tricu

archive_probe.py (4901B)

---

 #!/usr/bin/env python3
 """Probe the pinned archives' readiness for a C-loadable core shared library.
 
 Invoked by archive-probe.nix with the same reducer/API sources and ordinary -O2
 settings as build.nix. A recognized non-PIC link failure is diagnostic data;
 unrelated build errors fail the derivation. A successful link is not installed
 or presented as an accepted runtime: packaging/lifecycle/performance acceptance
 remains separate.
 """
 import argparse
 from collections import Counter
 import gzip
 import json
 from pathlib import Path
 import re
 import subprocess
 import sys
 
 
 def classify(returncode, log):
     if returncode == 0:
         return 'link-succeeded'
     errors = [line for line in log.splitlines() if 'error:' in line]
     def known_error(line):
         return bool(re.search(r'error: .*\.a\(.*(?:recompile with -fPIC|unsupported reloc \d+ against global symbol)', line)
                     or re.search(r'collect2: error: ld returned \d+ exit status', line))
     if (re.search(r'error: .*\.a\([^\n]+recompile with -fPIC', log)
             and all(known_error(line) for line in errors)):
         return 'blocked-non-pic-archives'
     raise RuntimeError('Archive probe failed for an unclassified reason; inspect its build log')
 
 
 def main():
     # Reject ordinary compile failures and mixed-cause link failures rather than
     # misreporting every unsuccessful build as a known toolchain limitation.
     assert classify(0, '') == 'link-succeeded'
     example = 'ld.gold: error: /tmp/libHSx.a(X.o): recompile with -fPIC'
     assert classify(1, example) == 'blocked-non-pic-archives'
     for bad in ('Main.hs: error: missing name', example + '\nld: error: undefined symbol'):
         try:
             classify(1, bad)
         except RuntimeError:
             pass
         else:
             raise AssertionError('unrelated failure accepted')
     parser = argparse.ArgumentParser(description=__doc__)
     parser.add_argument('--output', required=True, type=Path)
     args = parser.parse_args()
     args.output.mkdir(parents=True, exist_ok=True)
     command = [
         'ghc', '-O2', '-static', '-shared', '-fPIC', '-flink-rts', '-threaded',
         '-optl-Wl,-z,nodelete',
         '-XOverloadedStrings', '-XScopedTypeVariables', '-XMultiWayIf', '-XLambdaCase',
         '-Wall', '-isrc', '-iext/haskell/src', '-outputdir', 'objects',
         '-optc-Iext/haskell/include',
         'ext/haskell/src/EmbeddingCore.hs', 'ext/haskell/src/init.c',
         '-o', 'libtricu-core.so',
     ]
     ghc_version = subprocess.check_output(['ghc', '--numeric-version'], text=True).strip()
     if tuple(map(int, ghc_version.split('.')[:2])) >= (9, 14):
         command.insert(1, '-optl-Wl,-u,init_ghc_hs_iface')
     result = subprocess.run(command, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
     log = result.stdout.decode('utf-8', errors='replace')
     # Keep full diagnostics compressed rather than filling the console with tens
     # of thousands of repeated relocation errors. They are not runtime data.
     with gzip.GzipFile(args.output / 'link.log.gz', 'wb', mtime=1) as stream:
         stream.write(result.stdout)
     try:
         status = classify(result.returncode, log)
     except RuntimeError:
         print(log[-16000:], file=sys.stderr)
         raise
     errors = [line for line in log.splitlines() if 'error:' in line and 'recompile with -fPIC' in line]
     archives = Counter(name for line in errors for name in re.findall(r'/([^/ ]+\.a)\(', line))
     report = {
         'schema': 'tricu.core-archive-probe.v1',
         'ghc_version': ghc_version,
         'link_status': status,
         'link_exit_code': result.returncode,
         'command': command,
         'non_pic_archives': dict(sorted(archives.items())),
         'relocation_examples': [re.sub(r'/nix/store/[^ ]+?/(libHS[^ ]+\.a|bin/ld.gold)', r'\1', line)
                                 for line in errors[:3]],
         'runtime_artifact_accepted': False,
         'size_and_performance_measured': False,
     }
     if status == 'link-succeeded':
         # Immediate ABI sanity checks; full installed-runtime and timing checks
         # are deliberately not inferred from a successful linker invocation.
         subprocess.run([
             'cc', '-Wall', '-Wextra', '-Werror', 'ext/haskell/tests.c',
             '-Iext/haskell/include', '-L.', '-ltricu-core',
             '-Wl,-rpath,' + str(Path.cwd()), '-o', 'core-client',
         ], check=True)
         subprocess.run(['./core-client'], check=True)
         subprocess.run([sys.executable, 'ext/haskell/test_exports.py',
                         './libtricu-core.so', 'core'], check=True)
         report['c_client_and_exports'] = 'passed'
     else:
         report['c_client_and_exports'] = 'not-run-link-blocked'
     (args.output / 'report.json').write_text(json.dumps(report, indent=2) + '\n')
     print(json.dumps(report, indent=2))
 
 
 if __name__ == '__main__':
     main()