feat(#83): --archive / -z control surface + libarchive-backed reader - #436
Open
helly25 wants to merge 5 commits into
Open
feat(#83): --archive / -z control surface + libarchive-backed reader#436helly25 wants to merge 5 commits into
helly25 wants to merge 5 commits into
Conversation
First archive slice: the ratified control surface, with no diving yet. --archive[=none|roots|all] with the short -z-/-z/-z+ (chmod-style suffix signs, the -g gitignore family's spelling). The modes are nested, so one ordered enum expresses two separately-wanted behaviors: none keeps find's opaque-archive behavior, roots dives only into an archive named AS a root, all also dives ones met mid-walk. Bare --archive is all. The find style defaults to none, the xff family to roots. Fixes three real problems in the existing extras gate found while wiring this: - the short forms slipped past it entirely (-z+ produced a different, later error than --archive), because the gate looked up the arg without stripping the suffix sign; - --archive=none demanded a rebuild with the archive extra, although asking for find's behavior is exactly what a lean build already does; - the old --archive row had no value grammar, so --archive=none was rejected as an unknown option. The two "no" states are now distinct and never confusable: a build without the extra says "no archive support; rebuild with --//xff:archive", while a build that has it but cannot dive yet says "not yet implemented". Only an EXPLICIT request trips either, so the xff family's roots default never breaks a plain walk. Extracts ReadHashDefaults so RunFind stays under the clang-tidy function-size threshold (the guard pushed it to 802/800 statements). Tests: xff/cli/archive_test.sh pins every spelling on both sides of the split, the untouched default walk, and the documented modes. XFF.md regenerated.
The extras gate and --help=extras derived the rebuild hint as --//xff:<extra>, but the actual flags carry an xff_ prefix (//xff:xff_archive, //xff:xff_pcre). Both therefore printed a flag that does not exist, which is worse than no hint. Adds ExtraBuildFlag() next to ExtraEnabled(), mapping each extra key to its real flag label explicitly rather than by a derivation rule - the keys and the flags genuinely differ (pcre2 -> xff_pcre), so any rule would be wrong for one of them. An unknown extra yields no label and the caller drops the hint instead of inventing one. archive_test pins both surfaces.
Three more sites still printed or documented the non-existent flag: the help renderer's not-built note, the --archive flag's own details prose, and a BUILD comment. extras_test was pinning the wrong string, so it is corrected too (it asserted the bug rather than the behavior). All four surfaces now name --//xff:xff_archive, and XFF.md is regenerated.
helly25
enabled auto-merge (squash)
August 9, 2026 23:50
…ule (#437) * feat(#83): libarchive-backed archive reader as a standalone extra module Second archive slice: the dependency and the reader, still with no walk integration. extra_modules/archive/ is its own Bazel module (@xff_archive), the same shape as the PCRE2 extra: it depends on @libarchive (from the BCR, no vendoring) and the shared @xff_extras_api seams, never back into the xff core, so deleting the directory drops the extra entirely. Verified: the lean //xff/cli:xff has zero libarchive in its dep graph. archive_reader lists an archive's members from memory or by streaming a file. libarchive detects the format and compression filter from content, so tar / zip / cpio / ar and the gz / bz2 / xz / zstd / lz4 filters all come through one entry point. It reads headers only, never member content, so listing a huge archive stays cheap. The error contract matters for the walk and is pinned by tests: "not an archive" (InvalidArgument) must stay distinct from "corrupt archive" (DataLoss) - the first is an ordinary file, the second is a real error to report. Empty input is explicitly not an archive: libarchive opens zero bytes and reports EOF, which would otherwise make an empty file look like a valid archive holding nothing. Licensing: libarchive is BSD-2-Clause and its codec closure (zlib, bzip2, liblzma, lz4, zstd on its BSD arm) is permissive throughout; mbedtls stays off, so no crypto arm is pulled. The notice registers itself from the reader's translation unit through the existing license-notice seam, so --help=notice and the NOTICE file stay complete by construction. Tests write real tar and gzip-filtered tar archives with libarchive's write API and read them back, so there are no committed binary fixtures and the whole dependency is exercised end to end. * fix(#83): correct the remaining stale --//xff:archive references Three more sites still printed or documented the non-existent flag: the help renderer's not-built note, the --archive flag's own details prose, and a BUILD comment. extras_test was pinning the wrong string, so it is corrected too (it asserted the bug rather than the behavior). All four surfaces now name --//xff:xff_archive, and XFF.md is regenerated.
helly25
changed the title
Owner
Author
|
Note: this PR now carries both archive slices. The reader PR (#437) was stacked on this branch and auto-merged into it immediately - a PR based on a feature branch has no required checks, so auto-merge fires at once. So review here covers:
Full CI runs on this PR since its base is |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
First archive slice (#83): the ratified control surface, with no diving yet.
--archive[=none|roots|all]plus the short-z-/-z/-z+(chmod-style suffix signs, the-ggitignore family's spelling). The modes are nested, so one ordered enum expresses two separately-wanted behaviors:nonekeeps find's opaque-archive behavior,rootsdives only into an archive named AS a root,allalso dives ones met mid-walk. Bare--archiveisall. The find style defaults tonone, the xff family toroots.Four real bugs this surfaced, all fixed with tests
-z+produced a different, later error than--archive(the gate looked up the arg without stripping the suffix sign).--archive=nonedemanded a rebuild with the archive extra, although asking for find's behavior is exactly what a lean build already does.--archiverow had no value grammar, so--archive=nonewas rejected as an unknown option.--//xff:archive, which is not a real flag (it is--//xff:xff_archive). Four sites printed or documented it, andextras_testwas asserting the broken string.ExtraBuildFlag()now maps each extra key to its real flag explicitly, because a derivation rule cannot work for both (pcre2maps toxff_pcre).The two "no" states are now distinct and unconfusable: no extra -> "no archive support; rebuild with --//xff:xff_archive"; extra present but diving unbuilt -> "not yet implemented". Only an EXPLICIT request trips either, so the xff family's
rootsdefault never breaks a plain walk.Also extracts
ReadHashDefaultssoRunFindstays under the clang-tidy function-size threshold (the guard pushed it to 802/800 statements).Tests:
xff/cli/archive_test.shpins every spelling on both sides of the split, the untouched default walk, the documented modes, and that both surfaces name real Bazel flags. XFF.md regenerated.