Skip to content

v2026.11.22.1

Latest

Choose a tag to compare

@github-actions github-actions released this 22 Nov 12:34
367a593

Added

  • i686-windows-msvc, the 32-bit x86 Microsoft ABI, is a preview row of the target table,
    served by the msvc provider as x86_64-windows-msvc is; x86_64 Windows runs its programs
    (WOW64). A project needs no [target.i686-windows-msvc] section to build it, and no platform is
    looked up for it. --target i686-windows and x86-windows-msvc select it; before, they
    selected no row. Under clang 23 its import std builds at C++20 only (docs/20) (mcpp#996).

Changed

  • The framework mcpp scaffolds tests with and suggests where a package has none is mtest 0.11.3
    (mtest gate judges a ledger only for a set that runs every case; the library is
    source-compatible with 0.11.2).

Fixed

  • A process that finds another refreshing the package index waits for it when an index the
    home reads has no packages yet, and judges that as a reader does (<index>/pkgs): the
    provider's own index (xim) is one of them, and a directory the other process has made and
    not yet filled is absent. Before, only the declared directories were judged, by existence,
    so a reader beside a first fetch skipped the refresh and reported no local package index
    (mcpp#927, design 2026-11-12 §18 P3).
  • A request built without a platform refreshed the package index on every build more than the
    debounce apart: a miss of the index's platform column refreshed for every row the engine does
    not know and every planned one, and a custom triple a [target.<triple>] section defines, or
    a planned row built with a named toolchain, is not refused for want of a platform (12 to 19 s
    a build for i686-windows-msvc with a section). Only a request refused without a platform
    refreshes now, judged by the facts the refusal reads; platform = "none" in a table ends the
    lookup as --platform none does (SPEC-013 §14, mcpp#996).
  • Two entries of a project's effective [indices] that name one location -- a member's beside
    its workspace's [workspace.indices], or two in one manifest -- registered the location twice,
    and installing any package of it stopped at the provider as ambiguous, naming neither entry.
    The build and mcpp index update refuse it now, naming both entries, the location and the
    remedy (index-location-named-twice, SPEC-004 §9).
  • An installed GCC's companions (binutils, the Linux headers) were read from the index's current
    descriptor of the toolchain, not from the descriptor the payload was installed with. When xim's
    gcc recipe moved binutils from 2.42.1 to 2.47, a gcc@16.1.0 installed beside binutils 2.42.1 was
    given no binutils after mcpp index update, and its builds stopped at "cannot execute 'as'". The
    installed descriptor is read first now, then the index's, and the first declaration the store
    holds decides: a home that links the payload from an [xlings] upstream holds the companions
    the index names, beside the upstream's descriptor (SPEC toolchain-management §2.9, design
    2026-11-21 §3.5).
  • mcpp test counts a suite run as a case by its cases on the program's line as the summary
    does: an mtest program of one suite and seven cases read 8 cases on its line and
    7 passed in the summary.