Repository navigation
Conversation
The-OpenROAD-Project#4501 removed ORFS's bazel surface, MODULE.bazel among it. A few files whose only purpose was to be referenced from that MODULE.bazel are still here, and nothing reads them any more. .bazelrc registry pins, C++ toolchain flags and a MODULE.bazel.lock policy for a bazel build that no longer exists. No workflow in .github/ invokes bazel. patches/{bison,icu,fakeram}/BUILD.bazel patches/tcl_lang/BUILD.bazel exports_files(glob(["*.patch"])) so a root MODULE.bazel could name the patches in single_version_override. With no MODULE.bazel nothing names them. Each file's own docstring said "Drop when tools/OpenROAD drops its copy"; tools/OpenROAD still has its copies, and those are the ones that matter. patches/tcl_lang/ goes entirely: it held no patch at all, so its glob matched nothing. flow/scripts/memories/bazel-orfs-auto-memories.patch a patch against bazel-orfs, carried here and described in AutoMemories.md as "to be upstreamed to bazel-orfs once the feature lands here". bazel-orfs implements AUTO_MEMORIES support itself now, and this copy no longer matches its tree. The patch files under patches/{bison,icu,fakeram} stay. Only the BUILD.bazel label plumbing goes; the patches themselves still describe real fixes. .bazelignore and .bazelversion also stay, and are not leftovers: .bazelversion is what bazelisk reads in docker/Dockerfile.claude, and .bazelignore keeps tools/ from being walked by anything that does consume this tree from bazel. AutoMemories.md loses a `bazelisk test //flow:memories_tests` line -- that target went with flow/BUILD in The-OpenROAD-Project#4501 -- and its "Consuming from bazel-orfs" section now states what a sandboxed build system has to declare instead of pointing at the deleted patch. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
There was a problem hiding this comment.
Code Review
This pull request removes several Bazel configuration, patch, and BUILD files, and updates the AutoMemories documentation to describe how bazel-orfs handles sandboxed builds. The review feedback identifies two inaccuracies in the updated documentation: a minor typo in the glob pattern for liberty files and an incorrect order in the described dependency chain.
Two review notes on the AutoMemories "Consuming from bazel-orfs" section. read_liberty.tcl globs $(RESULTS_DIR)/memories/*.lib, not *.lib. The text said the latter, which reads as if the liberty views were picked up from the results dir root while the LEFs came from memories/. Both come from memories/. The dependency chain is now written as the make rules themselves. The prose form invited reading it as execution order, which is the reverse: memories_inferred.json is *built* first, but `yosys-dependencies` *depends on* memories.json, which depends on memories_inferred.json. Showing the three rules removes the ambiguity rather than restating it, and says which end of the chain goes missing in a sandbox. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
🔍 QoR checkMetrics reflect the PR merge build — i.e. what will land on the target branch. Advisory — results are log-only and do not affect build status. The authoritative QoR gate remains the local rules-file check. Commit 62 design(s) checked — 0 with regression(s), 0 without a comparable baseline.
|
|
Isn't |
|
The community now has "ownership" of it. I don't know if anyone is using it. It may be that automemories and fake memories for that matter has no users currently. I don't know exactly who and how ORFS/OpenROAD is being used... |
What this removes
#4501 took out ORFS's bazel surface,
MODULE.bazelamong it. A handfulof files whose only purpose was to be referenced from that
MODULE.bazelare still in the tree, and nothing reads them now..bazelrcMODULE.bazel.lockpolicy for a bazel build that no longer exists. No workflow in.github/invokes bazel.patches/{bison,icu,fakeram}/BUILD.bazelexports_files(glob(["*.patch"])), so a rootMODULE.bazelcould name the patches in asingle_version_override. With noMODULE.bazel, nothing names them. Each file's own docstring says "Drop when tools/OpenROAD drops its copy" —tools/OpenROADstill has its copies, and those are the ones that are actually used.patches/tcl_lang/BUILD.bazelwhose glob therefore matched nothing.flow/scripts/memories/bazel-orfs-auto-memories.patchAutoMemories.mdas "to be upstreamed to bazel-orfs once the feature lands here". bazel-orfs implementsAUTO_MEMORIESsupport itself now, and this copy no longer matches its tree.What deliberately stays
patches/{bison,icu,fakeram}. Only theBUILD.bazellabel plumbing goes; the patches still describe realfixes. In particular
patches/fakeram/0001-Add-ORFS-ASAP7-backend.patchis left alone — see the note below.
.bazelversion—docker/Dockerfile.claudeinstalls bazelisk andrelies on it to pick the right release.
.bazelignore— still load-bearing for anything that consumesthis tree from bazel. It keeps
tools/from being walked, and as anexternal repository bazel turns it into
--deleted_packages; withoutit
tools/OpenROADandtools/yosysbecome loadable packages andtheir own bazel files get pulled into loading.
Docs
docs/user/AutoMemories.mdloses abazelisk test //flow:memories_testsline — that target went with
flow/BUILDin #4501 — and its "Consumingfrom bazel-orfs" section now states what a sandboxed build system has to
declare, instead of pointing at the patch this PR deletes.
Unrelated finding, for whoever owns AUTO_MEMORIES
Not fixed here, and not a consequence of this PR, but worth recording
while in the area: the ASAP7 FakeRAM backend is not applied to the
copy ORFS vendors.
gen_memories.pyinvokestools/FakeRAM2.0/run.py --orfs_asap7_backend.That flag comes from
patches/fakeram/0001-Add-ORFS-ASAP7-backend.patch,which the deleted
MODULE.bazelused to apply to FakeRAM when it was anexternal bazel module. FakeRAM has since been vendored at
tools/FakeRAM2.0/(the standalone repo is archived), and nothingapplies the patch to the vendored copy — there is no
tools/FakeRAM2.0/orfs_asap7/and no--orfs_asap7_backendinrun.py.The visible effect on a clean
mastercheckout:So three of ORFS's own AUTO_MEMORIES unit tests fail, and
AUTO_MEMORIES=1cannot generate views at all. The fix is presumably toapply that patch to
tools/FakeRAM2.0(or vendor it pre-patched) andthen drop the patch file — deliberately left out of this PR, which is
scoped to removing dead plumbing rather than changing behaviour.
🤖 Generated with Claude Code