Revert "feat(xim): persist declared xpkg exports (#351)" - #352
Merged
Conversation
…kg-exports.json (#351)" The persisted file's only consumer was mcpp's toolchain link model, and the entire mcpp 0.0.83 verification matrix (PR CI, hermetic container, fresh installs across 3 platforms + 6 distro containers) ran green on xlings 0.4.62 — which never had this file — with mcpp's built-in fallback (per-arch triple map + glob) covering every real-world case. A general-purpose package manager should not carry a mechanism whose sole reader is one downstream tool: reverted per layering review. The declared exports in package lua (exports.runtime) are untouched — xlings' own elfpatch predicate still consumes those. If an installed-state metadata DB is ever needed, it should be designed with xlings itself as its first consumer.
Sunrisepeak
added a commit
to mcpp-community/mcpp
that referenced
this pull request
Jul 7, 2026
…rts.json reader (#197) New contributor-facing document describing the toolchain machinery end to end — resolution, the link model, the unified post-install fixup pipeline (trigger semantics, marker fingerprinting, the never-patch-in-place and ownership invariants), the human-facing cfg, the hermetic link check — and how to extend it: new toolchains, new CPU architectures, and the embedded/bare-metal outlook (no-loader / self-contained-sysroot toolchains map onto existing CLibMode semantics). Also removes mcpp's reader for the persisted .xpkg-exports.json (the xlings-side writer is reverted in openxlings/xlings#352): its only consumer was this resolver, while the triple map + glob fallbacks already cover every real payload — the entire 0.0.83 verification matrix ran green on xlings 0.4.62, which never had the file. The decision is recorded in the new doc (§3). Loader resolution behavior is unchanged for every existing payload. No release needed: doc + dead-code removal only.
Sunrisepeak
added a commit
that referenced
this pull request
Aug 5, 2026
…stack
Covers what slice 1 deferred: the payload. Two tracks whose constraints are
opposites — mesa can be built and published; NVIDIA's userspace cannot be
redistributed and must match the host kernel module bit for bit, so it is
borrowed rather than shipped.
Measured on this machine rather than assumed, and three of the measurements
contradict what the slice 1 design assumed:
* the payload is 241 MB, not the >800 MB O4 feared — closed, no split.
LLVM alone is 137 MB of it, and the xlings llvm package cannot supply it
(627 MB of clang toolchain, no libLLVM.so);
* mesa 25.x is one 41 MB libgallium megadriver plus symlinks, not the
"25 hand-written .so" the design worried about;
* libgallium tops out at GLIBC_2.38, so building against the subos's
glibc 2.39 is exactly the inverse of issue #352.
The architectural finding is libglvnd: it is the vendor-neutral dispatch
layer, so mesa and the host's NVIDIA stack coexist in one subos and the
choice is per-process through the environment — which is what slice 1
shipped. That makes these two tracks parallel rather than sequential.
Also names the failure mode Track B has to design for: a host driver upgrade
leaves dangling symlinks that pass every [ -e ] test, and surfaces as a
loader error that never mentions xlings.
Sunrisepeak
added a commit
that referenced
this pull request
Aug 5, 2026
…is gcc-runtime
The names I proposed were wrong on three counts, and the root cause is a layer
boundary the doc never stated.
`compat.` is **mcpp-index**'s namespace — compat.freetype, compat.glfw,
compat.glx-runtime — for C/C++ libraries an mcpp *build* consumes. The graphics
payload is what runs inside a subos, which is xim-pkgindex: default `xim`
namespace, kebab-case, and not one package name there contains a dot. So
compat.mesa becomes `mesa`, host.nvidia becomes `nvidia-runtime`.
And I had argued against calling it `llvm-runtime` on the grounds it would be
confused with the `llvm` package. That is backwards: xim-pkgindex already
ships `gcc-runtime`, split from `gcc` for word-for-word the same reason —
"xim:gcc is the full compiler (~1.1 GB) / the runtime libs are ~25 MB / tools
that only need to run C++ binaries don't need cc1/cc1plus/headers/*.a". With
that precedent in the same index, llvm-runtime is the least confusable name
available, and the split is a shape the index already has rather than a new
idea borrowed from Ubuntu.
One difference that must not be copied along with the shape: gcc-runtime's own
comment notes libstdc++ is forward-compatible, so one version covers every
binary built against an equal-or-older GCC. libLLVM is not — the @LLVM_20.1
symbol version pins it to major.minor. Same packaging shape, different version
policy, and consumers must hard-pin.
Also corrects the C++ runtime dependency: mesa should depend on gcc-runtime
(~25 MB, ships libstdc++.so.6.0.34 = GLIBCXX_3.4.34, covering the 3.4.29
libgallium needs) rather than on gcc, which is the 1.1 GB compiler it has no
use for at run time.
Records the mcpp-index relationship rather than leaving it implicit:
compat.glx-runtime is a payload-less host adapter declaring
capabilities = {x11.display, opengl.glx.driver}. It answers "how does an mcpp
app say it needs GL"; this design answers "does the subos have GL at all".
They meet later — an mcpp app sourcing GL from the subos instead of the host is
what actually cures #352 — but they are different packages in different
indexes.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Reverts #351 per layering review: the persisted
.xpkg-exports.jsonhad a single external consumer (mcpp's toolchain link model), while mcpp's built-in fallback (per-arch triple map + glob) already covers every real-world case — the entire mcpp 0.0.83 verification matrix ran green on xlings 0.4.62, which never had this file. A general-purpose package manager shouldn't carry a mechanism whose only reader is one downstream tool. The lua-levelexports.runtimedeclarations are untouched (xlings' own elfpatch predicate consumes those). mcpp removes its reader side in a companion PR.