5.0.0 

Breaking 

  • Remove support for TYPO3 v12.

  • The list item moved from a partial to a template: Resources/Private/Partials/TouristAttraction/ListItem.html is now Resources/Private/Templates/TouristAttraction/ListItem.html. The name is unchanged; only the directory differs.

    Integrators overriding the list item must move their override accordingly. An override left in Partials/ is silently ignored — the file is no longer read from there, and the shipped template renders instead. Nothing errors, so check this explicitly when upgrading.

    The reason is caching: the attraction list now renders each item separately and caches the resulting HTML per attraction, so an item is rendered once and reused by every list showing it. A separately rendered item is a template rather than a partial.

    This also changes what the list templates receive. List.html and SelectedList.html no longer iterate attractions — the controller renders the items and assigns them as ready HTML in {items}:

    <f:for each="{items}" as="item">
    <f:format.raw>{item}</f:format.raw>
</f:for>
    Copied!

    An override still iterating {list.paginatedItems} or {attractions} renders nothing, without an error. {list} remains assigned for pagination; {attractions} is gone from the curated list entirely.

    Both actions share one cache, so an attraction shown by a filtered list and a curated one is rendered once.

    Template paths are unchanged otherwise: the item is resolved through the plugin's configured templateRootPaths, so an override in an extension or sitepackage keeps working once it sits in Templates/.

Features 

  • Attraction lists are cached, so a visitor rarely waits for one to be built twice.

    Three caches do the work, all in the pages cache group:

    tx_thuecat_teaser
    One rendered list item, keyed by the record, the detail page it links to and the language. An attraction shown by several lists — filtered, curated, on different pages — is rendered once and reused by all of them.
    tx_thuecat_list
    A whole rendered list, keyed by the plugin, the filter selection, the pagination page and the language.
    tx_thuecat_searchmask
    A rendered filter form, keyed by the plugin, its page, the selection and the language. Deliberately not by the pagination page, so paging through results reuses one form.

    Because they belong to the pages group, everything that already clears page caches clears these too: the "Clear cache" buttons in the backend, and any editor saving a record. Editors need do nothing differently — a changed attraction, town or category takes its cached output with it and the next visitor sees the new content.

    Entries are discarded by the records they show rather than by age, so nothing goes stale waiting for a lifetime to run out. A filter combination nobody has requested before is still built from scratch, but only its unseen items cost anything: the rest come from the teaser cache.

    The three caches therefore configure a lifetime of one year. This is not a staleness setting — invalidation by tag is what keeps content current — it exists because TYPO3's default would otherwise expire entries after an hour. Override it per cache in ext_localconf.php or config/system/additional.php via $GLOBALS['TYPO3_CONF_VARS']['SYS']['caching']['cacheConfigurations'][<identifier>]['options']['defaultLifetime'], though shortening it only costs rebuilds without making anything fresher.

    Note that an import currently discards the caches of every record it visits, whether or not that record changed, so lists are rebuilt after each import run. A known limitation, to be addressed separately; it costs speed, never correctness.

  • Add support for TYPO3 v14.
  • Handle images and other files properly via FAL upon import. Target file directory is now part of the import configuration.
  • Events are imported with their images. An event's schema:photo and schema:image both land in the images field of tx_events_domain_model_event ; events have no separate main image, so the source ranking decides the order — the photo comes first. The previous cap of eight images per event is removed: every image upstream publishes is imported.

  • Media uploaded directly to a record is imported. Alongside the established form, where a record points at a reusable media resource, upstream also lets editors upload a file into the record itself. Such a file was previously rejected before any download was attempted, and — because the rejection was fatal — it cost the whole root URL every one of its records. It is now imported like any other image, with its author and licence.

    This applies to every record type, not only events. A point of interest carrying a direct upload imports it into main_image / media_files as it would a referenced image.

    A directly uploaded file is identified by its file name in the delivered address, which stays stable across runs, so re-importing reuses the stored file instead of downloading it again.

    These files are served from the API host, which refuses anonymous requests, so the configured API key is sent when an image is fetched from that host. Images on other hosts — every referenced media resource so far — are requested unchanged, without a key.

  • Import opening hours as inline database records ( tx_thuecat_opening_hours ) and expose a display-ready, computed shape via Domain\Model\Frontend\Place::getComputedOpeningHours() and getComputedSpecialOpeningHours() . The shape groups records into validity periods, lists every weekday Monday-first, keeps all time spans of a day and marks days without hours as closed. A OpeningHours/PerDayTable partial renders it. The current open/closed status is intentionally left to client-side logic.

  • An event imported without a single date is named in the import log. Such an event is stored but cannot appear anywhere dates are shown, and until now nothing said so — the absence had to be noticed in the frontend and traced back by hand. Each one now gets a tx_thuecat_import_log_entry row of the new type eventWithoutDates at severity warning, naming the event and the record to look at.

    The reason is deliberately not distinguished: an event publishing no schedule, one whose schedule yields no usable day, and one whose dates are all excepted are reported alike, because the outcome is the same. The event still imports and a run whose only findings are dateless events exits 0.

    Events whose dates have all passed are included, so a finished event will produce this warning. That is intended — the log entry says "this event shows nothing", which is true of a past event as well — but it means the type is expected to appear routinely on sites with older data.

Fixes 

  • Paging through an attraction list no longer shows the wrong page.

    Every pagination page and every filter combination of one list plugin shared a single page-cache entry, because currentPage and demand were excluded from the cache identifier. Whichever page was requested first was then served for all of them, until that entry expired.

    The symptom only appeared once a page had been cached, so it was reliably invisible on a developer machine and in tests, where every request starts from a cold cache.

    The list and search actions are now non-cacheable and serve their output from the caches described under Features, each keyed by the plugin, the filter selection, the pagination page and the language. Filter and pagination URLs are unchanged.

  • A referenced resource that cannot be fetched no longer aborts the whole root URL. Previously any failure while resolving a reference (a resource deleted upstream, a server error, a malformed response, a network failure) propagated out of the resolver and the importer discarded every record belonging to that root. Now only the affected reference is dropped: the owning record is imported without that relation, and all other records under the same root are unaffected.

    Each affected parent record gets its own tx_thuecat_import_log_entry row of the new type referenceSkipped at severity warning, carrying the owner's remote_id and table_name alongside the reference URL, the target field and the failure reason — so the log names the records to fix, not merely the URL that failed. A URL already found to be failing is not fetched again for the rest of the run, but every parent that loses a reference is still logged.

    Because these runs report warning rather than error, thuecat:importviaconfiguration now exits 0 for them. Failures of the outer calls — the sync-scope/contains-place listing and the root URL fetches — are unchanged and still fail the run, so a genuine upstream outage remains distinguishable.

  • A record no longer inherits relations from the record imported before it.

    Records that said nothing about a relation kept whatever the previous record in the same run had declared. The effect was visible with images — several events ending up with the same picture, none of which named it — but it applied to every imported relation: town, responsible organisation, nearby parking, categories. A record silent about a relation is now imported without it.

    Relations already stored from earlier runs are not corrected by this. The import stops writing them; removing one whose source is gone is a separate matter, and images that were wrongly related need to be removed by hand.

  • A media entry the import cannot interpret no longer costs the whole root URL. Upstream may publish an entry that is neither a reference to a media resource nor a direct upload. Such an entry previously aborted the import of every record under that root. It is now skipped like an image that cannot be downloaded: the record is imported, its usable images are related, and a tx_thuecat_import_log_entry of type referenceSkipped at severity warning names the record and the entry. A run whose only problem is an entry of this kind exits 0.

  • An image that cannot be downloaded no longer costs the run its media. Previously the download raised on any error status, the exception escaped the resolver and the importer discarded the whole root URL; because promotion of downloaded files was gated on the run's overall severity, that single failure also deleted every image already downloaded under all other roots. A run could finish having imported thousands of records and not one image.

    Now only the affected image is dropped: the owning record is imported without that file, its other images are related as usual, and records under the same root are unaffected. This holds for every failure mode — an error status, a refused connection, a timeout, or an empty body. Each affected record gets a tx_thuecat_import_log_entry row of type referenceSkipped at severity warning, naming the record and the URL that could not be fetched, so runs whose only problems are unfetchable images exit 0.

    The image download now sends through the PSR-18 client rather than RequestFactory::request() , whose Guzzle default raises on a 4xx/5xx before the response can be inspected. JSON-LD fetches are unchanged and still fail loudly.

  • Staged media is now promoted into the target folder even when the run recorded an error elsewhere. A file in staging downloaded successfully, so an unrelated mapping failure says nothing about it. This is observable: a run ending in error now leaves promoted files where it previously left none. A promoted file that no record ended up referencing is reused by name on the next run rather than downloaded again. The per-run staging folder is still discarded on every exit path, so a crashed run leaves no orphans.
  • thuecat:importviaconfiguration now prints a closing message stating whether the run completed, completed with warnings, or failed. It previously exited silently, so a non-zero exit arrived with no explanation.

  • A schedule day that cannot seed a date series no longer aborts the event import.

    schema:byDay may carry values that are valid upstream but name no weekday — schema:PublicHolidays among them. These were passed on for date expansion, where they raised and cost the whole root URL its records. Such a value is now skipped: the event imports with the dates its remaining weekdays produce, and a tx_thuecat_import_log_entry of the new type scheduleDaySkipped at severity warning names the event and the value. Editors keep the information that upstream asked for something the import cannot express.

    Where a schedule supplies more usable weekdays than the import can carry — a monthly schedule holds one — the surplus days are recorded separately as scheduleDayDropped, also at warning. The two are deliberately distinct: one is an upstream value we cannot use, the other a usable value our own import discards, and they call for different editor action.

  • schema:exceptDate is honoured.

    Dates a schedule explicitly excludes were imported like any other occurrence. They are now dropped, for every frequency. The comparison is by calendar day in the schedule's timezone, since an exception is date-only while occurrences carry a time.

  • An event's stored dates are reconciled with its schedule on re-import.

    Only new occurrences were ever written; one that stopped being produced stayed in the database indefinitely. Dates the import owns are now compared against what the current schedule yields, and the difference is deleted. This is the observable change of this release: occurrences that became excepted, fell after a shortened series end, or vanished with a changed schedule disappear on the next run, where they previously lingered.

    Only the import's own rows are affected — ownership is established through the remote_id the import assigns — so dates created in the backend survive. The reconciliation is scoped to the events actually imported by the run: an event outside the run, or one whose root URL was abandoned, keeps its dates.

  • A failed root URL is identifiable from the import log.

    mappingError and fetchingError entries recorded the exception's message and origin but not which configured URL produced them, so a run with several roots left the failure unattributable. Both now carry the root URL in remote_id and repeat it in the entry's context alongside the existing detail. Severity is unchanged: an abandoned root remains an error and still fails the run.

    The failing record itself is not named. It is not reliably known at the point the failure is caught — parsing may raise before any record is identified — and naming the wrong record would mislead more than naming none.

Tasks 

Nothing

Deprecation 

  • The json-based media storage has been superseeded by proper FAL handling. The json-based output logic is still in

place. The next import will copy and relate the files as needed. With the next fitting version, the json-based fields will be removed.

Deprecated, triggering E_USER_DEPRECATED on use:

  • Domain\Model\Frontend\Media::getMainImage() , getImages() , getExtraImages() , getEditorialImages() and getAllImages() (the json-blob read accessors).
  • Domain\Model\Frontend\Base::getMedia() (hands out the json-blob carrier).

Use the FAL fields instead, available as native Extbase properties on Base: getMainImage() ( main_image ), getMediaFiles() ( media_files ) and getEditorialImages() ( editorial_images ). Re-run the import to populate main_image and media_files ; editorial_images is maintained in the backend. Templates should switch from {record.media.*} to these properties.

  • The editorial_images field is now handled as the FAL field it always has been. Replace template output from {entity.media.editorialImages} to {entity.editorialImages} to make use of the full FAL provided functionality.

  • The json-based opening hours storage has been superseeded by inline database records. The json-based output logic is still in place. The next import will create the inline records as needed. With the next fitting version, the json-based fields will be removed.

    Deprecated, triggering E_USER_DEPRECATED on use:

    • Domain\Model\Frontend\Place::getOpeningHours() , getMergedOpeningHours() , getSpecialOpeningHours() and getMergedSpecialOpeningHours() (the json-blob read accessors).

    The legacy \Domain\Model\Frontend\OpeningHours , MergedOpeningHours and MergedOpeningHourWeekDay models are deprecated alongside them.

    Use getComputedOpeningHours() / getComputedSpecialOpeningHours() instead. Re-run the import to populate the inline records. Templates should switch from {record.openingHours} / {record.mergedOpeningHours} to the OpeningHours/PerDayTable partial.