Handle padded archive name tables

[zardus]

THIS MESSAGE WAS GENERATED BY AN AUTOMATED PROCESS

Problem

Some valid static archives use slash-newline entries in their GNU filename table and pad the table with NUL bytes. arpy selects NUL separation whenever any NUL occurs, so a later /offset member fails with ArchiveFormatError: file references a name not present in the index.

The affected samples are unavailable. Their verified SHA-256 identifiers and observed changes are:

• 0e712c71b3a16c8431e0d99fc4f0a3989744de05296a1793af72fb089a59c432: failed before; loads as an AMD64 StaticArchive with 77 members after.
• 4da54dd0ee922a62a3fdd17d0c0c08cd164a4a30f847bf4c2ef88c33fec49ff8: failed before; loads as an AMD64 StaticArchive with 2 members after.

The exact inputs and reproducer are unavailable.

Root cause

arpy selects NUL separation whenever any NUL occurs in a GNU filename table. That misclassifies trailing padding as a delimiter and makes valid slash-newline entries after the first one unreachable by offset.

Fix

Recognize NULs as padding only when the unpadded table is unambiguously slash-newline-terminated. Continue to treat all other tables containing NULs as NUL-delimited, and populate both arpy's eager and lazy lookup fields.

Testing

The regression test uses an existing tracked GNU archive to cover ordinary long-name lookup without distributing unavailable bytes. Both affected campaign objects changed from the recorded ArchiveFormatError to successful loads. 6 recorded-success StaticArchive controls spanning RISCV64, AMD64, X86, PPC64, PPC32, and AARCH64 retained identical backend, architecture, and member counts; 0 became worse.

The four commands below were executed from the same private measurement directory. The confidentiality rule forbids publishing private paths and control identifiers, so PRIVATE_TARGET_MANIFEST, PRIVATE_CONTROL_MANIFEST, and PRIVATE_CONTROL_1 through PRIVATE_CONTROL_6 replace only those exact values. The command structure and every argument are otherwise unchanged.

J297_BEFORE=/nix/store/c6jb9fgv21zql616kcq5ypxj803wmisa-python3-3.12.13-env
J297_TARGET_MANIFEST=PRIVATE_TARGET_MANIFEST
"$J297_BEFORE/bin/python" -P measure_archives.py --manifest "$J297_TARGET_MANIFEST"

J297_BEFORE=/nix/store/c6jb9fgv21zql616kcq5ypxj803wmisa-python3-3.12.13-env
J297_REGRESSION_MANIFEST=PRIVATE_CONTROL_MANIFEST
"$J297_BEFORE/bin/python" -P measure_archives.py --manifest "$J297_REGRESSION_MANIFEST" --sha PRIVATE_CONTROL_1 --sha PRIVATE_CONTROL_2 --sha PRIVATE_CONTROL_3 --sha PRIVATE_CONTROL_4 --sha PRIVATE_CONTROL_5 --sha PRIVATE_CONTROL_6

/nix/store/5hkdwc5mq2bxl6swbm11rvx6rd26kjd5-python3-3.12.13-env/bin/python -P measure_archives.py --manifest PRIVATE_TARGET_MANIFEST

/nix/store/5hkdwc5mq2bxl6swbm11rvx6rd26kjd5-python3-3.12.13-env/bin/python -P measure_archives.py --manifest PRIVATE_CONTROL_MANIFEST --sha PRIVATE_CONTROL_1 --sha PRIVATE_CONTROL_2 --sha PRIVATE_CONTROL_3 --sha PRIVATE_CONTROL_4 --sha PRIVATE_CONTROL_5 --sha PRIVATE_CONTROL_6

Validation: #851 (comment)

session: sharpen

[zardus]

THIS MESSAGE WAS GENERATED BY AN AUTOMATED PROCESS

Complete archive-loader result for both unavailable affected inputs, before and after this change.

Before — both archives fail while resolving a later GNU filename-table offset:

cle master 53d3d24

0e712c71b3a16c8431e0d99fc4f0a3989744de05296a1793af72fb089a59c432: arpy.ArchiveFormatError: file references a name not present in the index
4da54dd0ee922a62a3fdd17d0c0c08cd164a4a30f847bf4c2ef88c33fec49ff8: arpy.ArchiveFormatError: file references a name not present in the index

After — both archives load completely:

cle with this change 2591eed

0e712c71b3a16c8431e0d99fc4f0a3989744de05296a1793af72fb089a59c432: StaticArchive arch=AMD64 children=77
4da54dd0ee922a62a3fdd17d0c0c08cd164a4a30f847bf4c2ef88c33fec49ff8: StaticArchive arch=AMD64 children=2

[zardus]

THIS MESSAGE WAS GENERATED BY AN AUTOMATED PROCESS

Validation record for head 2591eedc5de92e03c81e2121ba4b78c60d88d76d against baseline 53d3d2419cfbb8a8c8741e97440cc1a9d6462559.

Candidate validation

• ./feature.sh build fix-c-276: passed; candidate environment /nix/store/5hkdwc5mq2bxl6swbm11rvx6rd26kjd5-python3-3.12.13-env.
• ../../../../nix/run.sh --feature fix-c-276 -- pytest --import-mode=append -q tests/test_static_archive.py: passed.
• Hosted-version merge-base pylint: cle/backends/static_archive.py improved from 9.67 to 9.81 with no W0201 diagnostics; tests/test_static_archive.py scored 10.00.
• Merge-base pyright: the source retained the baseline's one pre-existing parent_object error and introduced none; the new test had zero errors.
• ./feature.sh test fix-c-276: cle passed 277 tests with 9 skipped; angr passed 3,584 with 47 skipped and 2 xfailed. All family suites, package checks, pre-commit, feature lifecycle checks, Rust checks, and worktree cleanliness passed. angr-management did not run because this collection did not adopt it.
• The full gate exited 1 for two unrelated angr-agentic checks: campaign tests reported 25 failures and 20 errors, and mono's consistency test found that its driver omits pysoot. Both failures repeated on prior full gate runs and inspect only angr-agentic files.

Dataset validation

The baseline environment was /nix/store/c6jb9fgv21zql616kcq5ypxj803wmisa-python3-3.12.13-env; the exact candidate environment was /nix/store/5hkdwc5mq2bxl6swbm11rvx6rd26kjd5-python3-3.12.13-env.

The four commands below were executed from the same private measurement directory. The confidentiality rule forbids publishing private paths and control identifiers, so PRIVATE_TARGET_MANIFEST, PRIVATE_CONTROL_MANIFEST, and PRIVATE_CONTROL_1 through PRIVATE_CONTROL_6 replace only those exact values. The command structure and every argument are otherwise unchanged.

J297_BEFORE=/nix/store/c6jb9fgv21zql616kcq5ypxj803wmisa-python3-3.12.13-env
J297_TARGET_MANIFEST=PRIVATE_TARGET_MANIFEST
"$J297_BEFORE/bin/python" -P measure_archives.py --manifest "$J297_TARGET_MANIFEST"

J297_BEFORE=/nix/store/c6jb9fgv21zql616kcq5ypxj803wmisa-python3-3.12.13-env
J297_REGRESSION_MANIFEST=PRIVATE_CONTROL_MANIFEST
"$J297_BEFORE/bin/python" -P measure_archives.py --manifest "$J297_REGRESSION_MANIFEST" --sha PRIVATE_CONTROL_1 --sha PRIVATE_CONTROL_2 --sha PRIVATE_CONTROL_3 --sha PRIVATE_CONTROL_4 --sha PRIVATE_CONTROL_5 --sha PRIVATE_CONTROL_6

/nix/store/5hkdwc5mq2bxl6swbm11rvx6rd26kjd5-python3-3.12.13-env/bin/python -P measure_archives.py --manifest PRIVATE_TARGET_MANIFEST

/nix/store/5hkdwc5mq2bxl6swbm11rvx6rd26kjd5-python3-3.12.13-env/bin/python -P measure_archives.py --manifest PRIVATE_CONTROL_MANIFEST --sha PRIVATE_CONTROL_1 --sha PRIVATE_CONTROL_2 --sha PRIVATE_CONTROL_3 --sha PRIVATE_CONTROL_4 --sha PRIVATE_CONTROL_5 --sha PRIVATE_CONTROL_6

The two cited unavailable inputs changed from ArchiveFormatError to successful AMD64 StaticArchive loads with 77 and 2 members. 6 recorded-success controls across RISCV64, AMD64, X86, PPC64, PPC32, and AARCH64 loaded in both environments with unchanged backend, architecture, and member counts (1, 1, 15, 24, 123, and 253); worse: 0.

[angr-bot]

Corpus decompilation diffs can be found at angr/dec-snapshots@master...angr/cle_851

[ltfish]

Should we also send a PR to arpy?