build-cran-binaries/.crow/backfill-abi-classification.yaml
pat-s 1d450c7e5d
feat(abi): backfill classifier verdicts for pre-per-minor packages
bincraft now unions the ABI classifier cache into its risky-package set, but the
cache only holds verdicts for packages that passed through the build pipeline
after the per-minor scheme existed. Everything older has no verdict, and no
verdict means the flat binary is carried into every per-minor index. That is the
population the fix has to reach: a base64enc built under R 4.4.3 was served to
R 4.6 clients, which have no SETLENGTH to bind.

The verdict is computed from the package source, so it is independent of OS,
arch and R minor: one run populates the cache for every slot.

Only NeedsCompilation=yes packages are classified. A package with no compiled
code has no symbols to bind and cannot fail this way, which cuts the work from
23218 packages to 4585 on amd64/resolute.

This change will:

- add local/backfill-abi-classification.R, enumerating a slot's flat index and
  classifying each compiled package through bincraft's cached classifier
- add a manual .crow workflow to run it
- stay resumable, since cached verdicts return without touching the network, so
  a re-run only downloads what is still unclassified
2026-09-04 10:32:47 +00:00

90 lines
3 KiB
YAML

# Populate the ABI classifier cache for packages that predate the per-minor scheme.
#
# bincraft decides whether a flat binary may be carried into a per-minor index
# from its risky-package set. That set is now the union of the per-minor tarball
# scan and the `abi_classification` cache, but the cache only holds verdicts for
# packages that passed through the build pipeline after the scheme existed.
# Everything older has no verdict, and no verdict means "carry it" - which is how
# a base64enc built under R 4.4.3 reached R 4.6 clients and died at dyn.load()
# on SETLENGTH.
#
# The verdict is computed from the package source, so it does not depend on OS,
# arch or R minor: one run populates the cache for every slot. The arch/codename
# below only choose which flat index to enumerate.
#
# Re-running is cheap. Cached verdicts are returned without touching the network,
# so a second run costs one query per package and downloads only what is still
# unclassified.
#
# crow pipeline create devxy/build-cran-binaries --var backfill_abi=run
variables:
# A manual pipeline creation instantiates every file in .crow/, so the default
# must be a value the `when` clause rejects.
backfill_abi:
description: "Set to 'run' to classify the flat index into the ABI cache."
options:
- none
- run
default: none
BACKFILL_ARCH:
description: 'Arch of the flat index to enumerate.'
options:
- amd64
- arm64
default: amd64
BACKFILL_CODENAME:
description: 'Codename of the flat index to enumerate.'
default: resolute
BACKFILL_LIMIT:
description: 'Classify at most this many packages (0 = no limit).'
default: '0'
when:
- event: manual
evaluate: 'backfill_abi == "run"'
skip_clone: true
labels:
group: rpkgs-amd64
steps:
- name: 'Classify the flat index into the ABI cache'
image: reg.devxy.io/rpkgs/build-env-ubuntu:resolute
pull: true
environment:
OTEL_R_TRACES_EXPORTER: none
OTEL_R_LOGS_EXPORTER: none
OTEL_R_METRICS_EXPORTER: none
PGPASS:
from_secret: PGPASS
REPO_RO_TOKEN:
from_secret: REPO_RO_TOKEN
GITHUB_PAT:
from_secret: GITHUB_PAT
GIT_USER: pat-s
# Pipeline variables are substituted into the YAML, not exported, so they
# have to be bound here to be readable as env vars in the script below.
BACKFILL_ARCH: ${BACKFILL_ARCH}
BACKFILL_CODENAME: ${BACKFILL_CODENAME}
BACKFILL_LIMIT: ${BACKFILL_LIMIT}
R_LIBS_USER: /tmp/R-pkgs
commands:
- git clone -q https://pat-s:$$REPO_RO_TOKEN@git.devxy.io/devxy/build-cran-binaries.git .
- mkdir -p /tmp/R-pkgs
- /opt/R/4.5.3/bin/Rscript local/install-bincraft.R
- |
set -eu
LIMIT=""
if [ "$$BACKFILL_LIMIT" != "0" ]; then
LIMIT="$$BACKFILL_LIMIT"
fi
/opt/R/4.5.3/bin/Rscript local/backfill-abi-classification.R \
"$$BACKFILL_ARCH" "$$BACKFILL_CODENAME" $$LIMIT
backend_options:
docker:
resources:
requests:
memory: 4Gi
cpu: 1000m