npub1sh4dwqc9ypguemh6m8tmxxl8xuzfsz0j8av8vq7t3xd3mldsg9ls4ew80h/ngit-cli-mirror, branch v2.4.1

chore: release v2.4.1

2026-04-22T13:00:23+00:00

Fix fatal errors during clone/fetch when an open PR's git data is unavailable on the repository's specified git servers.

fix: prevent fatal clone/fetch errors when PR git data unavailable

2026-04-22T13:00:17+00:00

Only advertise `refs/heads/pr/*` branches once their tip OIDs are confirmed present locally; prevents `fatal: bad object` / `remote did not send all necessary objects` errors during clone/fetch when a PR tip lives on a different git server than the one that won the bulk prefetch race. After the bulk prefetch, collect remaining missing PR tip OIDs and do one batch fetch per repo git server using only the OIDs that server has advertised; break early once all are satisfied and skip servers that carry none. Avoids batch-poisoning (a server rejects the whole request if any single OID is absent) and redundant connections. Restrict mop-up fetches and run_fetch to the repo's declared git servers; do not fetch from clone-tag URLs in PR events - they are submitter-supplied and could let a malicious or slow server stall every clone/fetch operation. Also apply rustfmt and fix clippy warnings.

chore: release v2.4.0

2026-04-10T20:46:02+00:00

Add git worktree support and fix patch parsing, panic on bare npub clone, percent-encoding of reserved characters in identifiers, and push errors.

fix: percent-encode identifier in nostr:// URLs and GRASP HTTP paths

2026-04-10T16:30:02+00:00

Repository identifiers can contain any characters per NIP-01 d-tag rules. Encode them in nostr:// clone URLs (display and parse) and in GRASP //.git paths, aligning with NIP-34 and GRASP-01.

docs: update changelog with fixes since v2.3.0

2026-03-30T16:29:24+00:00

- more robust patch parsing and gracefully handle errors - panic when cloning a bare `nostr://npub/identifier` URL with no relay hints

feat: git worktree support

2026-03-30T16:29:23+00:00

Git worktrees don't have a .git directory with a parent, so we need to look for the git dir via git2's Repository::discover() and then look for the cache database there. This allows the client to work correctly when run from a worktree, and also allows the cache database to be shared between the main repo and its worktrees (since they share the same git dir and thus the same cache path).

fix(patch): accept reconstructed OID when commit id mismatch occurs

2026-03-30T16:29:22+00:00

When applying a patch, if the reconstructed commit OID doesn't match the tagged commit id (after the existing amend attempt), accept the reconstructed OID rather than erroring. The diff is applied correctly regardless of the mismatch; apply_patch_chain already threads the actual OID forward via next_parent_override so the chain remains consistent. Mismatches can occur for several reasons: GPG-signed commits where libgit2's commit_create_buffer produces subtly different bytes than the original git commit, patches produced with non-standard tooling, or edge cases like the diff.noprefix issue fixed in 7a36aed.

fix(patch): don't error on commit id mismatch for pgp-signed commits

2026-03-30T16:29:20+00:00

When applying a patch with a pgp signature, commit_create_buffer + commit_signed reconstructs the signed commit object. The resulting OID matches the original only if commit_create_buffer produces the exact same bytes that git originally signed — which isn't guaranteed across different git implementations or versions. The existing amend workaround (which re-applies author/committer to recover the correct OID) is preserved for unsigned commits. For pgp-signed commits the amend still runs but the mismatch bail is skipped, since the tree, author, committer, message and signature are all correct regardless of the OID difference.

fix(patch): handle diff.noprefix and normalize non-standard diff paths

2026-03-30T16:29:19+00:00

libgit2 respects the user's `diff.noprefix` git config when generating patches via `git_email_create_from_commit`, producing diffs without the standard `a/`/`b/` path prefixes. `git2::Diff::from_buffer` always expects the standard prefix format and fails with "header filename does not contain 1 path components" when it is absent. Two fixes: - In `make_patch_from_commit`: explicitly set `old_prefix("a/")` and `new_prefix("b/")` on the diff options so ngit always generates interoperable patches regardless of the submitter's git config. - In `create_commit_from_patch`: normalize no-prefix diffs before passing to `git2::Diff::from_buffer`, as a defensive measure for patches already published by affected ngit versions or other tools. Adds unit tests for the normalization function covering: no-op on already-prefixed diffs, single and multi-file patches, new/deleted files (/dev/null preservation), and end-to-end libgit2 parse verification.

fix(client): panic cloning bare npub nostr:// URL with no relay hints

2026-03-20T09:25:02+00:00

When cloning a nostr:// URL containing only an npub and identifier (no relay hints), and nothing is cached yet, the relay set was empty because fallback relays were intentionally skipped when a coordinate was present. This caused an unwrap() on a None from reduce() over an empty iterator. Fall back to fallback relays when the coordinate-derived relay set is still empty after processing all hints, and replace unwrap() with map_or() as a defensive guard.