Repository navigation
Conversation
* Add missing Smokeping validation group title
The validate page builds each group's title from
validation.validations.groups.{group}, but the smokeping group had no
entry, so the raw translation key was shown as its heading.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Correct spelling of 'Smokeping' to 'SmokePing'
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Jellyfrog <Jellyfrog@users.noreply.github.com>
not strings
…e` (librenms#21019) * Fix Oxidized backup dates when the API returns `time` instead of `date` Oxidized returns the revision timestamp in the `date` field or, with some backends (such as git), in the `time` field instead. mapVersions() only looked at `date`, so those revisions were listed with a blank date in the Config tab. Accept either field, preferring `date` when present and parseable, and add unit tests for the `time`-only format and the precedence. Fixes librenms#21018 * Simplify Oxidized date parsing per review Inline the date/time fallback in mapVersions() instead of the versionDate() helper, and apply the same fallback in latest() where the node timestamp may also arrive as time instead of mtime. Adds a test for the latest() fallback. * Revert the speculative node time fallback in latest() You are right, $node[time] does not exist: Oxidized serializes nodes with a top-level mtime, and time only appears inside the last job summary as a duration, not a timestamp. The fallback and its test are removed; the versions date/time fix is unaffected.
* Modernize VRF module merge vrf lite into vrf WIP * Proper module design * Handle vrf failures a little more gracefully * Better snmp error pattern * Update schema * BGP Peers in panel * Timos port mapping fixes * build assets * Refine settings page and how manual vrf contexts are stored * Refine settings page and how manual vrf contexts are stored * Remove unuseful help * Bug fixes * Cleanup
) * fix(discovery): avoid ambiguous LLDP neighbor IP matches Resolve management addresses against device and interface ownership together before accepting a unique neighbor. Keep the existing identifier priority and let ambiguous IP addresses fall through to MAC and sysName matching. Add dual-stack NAT, ambiguity, priority, cache, and persisted-link regressions. Developed with assistance from OpenAI Codex. No physical-device testing. * fix(discovery): constrain ambiguous neighbor resolution Retain the direct management fallback for MAC-less CDP/FDP, exclude deleted ports, and disambiguate LLDP only within the IP owner set. Cache address candidates and skip autodiscovery of an already ambiguous management IP. Add IPv4/IPv6 regressions for protocol/cache isolation, stale ports, candidate matching and autodiscovery. * test(discovery): document neighbor provider value types * perf(discovery): index IPv4 and IPv6 address lookup for neighbor matching * fix(discovery): record neighbor lookup indexes in schema Synchronize the expected database schema with the IPv4 and IPv6 lookup migration so DBSetupTest validates the migrated tables. Validated migration up/down/up and all 193 focused neighbor/discovery tests. AI-assisted development: OpenAI Codex. Signed-off-by: dongxuelian2 <2984896280@qq.com> * fix(discovery): remove address indexes from neighbor fix Keep ambiguity detection independent of an unproven schema optimization. Discovery caches address owners within each run. Other consumers, including OSPF polling and routing pages, could benefit from these indexes but are unchanged by this fix and have no production bottleneck evidence here. Remove the index migration and restore the expected schema. The existing 30,001-address-per-family fixture takes 5.6-6.5 ms per uncached owner lookup without the indexes; this is a local scale check, not production profiling. Validation: 193 focused discovery/address/MAC tests, 531 assertions; complete schema comparison with Windows table-name case normalization passes. OpenAI Codex assisted with this change. Signed-off-by: dongxuelian2 <2984896280@qq.com> * fix(discovery): disambiguate neighbors named by a shared IP Treat an IP-literal sysName as repeated address evidence only when it equals the management IP after normalization and that address has multiple owners. Skip both hostname and sysName matching for that case, allowing scoped MAC resolution or the existing MAC-less CDP/FDP fallback to decide. Preserve ordinary hostname priority, different literal-name addresses and unique IP ownership. Cover both MAC identifiers, IPv6 normalization, cache observation order, link endpoints and suppression of ambiguous autodiscovery. Validation: the final targeted regression selection has 8 failures on 13ff818 and passes with this fix. All 204 focused discovery/address/MAC tests pass (564 assertions), as do Pint, Rector, PHPStan, syntax and diff checks. OpenAI Codex assisted with implementation, tests and review. Signed-off-by: dongxuelian2 <2984896280@qq.com> --------- Signed-off-by: dongxuelian2 <2984896280@qq.com>
* Filter empty port ids in graph widget settings Widgets saved with an empty entry in graph_ports produce links like graphs/id=,954, which GraphsPageRequest rejects, redirecting back. Filter them like graph_sensors already does. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Filter empty graph_custom entries, drop redundant sensor filter Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Antonio J. Pérez <3948956+Psycomy@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Split snmp secret into two separate secrets Version did not belong in the secret * style: auto-format code Co-authored-by: murrant <39462+murrant@users.noreply.github.com> * fix v1 probes and other fixes * Cleanup some more bugs and issues * Clean up * every snmp polling method should have a secret * Migration fix --------- Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com> Co-authored-by: murrant <39462+murrant@users.noreply.github.com>
* Handle unknown polling types and secrets without crashing * Handle unknown polling methods and secrets cleanly
* Add AOS6 FDB, transceiver and NAC discovery * Refactor AOS6 FDB port mapping * Fix AOS6 FDB coding style * Refactor AOS6 NAC and transceiver discovery * Fix AOS6 coding style * Refactor AOS6 FDB discovery * Refactor AOS6 FDB discovery * style * Fix AOS6 FDB coding style * Use transceiver presence for AOS6 DDM sensor skip_values Per review: skip DDM sensors when no module is present (ddmSupplyVoltage = 0) and skip optical power / bias when Tx power is not applicable (copper SFPs), instead of skipping 0 readings, which would delete a sensor when a laser fails or Rx is weak. Claude-Session: https://claude.ai/code/session_01S8peTygx8Znj6RX6JrxaMQ * Code review fixes * Apply review fixes * Apply review fixes * Regenerate AOS6 test data after rebase * Fix AOS6 NAC timestamps and transceiver mapping Use observed epoch timestamps or SNMP agent uptime for NAC session age. Prefer vendor slot and port mappings for transceiver identity, and exclude expansion-card metadata in the legacy fallback. Recapture and regenerate the existing AOS6 fixture pair from a consistent real device dataset, including transceivers, port stacks, FDB and NAC coverage. * Fix AOS6 transceiver graph sensor association Keep the transceiver entity_physical_index aligned with AOS6 DDM sensors, which use port ifIndex with entPhysicalIndex_measured=ports. Preserve vendor slot/port metadata mapping and regenerate the existing fixture. --------- Co-authored-by: Jellyfrog <Jellyfrog@users.noreply.github.com> Co-authored-by: librenms-bot <librenms-bot@users.noreply.github.com>
* Fix missing SmokePing settings translations
Settings translations are resolved as settings.settings.<key>.description,
which Laravel walks segment by segment, so flat 'smokeping.<key>' entries
were never found and the SmokePing settings showed raw translation keys.
Nest them under 'smokeping' in every locale, and use the SmokePing spelling
in the English strings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Fix more settings translations stored under unreachable keys
Same lookup issue as the SmokePing settings: the settings UI resolves
settings.settings.<key>.description segment by segment, so these
translations were never found and their settings showed raw keys.
- ipmi.type, polling.selected_ports, processor.default_perc_warn,
webui.global_search.*: nest the flat dotted keys
- front_page_settings.top.{devices,ports}: were top_devices / top_ports
- prometheus.job: was 'Job'
- graph_stat_percentile_disable: is a top-level setting, was under webui
Applied to every locale that had them; translated text is unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Add missing English translations for settings
custom_map.edge_font_align, good_if, ignore_mount_type, networks_purge,
prometheus.user, prometheus.password and rewrite_if_regexp had no
translation, so the settings UI showed their raw keys.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Add missing English translations for the drive ignore settings
ignore_bsd_os_drives and ignore_junos_os_drives had no translation, so
the storage settings page showed their raw keys.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…ibrenms#21021) * Fix Browser Push user dropdown hiding users with duplicate realnames Browserpush::configTemplate() used the user's realname as the select option key. realname is not unique, so users sharing a display name overwrote each other and only one of them could be selected. Append the username to the label only when the display label would collide, so unique realnames keep their existing label. Add a feature test covering duplicate realnames, a unique realname, and a user without a realname. Fixes librenms#21014 * Build Browser Push user options with Collection methods Replace the foreach loop in configTemplate() with countBy() and mapWithKeys(). Behaviour is unchanged: labels are realname (or username when realname is unset), duplicates get the username appended, and the stored value remains the user_id. * remove useless comment * Keep numeric username labels in Browser Push options * Add numeric username label test for Browser Push options --------- Co-authored-by: Tony Murray <murraytony@gmail.com>
* BGP Peers WIP * Fixes * Adapt new vrf module * refactor * Fixes
* Move get_nagios_state() into the SensorState enum * Also move the backward mapping to the Enum * Changed tests * Also trim the descr passed to StateTranslation::define() The severity lookup used the trimmed descr, but define() still received the raw capture, so state_descr kept a trailing space (e.g. 'Primary ') which breaks the state translation lookup and the test data. * Remove stray dnos_stack.snmprec Accidentally committed local snmprec capture, unrelated to this PR.
…umber (librenms#21060) A rectifier that has been removed keeps its serial number in rectifier-Values-Table while rectifier-Registration-State is 0 (not-Detected), so discovery created five sensors for the empty slot, all reading 0. Skip on the registration state instead; rectifiers in any other state (including comms-Lost) keep their sensors. Test data: eaton-sc200_removed-rectifier, captured from a controller with one removed and one registered rectifier.
Handle FORTINET-FORTIGATE-MIB::fgTrapDhcp, which FortiOS sends when a DHCP server runs out of addresses, detects an IP conflict, or a DHCP client interface receives a NAK. fgDhcpTrapType is mapped to an event and severity (pool exhausted: error, IP conflict: warning, NAK: notice) and logged with the interface name and fgDhcpTrapMessage. snmptrapd may deliver the enum as a label, label(n), or bare number, so all three forms are accepted. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(polling): preserve precision of TiMOS BGP Counter64 values Reduce TiMOS octet counters modulo 2^32 directly from their decimal strings using bounded 16-bit halves before converting the result. Add poller regression coverage for boundary values, Counter64 maxima, leading zeros, and the counters reported in issue librenms#20868. Fixes librenms#20868. * fix(polling): validate and isolate TiMOS counter conversion Extract exact low-32-bit arithmetic to Number, handle missing counters as zero, and reject malformed decimal encodings without PHP diagnostics. Use exact int-or-float reconstruction for 32-bit PHP compatibility. Add database-free numeric cases and poller regressions for missing OIDs, invalid values and guaranteed cache cleanup on exceptions. * fix(polling): preserve null VRF lookup on PHP 8.5 Keep the existing empty-key fallback explicit to avoid PHP 8.5's null array offset deprecation when a TiMOS peer has no stored VRF OID. Retain strict warning checks in the poller tests and document the pure counter provider types required by level-6 analysis.
* Allow multiple polling methods per module Module decides how to combine/handle the data. Maybe only one gets even polled, maybe all get polled and the data is merged. mempools with snmp and unix agent polling methods. Unix Agent fixture based testing. * Fix tests * Remove connectivity helper * Update new modules * Fixes and cleanups * update schema * Simplify and improve design * reorganize * Introduce trait for shouldDiscover/shouldPoll default implementation * cleanup default usages * lint
- Count sensors per PON port from GW-EPON-MIB ponPortCurrOnu (registered ONUs) and ponPortLinkedOnuCounter (online ONUs) - Processors: skip empty board slots (255) and ONU rows of boardCpuUsage, name each CPU by board - Mempools: boardMemoryUsage is a percentage, use percent_used - State sensors follow GW-EPON-DEV-MIB enums, fix fan descr typo - Add GFA6900S test data
…ibrenms#21065) * BGP peer table: find the neighbour of peers identified by interface Since librenms#21054 a session whose remote address is unknown or link-local is identified by its interface name, e.g. swp25 for a Cumulus unnumbered peer, and keeps the learned address in bgpPeerRemoteAddr. The routing tab only looked the neighbour port up by bgpPeerIdentifier, so those peers lost their neighbour link. Fall back to bgpPeerRemoteAddr when the identifier is not an address. Split out of librenms#20187, the rest of which librenms#21054 made redundant. * BGP peer table: only link the neighbour when one port carries the address Link-local addresses (and others) are not guaranteed to be unique, so skip the link when more than one port matches rather than picking one at random. * tidy it up --------- Co-authored-by: Tony Murray <murraytony@gmail.com>
…ibrenms#21068) - Detect switch type from the connUnitRevs "Hardware version" row, fall back to fcFeModuleObjectID, then to the yaml entPhysicalDescr - Add newer switch types (X6/X7/X8, G630, G710-G820, 7810, 7850) - Move version and serial to yaml, fall back to entPhysicalSoftwareRev - Add SFP voltage, current and power-on hours sensors - Add connUnitStatus, swOperStatus and swDiagResult state sensors - Add fabos_6505 test data and update existing fabos test data
…enms#21074) - hardware/version/features/serial from NETGEAR-SWITCHING-MIB agentInventory* version is now the firmware (e.g. 1.1.0.8) instead of the VxWorks kernel version; the OS moves to features alongside additional packages - fix sysDescr fallback regex field order (hardware, version, features) - move fan speed, temperature and PSU state sensors to YAML - add fan state, fan duty level and temperature limits - fix PSU state flapping other(0) -> operational(2) on every discovery, the old code stored the enum text as the current value - add LB4M test data
* Ruckus SmartZone: AP license total, consumed and remaining count sensors * Support plural sensor category labels * Add Cluster sensor category * Add License sensor category * Add Access Point sensor category
* Add deprecation annotations for device_by_id_cache and getidbyname * Replace getidbyname with DeviceCache in api_functions * Return 404 for unknown devices in api_functions A deleted or renamed device used to silently drop the device filter in several list endpoints (returning data for every device the user can see), or turn into 500/400 responses. Resolve hostnames/IDs through a single api_resolve_device() helper that throws a 404 ApiException when a non-empty hostname or device ID was given but no device matches, while an empty value keeps the list-everything behaviour. The rendered response body is identical to api_error(404, ...). - list_bgp, list_cbgp, list_ospf, list_vrf, list_mpls_services, list_mpls_saps, list_vlans, list_links, list_logs, list_services, list_ipsec: 404 instead of filter drop / 500 / 400 - add_edit_rule: 400 naming the unknown device instead of passing null to devices()->sync() (NOT NULL violation) - device_under_maintenance: drop redundant second DeviceCache::get() and dead check - get_graph_generic_by_hostname, get_graph_by_service: use the model from DeviceCache directly instead of device_by_id_cache() - phpstan-baseline-deprecated: device_by_id_cache count 5 -> 3 for includes/html/api_functions.inc.php * Add API tests for unknown device handling Cover the new 404/400 behaviour for deleted/renamed devices: list endpoints (bgp, cbgp, ospf, vrf, mpls, vlans) with an unknown hostname or device ID, eventlog/services/links/ipsec/maintenance by device, and alert rules referencing unknown devices. Also guard the unchanged list-everything behaviour without a hostname filter. * Use Device model instead of device_id checks in api_functions * Add API tests for inventory and parents endpoints * Inline device resolution and use Device model directly in api_functions Remove the api_resolve_device() helper (and the ApiException import) and inline the DeviceCache::get() + exists check in each caller. List endpoints keep the 404 for an unknown device; ipsec, inventory and the parents endpoints keep their original 400 status and message. Drop the intermediate $device_id variable in favour of $device->device_id, and align add_edit_rule's loop variable naming ($hostname / $device). Update the API tests accordingly (ipsec/inventory/parents expect 400). * Eloquent rewrites * Fix phpstan iterable value type in ApiUnknownDeviceTest * Use DeviceCache in get_fdb, maintenance_device and get_device_groups * Use Device model in get_port_stats_by_port_hostname and parent resolution * Apply rector ClosureToArrowFunction in get_fdb * Fix eventlog API test URLs (logs prefix)
asamNumberofAlarmEntries is the number of active alarms, but it was a state sensor with states for 0 to 6 only. Three minor alarms showed as critical, and seven or more alarms matched no state at all. As a count sensor it graphs the number of alarms; the severity is already alerted by the Alarm: Current Status sensor. The sensor gets a new RRD file.
* Detect Zscaler ZPA Service Edge running on Linux as zscaler ZPA Service Edge VMs run RHEL with a stock net-snmp agent and were detected as linux. Match them by sysDescr (Linux zpa-service-edge) and sysObjectID, exclude them from linux, and take version (kernel) and architecture from sysDescr. Add test data. * Drop redundant linux exclusions for ZPA service edge The zscaler definition is checked before the generic linux one, so the exclusions are not needed. They also stopped other hosts with a matching sysDescr but a different sysObjectID from being detected as linux.
* Centralize eventlog severity label classes into a Severity enum
The severity-to-bootstrap-label-class mapping was duplicated in the
global eventlog table (Table\EventlogController) and the device
overview eventlog component. Move it into a single
LibreNMS\Enum\Severity::labelClass() method.
* Remove port eventlog subview and legacy eventlog table handler
The per-port eventlog subview (includes/html/pages/device/port/events.inc.php)
and its print-event include are removed. The port page menu now links to the
existing eventlog table (device logs tab) pre-filtered by
eventtype=interface, which already renders interface events with port links.
Legacy ?view=events port URLs are no longer supported.
The bootgrid instances for the global eventlog page and the eventlog widget
POST to /ajax/table/eventlog, which is served by the modern
Table\EventlogController (route table.eventlog). The legacy handler
includes/html/table/eventlog.inc.php was only reachable via
ajax_table.php?id=eventlog, which no caller uses anymore.
Also removes the eventlog_severity() helper, which loses its last remaining
callers (print-event and the table handler) with this change.
Adds a feature test covering the new port page menu link.
* Link ports to their eventlog and support a reference filter
The port list (interface overview) now shows an eventlog link per port
that has events, pointing at the eventlog table pre-filtered by
eventtype=interface and reference=<port id>. The port page menu links
the same way, replacing the device-level interface filter.
The eventlog table accepts a new reference filter field; the device
eventlog page shows the filtered port in the filter bar with a clear
link and passes the filter through to table requests.
* Search eventlog by port label
Interface events display the port label in the type column but the
table search only covered the message field. Add a port relation to
the eventlog model and extend the table search so a search phrase also
matches interface events whose port ifName, ifAlias or ifDescr
contains the phrase, constrained to interface type so non-interface
events with colliding reference ids do not match.
* Move eventlog port label after the filter button
The port label of the filtered reference belongs at the end of the
eventlog filter bar, to the right of the filter button.
* Show a clear control for unresolved references
* Use Html::severityToLabel() for eventlog severity labels
The severity-to-label-class mapping already exists in the shared
Html::severityToLabel() helper. Use it in the global eventlog table
and the device overview eventlog component instead of the
Severity::labelClass() method and drop that method again.
* Escape the port label in the eventlog filter bar
The port label is device-controlled data (ifName/ifAlias/ifDescr) and
the filter bar is built by appending a JavaScript string as HTML.
HTML-encode the label before it is serialized into the script so it
is inserted as text instead of being parsed as markup.
* Document the eventlog search override and port relation
* Render the eventlog severity label escaped in the device overview
The device overview eventlog emitted the pre-rendered HTML of
Html::severityToLabel() with unescaped {!! !!} output. Extract the
severity to label class mapping into Html::severityClass() and render
the alert-status span with the class as escaped output instead.
Html::severityToLabel() now delegates to the new method and produces
unchanged output for all existing callers.
* Use Html::severityClass() in the label component
The x-label component carried its own copy of the severity to label
class mapping. Use the shared Html::severityClass() helper instead.
* Tidy up
---------
Co-authored-by: Tony Murray <murraytony@gmail.com>
GHSA-3772-gpfc-vxgw
GHSA-c6fm-453r-fqww
Also reset the matched port per lease so an earlier match does not mark later MAC addresses as raw html. GHSA-9h74-w58p-25rr
GHSA-9qqx-578p-5jvx
GHSA-3h4r-27fv-7g68
Covers nfs, oslv_monitor and the apps overview. Port links built from models now use the port label and escape it. GHSA-m6c4-f8pq-4hmq
Rediscover now requires access to the group and every member device. Static group create/update only accepts devices the user can access and keeps existing members the user cannot see. GHSA-v9xq-7j25-w35q GHSA-m97c-xmpx-2pjc
Settings are stored unvalidated, output them with @JSON so they cannot break out of the script context. GHSA-9g2c-v4wv-cxqp
GHSA-hjw6-647w-9c3q
Also never use a hostname containing path separators as a RANCID config file name. GHSA-8xx2-xqh2-c836
Co-authored-by: murrant <39462+murrant@users.noreply.github.com>
Add lnms plugin:sync to install all packages recorded by plugin:add. LIBRENMS_COMPOSER_PLUGINS sets the location of composer.plugins.json. When COMPOSER points at another composer.json, the plugin commands create it from the one in the install directory, so packages can be added without changing the install directory.
…s#21195) * alerts: namespace incident ids per dashboard widget instance * add validation --------- Co-authored-by: Tony Murray <murraytony@gmail.com>
* Fix ambiguous user_id in dashboard select Signed-off-by: Niko Naumanen <219071881+Niibizness@users.noreply.github.com> * Retrigger CI Signed-off-by: Niko Naumanen <219071881+Niibizness@users.noreply.github.com> --------- Signed-off-by: Niko Naumanen <219071881+Niibizness@users.noreply.github.com>
The device page (App\View\Components\Device\PageLinks) builds its SSH, Telnet and Web links with the override_device_ssh_port, override_device_telnet_port and override_device_http_port device attribs. The device list row actions only applied the http port, and silently dropped the ssh and telnet overrides. Append the ssh/telnet port overrides to the generated hrefs the same way the http port already is, and widen the attribs eager-load constraint to fetch all three types — getAttrib() reads the already-loaded relation, so an unconstrained lazy load would not kick in and a constrained one would silently return null. The gateone server URL keeps precedence over the direct ssh:// link, matching the device page behavior. Signed-off-by: Dan Meyers <danjbh@users.noreply.github.com> Co-authored-by: Dan Meyers <danjbh@users.noreply.github.com>
…oups (librenms#21069) * Fix service template Apply failing when the template has no device groups Device::scopeInDeviceGroup() and scopeNotInDeviceGroup() passed their argument through Arr::wrap(), which nests a Collection inside an array. With a non-empty collection Arr::flatten() hid this, but an empty collection (a dynamic service template with device rules only, no device groups) made whereIn() throw 'Nested arrays may not be passed to whereIn method', so ServiceTemplateController::apply() returned 500 and the cleanup of stale services never ran. Unwrap Collections to plain arrays before Arr::wrap() in both scopes. Fixes librenms#21039. * Remove unneeded comments --------- Co-authored-by: Tony Murray <murraytony@gmail.com>
…1267) * Update Health-Information.md Verbiage and option flag updates * Update v2 for verbiage and option flags in Health-Information.md Corrections of the corrections.
…nms#21281) Any sub-minute scheduled event keeps schedule:run looping for the whole minute, even when its when() condition is false.
Owner
Author
|
Test workflow does not run on the fork; closing. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fork-only CI run before opening upstream.