Parents and children of check-in [a71bf662ee]
|
2026-10-19
| ||
| 01:53 |
fix(stash,undo): write working files through the checkout's storage
Stash and undo read, wrote and removed working files with the os package, so a MemStorage checkout and simulation tests never saw those writes, and no write recorded its mtime. They now take a *simio.Env and go through its Storage, recording each written file's mtime via vfile.WrittenMTime (moved with RecordableMTime from internal/checkout so these packages can share the same-second guard). Stash save marks reverted rows unchanged with their written mtime, as fossil's revert does. Apply and pop add a stashed added file again, as fossil's stash_apply does, and give a restored edit's row its changed state and written mtime. Undo and redo record a written file's mtime only when the restored row's mtime matched the file when it was saved. Tracked in #448 check-in: 41951ffbc5 user: [email protected] tags: feat/submanifest-reader, main, trunk | |
| 01:50 |
perf(verify): check blob integrity by walking delta trees once
Verify and Rebuild expanded every blob one at a time through a 256 MiB LRU cache. A real repository's delta chains run hundreds of revisions deep, so evicted chains were walked again from their roots: the blob check on the Fossil SCM repository ran for over 45 minutes. content.ExpandAll walks each delta tree from its root, as fossil's rebuild_step does, expanding every blob once from its parent's content and binding it to its name. It reports for each blob what a caching Expand of that blob alone reports, and expands blobs no sound root reaches (cycles, phantom-grounded or too-deep chains) on their own for that error. checkBlobs and Verify's checkCheckins both use it and sort their issues by rid, so reports are unchanged. check-in: 9a9bc65f92 user: [email protected] tags: feat/submanifest-reader, main, trunk | |
| 01:50 |
fix(merge): write fossil's merge3 output byte for byte
Port fossil 2.28's merge3.c and the diff it runs on (break_into_lines, break_into_tokens, longestCommonSequence, diff_step, diff_optimize, the UTF-8/UTF-16 BOM handling) so Merge and Update align, merge and mark conflicts exactly as fossil does: line-numbered markers, the token-level suggested resolution, CRLF markers, and binary input left unmerged. HasConflictMarkers follows contains_merge_marker in skipping CR runs. A fossil-binary comparison test checks 1237 generated and hand-written cases against "fossil 3-way-merge" and UTF-16 cases against "fossil merge"; fossil commit refuses a file holding our markers. Tracked in #447 check-in: dcec15d91f user: [email protected] tags: feat/submanifest-reader, main, trunk | |
|
2026-10-18
| ||
| 21:59 |
fix(checkout): keep uncommitted changes across Update, as fossil does (#450)
* fix(checkout): keep uncommitted changes across Update, as fossil does Update compared only committed versions: it wrote the target over a locally edited file, deleted an edited file the target removed, and merged the current and target check-ins with their ancestor while ignoring the working file. It now follows fossil's update (src/update.c): scan the checkout, then for each file take the target's content when there is no local change, merge a local edit three ways (the checkout's version as base, the working file as local, the target as other) with conflicts listed in Conflicted, keep an edited file the target deletes as an untracked file, keep pending adds and removals, restore a missing file, and copy an untracked file the target adds over to NAME-original first. A dry run writes nothing. Files written with the target's content record the mtime of the write. Files are matched by name; a target-side rename is not followed. The two tests that relied on merging two branches' committed changes now make the clashing edit locally. A fossil 2.28 interop test runs the same edits through fossil update and compares files, fossil changes and Status. Tracked in #443 * fix(checkout): make Update safe on binary files, failures and dry runs A binary file (a NUL byte, or a line over 32,767 bytes, as fossil's line diff refuses) is no longer line-merged. Update saves the working copy as NAME-original and writes the target, as fossil's update does; Merge keeps the checkout's copy, as fossil's merge does. Both used to splice the versions and report a clean merge. merge.IsText is fossil's test. Update refuses before writing anything when a file it must write or delete has a directory or a read-only file in the way, moves the checkout database to the target in one transaction, and in a dry run leaves the database untouched (the scan's findings stay in memory). Every target file gets the target's executable bit, as fossil's checkout_set_all_exe does, so a bit-only change reaches the disk. UpdateResult.Unmerged lists files left to reconcile without markers: edited files the target deletes, files added both locally and by the target, and binary conflicts. The interop test now covers a binary conflict against fossil 2.28. Tracked in #443 * fix(checkout): check real write access before Update writes anything The pre-write check read the owner-write bit: it missed an unwritable directory (a remove failed after a merge had been written), files owned by others, and wrongly refused for root. It now asks the storage, via an optional CheckWritable: OSStorage uses access(2) with W_OK on unix and a read-only-attribute check elsewhere (js/wasm, Windows). Every step needs its directory writable; a write or merge over an existing file needs the file writable too. A read-only untracked file where the target adds one is backed up to NAME-original, deleted and replaced, as fossil does, instead of refused. Merge keeps a binary file whatever its strategy, as fossil's merge does. setVVar's precondition now catches a nil *sql.DB or *sql.Tx inside the interface. Tracked in #443 check-in: 6e6969bf97 user: [email protected] tags: main, trunk | |
| 20:59 |
fix(checkout): record and trust vfile.mtime as fossil does (#444)
* fix(checkout): record and trust vfile.mtime as fossil does go-libfossil wrote vfile.mtime = 0 everywhere and hashed every file on every scan, so it and fossil rehashed a go-libfossil checkout on every command. The scan is now fossil's vfile_check_signature: a file whose size matches its artifact (blob.size of mrid) and whose mtime matches vfile.mtime is taken as unchanged without being read, unless ScanHash is given or the mtime-changes setting is off; the scan stores the file's mtime in whole seconds and runs its updates in one transaction. Status, commit, merge, revert and extract use the default scan, as fossil's commands do. Mtimes are recorded where fossil records them: extract and update store the mtime of each file written with its artifact's content, an unchanged file keeps its mtime through LoadVFile (fossil's update transfer and in-place commit), committed files keep theirs, revert stores the restored file's, and SetMTime stores the time it gives. Undoing a rename drops the mtime, since the file was not checked. One deliberate difference: an mtime not older than the current second is stored as 0, so the next scan (ours or fossil's) hashes the file and an edit that keeps the size within that second is still seen. A stored 0 is never trusted. MemStorage now stamps each write with a new mtime from its own clock, as a filesystem does, so in-memory checkouts see same-size edits. Status on the sqlite repository's 2,220-file checkout: about 250 ms before, about 30 ms after. * test(checkout): pin the same-second guard against fossil A same-size edit in the second a file was scanned is committed, and fossil 2.28 reading a row the guard stored as 0 reports the edit. Both tests fail with the guard removed. The guard's comment and the CHANGELOG name it as the one deliberate difference from vfile_check_signature. * fix(checkout): record the mtime of each write as it happens Extract and update statted their files only after every file was written and vfile reloaded, so an edit in between was recorded as the written content. They now stat each file right after writing it, as fossil's vfile_to_disk does, and carry that mtime to the row. Revert and the new writtenMTime read the clock before the stat, as the scan does. A caller's Env without a Clock now gets the real clock, as Storage defaults to the OS; scans read it. The mtime-changes setting is read from the repository only: go-libfossil reads no global fossil settings, which the comment and CHANGELOG now say. New tests: an edit after extract's or update's write is seen, a stored 0 is never trusted, and a simulated clock at a realistic time drives the mtime fast path. check-in: a71bf662ee user: [email protected] tags: main, trunk | |
| 20:01 |
fix: handle time exactly as fossil does (#442)
* fix(manifest): round a caller's check-in and tag time as fossil does Fossil reads a --date-override through SQLite's strftime, which rounds the seconds to the millisecond (holding a fraction past .999 in its second) rather than truncating them. A caller time finer than a millisecond now gets the D card fossil would write for it; the default clock time is still truncated, as fossil reads its clock. fsltype.CardTime and CardTimeNow are the one place both rules live, and the checkout commit path and branch creation take their default time through them. * fix(repo): stamp config and user rows in seconds since 1970 Fossil writes config.mtime and user.mtime with its now() SQL function, time(0). SetConfig, SetHashPolicy and CreateUser wrote julianday('now') instead, so fossil read those rows as dating from 1970 when comparing times in a configuration merge, and SetCaps and SetPassword left the user's mtime alone where fossil's user command restamps it. * fix(auth): read and write user.cexpire as a Julian day Fossil keeps user.cexpire as a Julian day (src/login.c). A sync login parsed it as RFC 3339 or 'YYYY-MM-DD HH:MM:SS' text, so any value fossil had written failed to parse and the login was refused as expired. The expiry is now read through julianday(), which takes both forms, and repo invite --ttl writes a Julian day; its write error is now checked. * fix(checkout): set file times as fossil's --setmtime does Fossil's --setmtime gives each unedited file the time of the newest check-in, among the version and its primary ancestors, that recorded the file's content, truncated to whole seconds (mtime_of_manifest_file), and it does so for every unedited file of the version once the checkout is up to date (vfile_check_signature with CKSIG_SETMTIME). SetMTime gave every written file the target check-in's time to the millisecond, and an update left the files it did not write alone. * fix(stash): keep and show stash and bisect times as fossil does Fossil stores stash.ctime as julianday('now') and lists it with datetime(ctime); go-libfossil stored CURRENT_TIMESTAMP text, which the two drivers returned in different forms, and showed a fossil-written time as a bare number. bisect's list now shows datetime(event.mtime), as fossil's bisect ls does, instead of the Julian day. * test(verify): fossil rebuild keeps the times go-libfossil stores A repository go-libfossil wrote, with check-ins at caller and clock times, a tag and a branch, is rebuilt by the fossil binary, and every check-in event, plink and tagxref time must come back unchanged. * test(sync): a fossil clone and push agree on every synced time The fossil binary clones a go-libfossil server, unversioned files included, commits at a time finer than a millisecond and pushes back. Each side derives check-in event, plink and tagxref times from the artifacts itself, and the unversioned time travels on the wire; both repositories must then hold the same values. * fix(resolve): bound NAME:TIMESTAMP with julianday() as fossil does Fossil resolves BRANCH:TIMESTAMP with event.mtime<=julianday(BOUND). The bound was computed in the text form event rows hold, so on a millisecond where that form is larger than julianday()'s, a check-in at the bound itself resolved here and not in fossil. * fix(manifest): round a caller's card time with integer arithmetic CardTime followed SQLite's printing of the seconds as a double, which is not a well-defined rounding: at exact half-millisecond ties fossil 2.28's SQLite gives .124 for .1245 but .020 for .0195, and other SQLite versions differ again. A caller time is now rounded to the nearest millisecond on its nanoseconds, ties up, held at .999 as fossil holds it. The fossil binary test compares every non-tie value, across second, minute, day and year ends; ties are asserted on their own. * fix(auth): ignore user.cexpire at sync login, as fossil does user.cexpire is the expiry of fossil's web-login cookie: fossil sets it at web login, writes 0 at logout, and its sync login (check_login) never reads it. Refusing a sync login on it locked out anyone who had logged into or out of fossil's web UI, or whose cookie had lapsed. The invite TTL that relied on it now lives in a config row of its own, invite-expire:LOGIN, as a UTC instant checked at sync login; repo invite no longer writes cexpire, and the expiry is no longer written in local time and read as UTC. Deleting a user drops its invite expiry. Tracked in #441. * fix(repo): convert Julian config and user mtimes on open Repositories go-libfossil wrote before config and user rows were stamped in seconds since 1970 still hold Julian days there, which fossil reads as January 1970. Open converts those rows to the seconds they stand for. It checks first and writes only when a row needs it, so it converts once and takes no write lock on an up-to-date repository. * fix(checkout): set a file no check-in recorded to 1970, as fossil does With no mlink row for a file's artifact in the version's ancestry, fossil's max(event.mtime) is NULL, which mtime_of_manifest_file reads as 0, so --setmtime sets the file to 1970. SetMTime skipped the file. * fix(stash): list stashes by creation time, as fossil's stash ls does fossil stash ls orders by ctime, oldest first. List ordered by stash id, newest first; it now orders by ctime with the id breaking a tie. * fix(auth): clear a login's invite expiry with the user, atomically DeleteUser removed the user and its invite expiry in two statements; a failure between them left the user's expiry behind or the user without it. Both now go in one transaction. A user fossil deleted leaves its invite-expire row behind, so CreateUser clears it in the same transaction as the insert; repo invite sets its own expiry afterwards. The on-open conversion of Julian config and user mtimes is now best-effort: while another connection holds the write lock, Open skips it and a later open converts the rows, instead of failing. check-in: d810a8118f user: [email protected] tags: main, trunk | |