datalad/datalad: 1.5.0

Le résumé fourni par la source

🚀 Enhancements and New Features Add --cfg-proc option to install, clone, and get to run configuration procedures on datasets post-installation. Fixes #7468 via PR #7477 (by @bpinsard) NF+BM: extend create_test_dataset spec, add save/run benchmarks on heavy hierarchies. PR #7839 (by @yarikoptic) 🐛 Bug Fixes Fix AltConstraints.__or__() and Constraints.__and__() mutating their left-hand operand in place and returning self. This caused silent side effects when constraint objects defined at module level (e.g. reckless_opt) were reused across multiple | or & chains. Both operators now return a new instance, preserving the original. Identified via the constraint system overhaul in datalad-next. Fixes #7164 (by @yarikoptic) Propagate datalad -c key=value CLI config overrides and DATALAD_* environment variables to subprocesses via Git's native environment-based configuration mechanism (GIT_CONFIG_COUNT / GIT_CONFIG_KEY_N / GIT_CONFIG_VALUE_N). Previously the overrides were only stored in datalad.cfg.overrides and were invisible to subprocesses spawned by commands like datalad run, so e.g. datalad -c user.name=Test run git config user.name did not return the override. Adapted near-verbatim from the cli_configoverrides patch in datalad-next, with its get_gitconfig_items_from_env / set_gitconfig_items_in_env helpers also added under datalad.config. Fixes #4119 (by @yarikoptic) ENH add --data option to get command, matching push one. PR #7476 (by @bpinsard) chore: deduplicate code across interface docs, create-sibling, s3, and benchmarks. PR #7841 (by @yarikoptic) Do not fail add-archive-content when adding files ignored due to .gitignore rules. PR #7844 (by @asmacdo) Fix the dataset-level "Total" progress bar of datalad get (and other JSON-progress annex commands): advance it on per-file byte-progress deltas (throttled to 0.5s) instead of only on completion, surface errored files in the bar label ([errored: N / ]) and remove their key-sizes from the effective total, and roll back any provisional credit when a partially-streamed file errors. The bar now tracks cumulative on-disk bytes for non-errored files; resumed-from-disk bytes are tracked as a separate anchor pushed to tqdm via initial= so the rate / ETA display reflects session work only. PR #7865 (by @yarikoptic) Fix tqdmProgressBar.update so a changed total= is applied directly rather than via tqdm.reset(), which used to zero the elapsed-time and EMA state and produce momentary nonsense rates (e.g. [00:00<00:00, 16.0T Bytes/s]). PR #7865 (by @yarikoptic) Resolve --jobs auto for push, drop, and remove so the literal string auto is no longer passed through to git annex. Previously push --jobs=auto crashed with option --jobs: cannot parse value 'auto', while drop --jobs=auto (and remove --jobs=auto, which delegates to drop) swallowed the same CommandError and silently left content in place. Fixes #7867. PR #7868 (by @yarikoptic) 🏎 Performance datalad push no longer runs a pointless git push --dry-run and refspec computation when the target is a git-annex special remote (e.g. WebDAV, S3, directory) that has no git URL. This was wasteful at best and could cause errors or significant slowdowns for remotes like WebDAV. The target_is_git_remote check is now performed before the dry-run block and guards the entire dry-run + refspec computation. Ported from the push_optimize patch in datalad-next. Fixes #6657 (by @yarikoptic) 🧪 Tests Resolve @use_cassette('name') against /vcr_cassettes/ .yaml so cassettes can ship next to their tests. PR #7852 (by @yarikoptic) Ship VCR cassette for test_anonymous_s3 so it stops hitting live S3. PR #7852 (by @yarikoptic) Drop the annex.pidlock=true NFS workaround. Its v7-era polling-based locking became the dominant cost on NFS after the v8→v10 git-annex bump and was the root cause of the cron NFS-job 6h timeouts; v10 + NFSv4.2 handles fcntl natively. PR #7852 (by @yarikoptic) Retry SSH-target docker setup on host port collisions in tools/ci/prep-travis-forssh.sh. PR #7852 (by @yarikoptic) Export the cron NFS test mount with async for more-realistic-than-default test conditions and a small per-write speedup. PR #7852 (by @yarikoptic) Capture per-test timings as a pytest-report.jsonl artifact via pytest-reportlog, preserving timing data even when a job is cancelled. PR #7852 (by @yarikoptic) BF(TST): add signal_timeout helper and bound test_gin_cloning. PR #7859 (by @yarikoptic)

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.