that 32-bit works.
- Add trace_demo.cpp example: demonstrates dispatch/next/tracing with
  BOOST_OPENMETHOD_TRACE=1.

Verified: MSVC 14.5 x86_64 full suite pass; MSVC 14.5 x86_32 full suite pass
(C1060 gone); gcc 13.4 -Werror clean; trace output identical (byte-for-byte
after address normalization).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Explicitly convert function pointer to std::uintptr_t in write_pf() instead of
relying on implicit conversion through operator<<. This avoids the -Wmicrosoft-cast
warning while using the same reinterpret_cast approach that ostdstream.hpp already
uses for function pointer output.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
@cppalliance-bot

cppalliance-bot commented Jun 13, 2026

Copy link
Copy Markdown

An automated preview of the documentation is available at https://75.openmethod.prtest3.cppalliance.org/libs/openmethod/doc/html/index.html

If more commits are pushed to the pull request, the docs will rebuild at the same URL.

2026-06-13 19:06:01 UTC

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants