[saikatchatrg]

Fixes #4717.

This fixes the build-script path that 26Q3, 26Q4 and the Docker images use today. If kepler-formal moves to a Bazel build, as suggested in #4717, this block goes away with the rest of the script's kepler-formal section.

What. kepler-formal's install RPATH becomes $ORIGIN/../lib (@loader_path/../lib on macOS) instead of the absolute ${INSTALL_PATH}/kepler-formal/lib.

Why. The absolute RUNPATH ties bin/kepler-formal to the directory it was built in. After the checkout moves, or when that path later points at another ORFS checkout, the binary loads that build's libnaja_*.so, or none. In #4717 the LEC step stopped the flow in 4_1_cts with a symbol lookup error. The naja libraries already install with $ORIGIN-relative RUNPATHs (since kepler-formal f70c2e3, master's pin), so the binary is the last absolute path. For an install that never moves, the new RUNPATH resolves to the same directory as before. CMAKE_INSTALL_LIBDIR=lib already keeps the libraries in lib/ on EL, and the backslash keeps bash from expanding $ORIGIN. macOS has no $ORIGIN, so it gets @loader_path, as naja's own CMakeLists.txt does for Apple.

How verified.

• 26Q3, kepler-formal rebuilt with only this RPATH change (Ubuntu, x86-64): the binary's RUNPATH is $ORIGIN/../lib, every libnaja_*.so resolves from the install's lib/, and a flow with LEC_CHECK=1 passes 4_1_cts.
• 26Q4 install copied to another directory: unchanged, the binary loads all seven libnaja_*.so from the original checkout. With its RUNPATH set to $ORIGIN/../lib, all seven load from the copy's lib/ and the binary runs.
• Made on master ef324e55; bash -n build_openroad.sh passes.
• Not tested: a full build_openroad.sh --local on master, EL, or macOS.

[gemini-code-assist]

Code Review

This pull request updates the build_openroad.sh script to use relative RPATHs for the kepler-formal tool, allowing moved installations to correctly load their own libraries. It sets the RPATH to @loader_path/../lib on macOS and $ORIGIN/../lib on other platforms. There are no review comments, so no additional feedback is provided.

[openroad-ci]

🔍 QoR check

Metrics reflect the PR merge build — i.e. what will land on the target branch.

Commit 2bdabbe · Jenkins build #1 · Baseline: build · View build on dashboard

62 design(s) checked — 0 with regression(s), 0 without a comparable baseline.

[saikatchatrg]

The two Jenkins CI-Public jobs fail after the flows, not in them: every design flow passed, the QoR check reports 62 designs with no regression, and both jobs end in the post-report metadata-upload step ("process apparently never started", exit -2). That step fails the same way on other recent PRs (#4723, #4725, #4730–#4732), so a re-run would wait on it being fixed.

When someone has a moment, two things need a maintainer: approving the five GitHub Actions workflows (first-time contributor), and a review of build_openroad.sh from openroad-devops. @vvbandeira, since you extended this path in #4700, could you take a look?