Skip to content

binutils, ghc-9.6.7: amd64 onto gen-16's plain-loader objects (world rebuild; needs gen-16 published) - #1499

Closed
bryan-minimal wants to merge 2 commits into
mainfrom
gen16-repin
Closed

bryan-minimal wants to merge 2 commits into
mainfrom
gen16-repin

Conversation

@bryan-minimal

Copy link
Copy Markdown
Member

Repins the two amd64 bootstrap objects that gen-16 changed. Nothing else moves: gen-16 is add-only, and every other pin stays on its byte-identical gen-13 or gen-15 object (no pkgs-flip.py run).

package arch from to
binutils (replace_on_cycle) amd64 gen-13 binutils-2.46-glibc 7fe673db… gen-16 binutils-2.46-glibc-seed 741ece75…
ghc-9.6.7 amd64 gen-15 fd4510bb… gen-16 fc0d0c2b…

arm64 is unchanged. Its binutils was never on the bedrock loader, and gen-16's arm64 ghc-9.6.7 is byte-identical to the gen-15 object pinned now (1fbf1656…).

Why

The gen-13 amd64 binutils object named /usr/lib/glibc-bedrock-2.42/lib as the PT_INTERP and RUNPATH on all 22 of its ELFs. A build without the bedrock tree can't exec as. minimermetic's hermetic glibc pass hit exactly that, and it gates 304 of its queue tasks. bootstrap#12 published the same binutils as a new leaf, binutils-2.46-glibc-seed, on the plain /usr/lib/ld-linux-x86-64.so.2 loader with no RUNPATH, behind a build-time guard against any glibc-bedrock reference. Its outputs use the same globs (usr/bin, usr/lib, usr/include), so only the url/sha change. It needs the same thing gcc's gen-13 plain-loader cycle-breaker already relies on: the plain loader and libc in /usr/lib.

ghc-9.6.7 amd64: Hadrian built unlit-ghc-9.6.7 and lib/bin/unlit with ghc-9.2.8's ghc-cc, which appended the bedrock loader. gen-16 rewrites them to the plain loader.

Evidence (bootstrap side, gen-16 manifest = bootstrap#13)

  • Built on four lab VMs, two per arch. The two amd64 machines produced identical trees and tarballs.
  • ELF scan of the staged objects: the seed has 22 ELFs and ghc 55. None names glibc-bedrock, and their programs use /usr/lib/ld-linux-x86-64.so.2.

Before merge

  • Bryan publishes gen-16 (sh publish-gen16.sh, bootstrap branch gen-16). Until then these URLs 404, which is why this is a draft.
  • Rebuild scope: effectively the world on amd64. binutils sits in the gcc/glibc cycle, so its cycle-breaker's hash moves every amd64 spec. ghc-9.6.7 moves ghc and its dependents. Queue this on its own, not alongside another world rebuild.

Follows gominimal/bootstrap#12 and #13. Pairs with #1493 (rustc-1.99.1 from the same generation).

🤖 Generated with Claude Code

binutils' amd64 cycle-breaker moves from gen-13 binutils-2.46-glibc
(bedrock loader on all 22 ELFs) to gen-16 binutils-2.46-glibc-seed
(plain /usr/lib loader, no RUNPATH; bootstrap#12). ghc-9.6.7 amd64
moves to gen-16, where unlit is on the plain loader too. arm64 is
unchanged (gen-16's arm64 ghc is byte-identical to gen-15).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@coderabbitai

coderabbitai Bot commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

Warning

Review limit reached

  • Run on-demand review

This review includes 2 billable files and costs up to $0.50.

  • Ask an admin to make reviews automatic

Open in CodeRabbit

Reviews can continue after your included limit without a manual trigger. An admin must approve usage-based billing.

Or wait 43 minutes for your next included review.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available. Your 85 included PR review attempts over the past 7 days set your current allowance at 1 review per hour.

Learn how review limits work.

Review configuration:

Run configuration

Commits

Reviewing files that changed from the base of the PR and between ded4f65 and b26cf68.

Files selected for processing: 2

  • packages/binutils/build.ncl
  • packages/ghc-9.6.7/build.ncl
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

build-bot builds on push, not on ready_for_review.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

@norrietaylor norrietaylor left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Scope matches the title. The diff repins only the amd64 binutils cycle-breaker object and the amd64 ghc-9.6.7 object. The arm64 pins are not changed, and nothing else moves. Clobber none. git log on both files shows the previous change was the gen-15 move of ghc-9.6.7 in #1462. This PR keeps that move for arm64 and goes past it for amd64. Correctness In each URL, the sha256 path segment matches the sha256 field. The binutils replace_on_cycle outputs (usr/bin, usr/lib, usr/include, usr/share) do not change, which agrees with the PR body. The prebuilt binutils keeps runtime_deps = [zlib], the same shape as the gcc gen-13 plain-loader cycle-breaker in packages/gcc/build.ncl:125-160. The plain loader comes from the glibc cycle-breaker. Not checked I did not fetch the gen-16 objects, so I did not confirm the ELF claims or the hashes against the bucket. I did not run a build. The rebuild scope (all of amd64) is as the PR states. Queue it alone. One formatting nit is below.

# -> 7.6.3 -> 7.8.4 -> 7.10.3 -> 8.2.2 -> 8.6.5 -> 8.10.7 -> 9.2.8 -> 9.6.7, every rung built on the
# hex0-rooted toolchain and this one reproduced byte for byte on two independent machines before
# publishing (generations/gen-15/manifest.json). Installs the whole compiler at /usr/lib/ghc-9.6.7;
# publishing (amd64: generations/gen-16/manifest.json; arm64: gen-15, which gen-16 left byte-identical). Installs the whole compiler at /usr/lib/ghc-9.6.7;

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit · The edit makes this header comment line about 150 characters long. The lines around it wrap at about 100. Rewrap the paragraph so the block keeps one width. Line 8 has the same problem on a smaller scale.

@bryan-minimal

Copy link
Copy Markdown
Member Author

Plan: fold this into the gcc 16 rotation rather than merging it on its own. The rotation is pkgs#1340 now, and a PR queued alone later. Each of #1499 and #1535 is a world rebuild. The registry spec's ruling (gominimal/arch#183, decisions §4) is to make gcc, glibc and bootstrap changes in one rotation, so the catalog is rebuilt once. That rotation also takes the OQ-42 breaker output fixes (bash-bootstrap, bzip2, binutils arm64, attr).

#1535 specifically: once gcc-runtime lands, gcc_runtime_guard.py requires gcc-runtime's breaker to pin the same object as gcc's. The arm64 repin to gen-17 has to move both breakers together.

This waits for the current canary run (230721a) to finish.

🤖 Generated with Claude Code

bryan-minimal added a commit that referenced this pull request Nov 17, 2026
Both rebuild the world, so they ride the same rotation as gcc 16
instead of each costing one of their own.

- #1499: amd64 binutils breaker and ghc-9.6.7 onto gen-16's
  plain-loader objects (binutils-2.46-glibc-seed, ghc-9.6.7).
- #1535: arm64 gcc breaker onto gen-17's gcc-15.2.0-glibc-seed.
  gcc-runtime's breaker moves with it, since gcc_runtime_guard.py
  requires the two pins to match.

All three objects exist in gs://minimal-bootstrap. #1499's build was
green. #1535's failed only on tmd's zig fetch from
deps.files.ghostty.org (connection reset), which is unrelated.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
bryan-minimal added a commit that referenced this pull request Nov 17, 2026
…308f31)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@bryan-minimal

Copy link
Copy Markdown
Member Author

Closing: this is folded into #1545 (gcc 16.2.0 with gcc-runtime).

Repinning a cycle breaker rebuilds the world, and so does gcc 16, so both ride one rotation instead of paying for two. #1545's first commit (134712d4) carries this PR's change unchanged:

  • the amd64 binutils breaker moves to gen-16's binutils-2.46-glibc-seed (741ece75…);
  • ghc-9.6.7 moves to gen-16's object (fc0d0c2b…);
  • arm64 pins are unchanged.

That tree built and checked green on both arches as canary #1340 (9800cb3c). Thanks, @norrietaylor, for the review here; it applies to the same lines in #1545. If #1545 doesn't land, reopen this.

🤖 Generated with Claude Code

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