404 Solution

Registro de alterações

Version 4.3.4 (August 20, 2026)

Bug Fixes

• Fixed concurrent diagnostic writers being able to discard the newest retained journal generation. A writer whose file had already been rotated could retain its old near-limit size and mistakenly rotate a much smaller live file; destructive rotation now revalidates that the writer still owns the live path before renaming or deleting any journal.
• Fixed sites with more than 51 active regex redirects evaluating only the 51 oldest rules. The cache safety check used a 51-row query to decide whether the result was small enough to cache, but when it found more rules it kept using that same limited query instead of continuing through the table. Newer regex redirects therefore stayed at “Never Used” and matching requests fell through to the 404 page. Regex rules are now read completely in bounded batches, preserving their existing oldest-first precedence without imposing a matching limit.
• Fixed a fatal error on the plugin’s own Page Redirects and Captured 404s screens for sites running WordPress 5.0, 5.1 or 5.2. The plugin supports WordPress 5.0 and above, but eight places that render a date called a WordPress function that only arrived in 5.3, so those screens died with “Call to undefined function wp_date()” instead of rendering.
• Fixed a redirect loop in which a request was answered with a redirect back to that same request. When a 404’s destination was the home page and the request carried a query string such as ?page=1, the destination was rebuilt into the original URL, so a browser or crawler looped until it gave up. Every hop was counted as another 404 hit, and one site recorded 3,892 hits on a single row this way. The home destination now resolves to the home page’s own path, and no redirect is sent when the destination is the request being answered.
• Fixed a fatal error during a plugin update, when a request that started under the old version was still running as WordPress replaced the plugin files. The plugin now notices its own files changed mid-request and re-reads its class list, instead of looking up the new files through the previous version’s list and failing with “Class ABJ_404_Solution_… not found”.
• Fixed the Confidence column on the Captured 404s tab reading as a dash for every row on every install. The match score was worked out when the 404 was captured and then thrown away, so the column, its tooltip and the two indexes that exist to sort by it had no value to work with. Captured 404s now record the score of their closest match.
• Fixed apostrophes and quotes being stored with a backslash in front of them, so text typed as “didn’t” was saved as “didn\’t”. This affected support and uninstall feedback and other text saved from the admin screens. The redirect destination and 404 log search boxes had the same problem, so searching for a term containing an apostrophe matched nothing.
• Fixed the plugin emailing “Failed to schedule cron hook” reports for scheduling requests that had not failed. WordPress refuses a request when an equivalent scheduled task already exists, and reports that the task list “could not be saved” when the list is written with no change; both are normal outcomes that mean the task is scheduled. One site received 18 copies of the same message, 14 of them inside one second.
• Fixed the reverse scheduled-task check: a failed recurring schedule could be reported as successful when a different occurrence of the same task already existed. Exact schedule and removal checks now require the requested timestamp, and malformed WordPress cron responses are logged with their original context instead of being treated as success.
• Fixed two simultaneous requests being able to schedule separate links in the permalink-cache background chain. The existing hook-wide check stopped a later request from restarting an already-visible chain, but the check and cron write were separate, so racing requests could both observe an empty store and queue events with different arguments. Scheduling that chain is now serialized by an expiring database claim, and the claim holder refreshes WordPress’s process-local cron cache before deciding, preserving the chain’s execution limit without leaving a permanent lock if a request dies.
• Fixed canonical pagination handling terminating the request and recording a successful hit after WordPress rejected the redirect. A rejected redirect now keeps normal 404 handling active and records the destination, status and source in the debug log. Its captured evidence row is now inserted only when the source is absent, inside a serializable database transaction that prevents two simultaneous requests from both inserting it, and duplicate cleanup preserves a user-authored redirect over newer captured evidence.
• Fixed schema checks, database upgrades and internal database lock rows being able to wait effectively forever behind another connection’s schema change. Metadata-lock waits for those operations are now bounded, the site’s previous database-session setting is restored afterward, and the original database error remains available in the debug log.
• Fixed database index repair retrying an ALTER TABLE after failures that removing the optional online-DDL clauses could not cure, such as permission or storage errors. The bare fallback now runs only when the database specifically rejects ALGORITHM=INPLACE or LOCK=NONE.
• Fixed the plugin retrying forever when WordPress refused to remove one of its scheduled tasks. Nothing in the retry could make progress, so a deactivation or cleanup request kept asking until the server’s time limit killed it. Security and scheduled-task manager plugins can refuse the removal, which is what made this reachable.
• Fixed a code path that could wait for one of the plugin’s internal locks with no time limit at all. On a read-only database replica, a full disk, or an unwritable uploads directory the wait could never succeed, so the request slept and retried until the server’s time limit killed it.
• Fixed the same internal lock being granted to two requests at once on sites with no persistent object cache. Ownership was decided by writing a value and reading it back, but WordPress answered the read from its own in-memory copy, so each competing request read back its own write and concluded it had won. Two database upgrades were seen starting in the same second on one site.
• Fixed two simultaneous requests entering the same database upgrade and the losing one reporting the winner’s success as a string of errors, including a second change that could not succeed because the change it asked for had already been made.
• Fixed “COLLATION ‘latin1_swedish_ci’ is not valid for CHARACTER SET ‘utf8mb4′” database errors repeating every few seconds on sites whose database settings still name a latin1 collation. The permalink cache never refreshed on those sites and the published-content lookup returned nothing.
• Fixed “Illegal mix of collations” errors when the plugin compared its own log rows against WordPress’s posts table on sites where the two tables were created with different collations. The comparison returned no rows rather than failing visibly, so the check that depends on it silently found nothing.
• Fixed the plugin repeatedly trying to add a column named “using” of type “btree” to its redirects table, and logging the resulting database error, on sites where an index definition ended in USING BTREE. The error was logged three times per upgrade and three more times per scheduled run, indefinitely.
• Fixed the plugin rewriting a database table that was already correct. A table’s character set was read from the first column that named one rather than from the table’s own settings, so a correctly configured table could be read as out of date and converted when nothing had drifted.
• Fixed one of the plugin’s index repairs naming a column without first checking that the column exists. That repair removes and re-adds an index in a single statement, so a database server that applied only half of it could leave the log table with neither index.

Improvements

• The Simple Mode setting for automatic redirects now says what the match score has to reach and where to change it, instead of only saying that a good match is required. It names the score actually in force on the site rather than the shipped default.
• The page that offers suggestions for a 404 now explains, to administrators only, what the number beside each suggestion means, what the current bar for an automatic redirect is, and the two ways to act on it: redirect that URL, or change the bar.

Internationalization

• Completed the translations for every language the plugin ships. 80 entries in Swedish and 846 across the other catalogs were still being shown in English; each is now translated, or marked to record that the English word is the right one in that language.
• The support request dialog, the confidence chart on the stats screen and the plugin migration tool now have translations. Their text previously rendered in English in every language, no matter how complete that language’s translation was.
• Restored accented characters in seven Latin-script translations where entries had been typed without them.

Version 4.3.3 (August 12, 2026)

New Features

• Regex redirect destinations can now be site-relative paths such as /archive/$1, not only full http(s):// URLs.

Bug Fixes

• Fixed the Page Redirects and Captured 404s tabs being able to hang after their data was already built. When something else on the site holds an output buffer open that cannot be closed, whether that is PHP output compression or another plugin, the response could keep trying to flush buffers until the host’s execution limit killed the request. Flushing is now bounded, so the table finishes rendering instead of spinning.
• Fixed a fatal error that could happen while updating from 4.2.x, when the server’s PHP bytecode cache was still handing out the previous version’s files as the new ones loaded. This guard protects future updates; it cannot protect the update that installs it.
• Fixed a crashed or killed request being able to leave an internal lock behind, which left redirect lookups on a slower fallback path until the lock expired (more than 24 hours was observed on one site). Locks are now released even when the request dies, and an abandoned lock is reclaimed after at most 5 minutes instead of after the server’s maximum execution time.
• Fixed “Illegal mix of collations” database errors on sites whose plugin tables do not all share the same collation. This affected 404 log cleanup, the admin list counts, the email digest, and the dead-destination check.
• Fixed the plugin running a schema-wide collation repair in the middle of an admin page load, which could make that page slow or time out. The repair now happens outside the page request.
• Fixed slow sorting on the Captured 404s tab. The timestamp sort had no supporting database index, so a large site re-sorted the whole table on every page load.
• Fixed the plugin trusting one of its own sort indexes by name alone. A database server silently narrows an index when a column it referenced is dropped, and an index narrowed that way was treated as correct forever, so the sorts it was built for scanned the entire table instead of reading a single page of rows. Indexes are now compared against the definition the plugin ships and rebuilt when they differ.
• Fixed the plugin rebuilding a database index on the strength of index information it could not read. When a database server left out or garbled the field that says whether an index is unique, the plugin decided the index differed from the one it ships and repaired it, which on the page-suggestion cache meant emptying that table and rewriting the index. Index information that cannot be read is now treated as unknown, and an index the plugin cannot describe is left alone.
• Fixed the Page Redirects and Captured 404s tabs waiting on their own row counts on large sites. Pagination and the per-status counts now load in separate stages, and a count that is still being computed is shown as incomplete instead of holding up the table.
• Fixed a regex redirect whose source pattern starts with an anchor (for example ^/products/(.*)) being rewritten on save into a pattern that could never match.
• Fixed the Add Redirect dialog rejecting valid regex destinations, and checking them against a stale copy of the validation rules after an update.
• Fixed the “Hits” counts on the admin tabs falling behind. The roll-up that keeps them current was no longer running from a page view, and the counts could also sit empty on sites whose WordPress cron is backed up. Both paths are restored.
• Fixed 404 URLs containing @ (retina images such as logo@2x.png) and PHP static-call frames in stack traces being mangled by the redaction that runs before a debug log or automatic report is written.
• Fixed page suggestions staying poor for months after an interrupted rebuild. A partially built suggestion cache recovered at about 50 entries a day rather than rebuilding; it is now completed at full rebuild speed.
• Fixed a multisite network where deleting a site part way through the page-suggestion rebuild could leave a later site with an empty suggestion cache that was marked fully built and never rebuilt. The rebuild now tracks the last site it finished instead of counting positions in a list that shifts whenever a site is removed.
• Fixed transient database failover and connection-state errors, including Galera nodes that are temporarily unavailable and “Commands out of sync,” being treated as permanent failures. Affected queries now recover and retry through the shared database path.
• Fixed an admin table database failure being mistaken for a successful empty result, which could make the table appear empty instead of showing the real error.
• Fixed the 404 and redirect trend charts reading all zeros on sites whose database server is not set to UTC, which is common on shared hosting. Each database row was matched to a calendar day by comparing two dates produced in different time zones, so on those sites no row ever matched a day and the chart drew a flat line with no error shown. The Search Console date range was built the same way and is fixed too.
• Fixed the “Post types” and “Categories” settings fields being able to alter the database queries built from them. Values from those two fields were placed into queries without being escaped, so a value containing a quote could change what the query did. Both fields are now escaped, and text carrying invalid characters is repaired before it reaches the database.
• Fixed automatic redirects being able to save a destination that cannot be reached. The destination was checked as a positive number but stored exactly as entered, so a value such as -3 was checked against page 3 and then stored as -3. Automatic redirects are created with nobody watching, so nothing later would have questioned it.
• Fixed 404 URLs losing their hit counts and their “View Logs” row action when their log rows had no canonical URL recorded. Those rows were dropped from the hit-count roll-up entirely, with nothing to say so.
• Fixed the “Hits” columns going stale, or being rebuilt on every check along with a repeating admin notice, on sites whose database server is not set to UTC. The age of the hit-count roll-up was measured by comparing a time the database rendered in its own time zone against a time measured in UTC, so it was wrong by that server’s offset, in one direction or the other depending on which side of UTC the server sits.

Improvements

• The hit-count roll-up no longer runs at the very end of an admin request, where it added as much as 21 seconds to the page on a 900,000-row test site.
• On LiteSpeed servers the admin table response is handed back as soon as it is ready, instead of waiting for the rest of the request to finish.
• Creating, editing or deleting a redirect no longer scans the whole WordPress options table. The plugin was still clearing a results cache it stopped using in 4.3.1, and the pattern it searched for could not use an index, so on a site with a large options table every redirect change read the entire table to delete nothing.
• When an admin action fails because the server or its gateway rejected the request, the message now names the underlying failure (for example “502 Bad Gateway”) instead of a generic “An error occurred”, and the full detail is written to the browser console. Saving settings, restoring defaults, switching settings mode and trashing a redirect each had their own copy of this handling, and two of them discarded the failure entirely.
• Automatic error reports and support previews now include log evidence for the error they describe, even when the error falls outside the most recent log tail. Admin table retries also preserve a diagnostic trail under the plugin’s shipped default settings.

Version 4.3.2 (July 11, 2026)

• FIX: Fixed the email digest sending every day even when the notification frequency was set to Weekly (thanks to gardendarts for reporting this). The weekly/daily cadence is now enforced against the time the last digest was actually sent, and changing the frequency in Settings now takes effect on the very next save instead of the next unrelated one.
• FIX: Fixed automatic error and heartbeat reports silently failing to send on some sites when an internal reporting component was unavailable.
• FIX: Fixed debug logs and automatic reports sometimes over-redacting the plugin’s own file names, option names, and URLs, including incorrectly truncating URLs that contained a query string.
• FIX: Fixed a rare issue where saving, publishing, or deleting content, or changing a category, during a plugin update could interrupt a scheduled background maintenance task.
• Improvement: Hardened several internal WordPress integration points (404 handling, REST API setup, log maintenance, and permission checks) so an unexpected internal error in one can no longer interrupt that request or affect other plugins.
• Improvement: Out-of-memory diagnostic reports now preserve the exact memory sizes involved, making them easier to diagnose.
• Improvement: Removed an internal AJAX diagnostics footer that no longer reflected real admin activity.
• Improvement: If you’ve opted in to “Help the developer with error logs and usage stats,” the optional weekly status ping now sends on a predictable weekly cadence (once the plugin’s been active 5+ days) instead of a rare random daily chance that averaged about once every 6 months. Same opt-in checkbox and same aggregate-only data (site version, URL, 404/redirect counts); just a steadier heartbeat.
• New: Added a “Never” option to the Email notification frequency dropdown, so you can fully turn off 404 email notifications without switching to Instant mode and separately zeroing the threshold.
• New: Added a “Your Diagnostic Data” card to the Options tab so you can download or delete the diagnostic data this plugin has reported for your site, in line with GDPR data access and erasure rights.

Version 4.3.1 (June 26, 2026)

• FIX: Fixed an out-of-memory error (thanks to johnegg and kalshyre for reporting this).
• FIX: Fixed other things that looked like possible out-of-memory errors during maintenance.
• Improvement: Made automatic error reporting more robust to cover OOM as best as possible as well as other failure types.

Version 4.3.0 (June 12, 2026)

Bug Fixes

• Fixed locale-stale labels in the Page Redirects view cache by storing status and type as integer codes only, translating labels at render time, and dropping the legacy translated-label columns during upgrade.
• Fixed the Page Redirects and Captured 404s admin tabs still occasionally failing to load on high-traffic sites running 4.2.0. The 4.2.0 release notes claimed this was fixed, but only the cache-warmup race was addressed. A separate internal “mutation watermark” gate could still abort the cache rebuild mid-stage when 404 traffic arrived during the rebuild, and surface the same “Could not finish refreshing data” message at the 45-second budget. That gate and its supporting machinery (watermark counter table, admin-mutation gate, view-build watermark stamping) have now been removed in full. On Bruno-class data (about 30,000 active redirects, about 384,000 hit rows, sustained 1 request per second or higher 404 traffic), the admin tabs hydrate within budget across consecutive refreshes.
• Fixed a stale “No ID(s) found” notice appearing after a redirect was edited and saved successfully. The notice from the prior screen is now cleared so the success message stands alone.
• Fixed a stuck “Refreshing data… (stage 4)” badge that could remain on the Page Redirects pagination strip after a background refresh failed. The strip now collapses placeholder-only rows on a failed table load, and the error notice no longer competes with the 5-star review prompt for attention.
• Fixed generic “Something went wrong” messages on admin AJAX errors. The underlying cause is now appended in parentheses so the error is self-diagnosable, and the multi-stage diagnostic detail is tucked behind a collapsed Details panel.
• Fixed wp abj404 list not returning redirects that were just created via wp abj404 create in the same session, and corrected the --status=manual filter so manually-created redirects are returned.
• Fixed CSV-imported redirects not appearing in the Page Redirects table immediately after import. The cached view is now rebuilt synchronously when the import completes, so the table reflects the new rows on the next page load.
• Fixed a “table doesn’t exist” error path that could fire during a partial install or upgrade. The data layer now tolerates an unprepared $wpdb and surfaces a clean degraded state instead of a fatal.
• Fixed the plugin’s settings defaults not being available on the uninstall path, which could leave a stray option behind on some sites.
• Fixed a duplicate-accessible-name accessibility warning on the redirects list. The “Add Redirect” submit button inside the modal is now labeled “Add”, which removes the collision with the page-level “Add Redirect” action button.
• Fixed a fatal that could occur on the Plugins admin screen when wp-admin/includes/plugin-install.php was unavailable on hardened hosts (the include is now guarded with is_readable).
• Fixed a fail-open default in the REST captured-404s status filter that could return rows from an unintended status when an unknown filter value was supplied.
• Fixed the URL and Destination column-sort “being prepared” progress indicator that could stay stuck at 0% on large sites where a security plugin blocks the WordPress cron loopback. The one-time sort-key preparation is now driven by the admin’s own page view in the background, with a longer and gentler retry window, so column sorting becomes available without depending on WP-cron, and a transient timeout no longer halts the progress.

Improvements

• Several admin surfaces were restyled to match native WordPress chrome and feel quieter on the eye: the top sub-navigation now uses standard .nav-tab styling; the Edit Redirect form follows the native form-table shape; list-table filter rows use subsubsub pipe-separated links; status badges and pills use the bordered-label shape; checkbox columns and column-header padding match core WP list tables; AJAX error notices use the native .notice-error shape; and the self-healing dashboard notice routes through the standard notice template.
• Several internal queries that scanned the full information_schema, post-type table, or content-keywords index on large sites are now bounded, removing a needless cost during background cache rebuilds. No behavior change for typical sites; smoother performance on sites with thousands of post types or very large content tables.
• Changed the internal admin-notice payload schema to store translated message and guidance text directly, removing the message catalog indirection. No user-facing behavior change is expected.
• When a background refresh fails, a Support button is now mounted directly on the failure notice so the report path is one click instead of a hunt through Settings.

Version 4.2.0 (May 23, 2026)

Bug Fixes

• Fixed the “Page Redirects” and “Captured 404s” admin tabs occasionally failing to load on busy sites. The cache-warmup pipeline could race against ongoing 404 traffic and report that the data snapshot was missing even when the underlying query had succeeded. Admin tables on high-volume sites now load reliably.

Improvements

• Substantial internal refactor of the data-access, view-build, and admin-mutation layers for long-term maintainability. No visible behavior change is expected; the version bump from 4.1.x to 4.2.0 reflects the size of the change.

Version 4.1.19 (May 17, 2026)

New Features

• Regex redirects are now auto-detected. When you save or import a redirect that contains regex characters (wildcards, brackets, pipes), the plugin automatically marks it as a regex redirect and applies the correct syntax. No need to manually check the “Treat as regex” box. An admin notice confirms the auto-promotion with Edit and Undo links.
• Debug logs now automatically redact sensitive server details (database name, table prefix, filesystem paths) so logs can be shared safely without manual editing.

Bug Fixes

• Fixed the Confidence dropdown filter on admin tables only working on the initial page load. Changing the filter and then navigating pages or waiting for the background refresh would reset it to “All”, showing every row.
• Fixed invalid regex patterns in CSV imports crashing the import. Invalid patterns are now validated during import and suppressed at runtime instead of causing errors.
• Fixed the repair-and-retry path continuing to report an error after the repair succeeded. The error is now cleared immediately after a successful repair.
• Fixed a potential SQL ambiguity error in log queries on sites with certain JOIN configurations.
• Fixed all 6 scheduled tasks not being removed when the plugin is deactivated. Previously some cron hooks were left behind and would trigger errors until manually cleared.
• Fixed the “Undo” link on regex auto-promotion notices not reverting the URL on the next page load. The original URL is now correctly restored along with the Manual status.
• Fixed a rare race on shared hosting where a dropped database connection during the admin table rebuild could let a second worker drop the rebuild’s working table out from under the first, producing “Table doesn’t exist” or “Can’t find .frm file” errors. The rebuild now releases its database lock between each step so a dropped connection loses at most one step of progress instead of the entire 11-step run.

Improvements

• Bulk CSV and WP-CLI imports are significantly faster. A 10,000-row import that previously issued ~60,000 extra cache-invalidation queries now batches them into a single invalidation at the end.
• The admin settings page now validates URL length and rejects negative numeric values, preventing misconfigured redirects.

Internationalization

• Added translations for 6 new strings in all 12 language files.

Version 4.1.18 (May 13, 2026)

Bug Fixes

• Fixed the admin table cache rebuild getting stuck on the same step after a dropped database connection. The rebuild now auto-resumes from where it stopped on the next request, instead of retrying the same failing query forever.
• Fixed admin actions that hit an expired session nonce showing a generic “security check failed” error. The session is now silently refreshed and the action retried, so the expiry is invisible to the administrator.
• Fixed transient invalid AJAX responses on admin tables producing “undefined” errors. The admin pages now validate the response shape before reading it and surface a clear error instead.
• Fixed the “missing database table” admin notice being cleared by the next successful query in the same request, so the admin never saw it. The generic auto-clear now skips the missing-table notice; that notice is cleared only by the dedicated repair-success path.

Improvements

• Captured 404s and Page Redirects admin tables load noticeably faster on large sites. The total log-count query is now cached on the maximum log id (avoiding a full log-table scan on every page load), log queries no longer sort by non-indexed columns, the Most Unused Redirects report reads from the pre-aggregated hits rollup instead of scanning raw logs, and the dashboard activity trend is cached between admin page loads.
• When WordPress cron is broken and the log hits rollup falls behind, the plugin now surfaces an admin notice on its own pages explaining how to fix it, instead of silently letting the admin tables show stale numbers.

Internationalization

• Added translations for 13 admin strings that were previously displayed in English even on non-English sites.