Repository navigation
Conversation
Rewrite eight reason_display_messages strings that used internal terms
("coverage gap", "unread tail remains unproven", "inventory") or gave no
detail. Each new string says what happened, what was kept, and whether
the next run retries, checked against the code that emits the code.
icloud-notes promises no retry, because an unrecognized response waits
for a connector upgrade.
Patch-bump each connector and regenerate the implementation index and
library metadata.
Assisted-by: AI
Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
Connectors throw terminal codes (createConnectorFailure, code-shaped bare Errors, and the shared Amazon and browser sign-in helpers) that the completeness test never saw, so the Desktop app kept its own copy for them. Declare plain copy for the reachable codes it shows, and extend the scan so a thrown code without copy fails CI; the remaining undeclared codes are listed in a shrink-only allowlist for a later review. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
About 330 codes that connectors throw had no owner-facing copy, and the
copy that existed repeated the same few sentences in many spellings. An
audit of all 693 (connector, code) pairs found 11 owner situations: sign
in again, run again later, internal problem, items skipped, export file
problem, continues next run, settings problem, not offered, nothing to
import, site busy, and capped by a limit.
packages/polyfill-connectors/src/reason-copy.json now holds those 11
sentences once, and maps each connector's codes to one of them. 22 codes
keep a connector-specific sentence because the owner's next step depends
on that site. scripts/generate-reason-display-messages.mjs writes the
resolved text into each manifest's reason_display_messages, replacing
{site} with the manifest display_name. Manifests keep their format, so
Desktop and the reference implementation read plain sentences as before
and need no change. Connectors keep emitting their own codes.
reason-copy:check runs in npm test and fails when a manifest drifts from
the copy file. The shrink-only allowlist of thrown codes without copy is
gone: every thrown code now needs copy or must be runtime-generic.
Four connector tests that compared the copy set to the SKIP_RESULT
vocabulary now include the connector's thrown codes.
Every connector whose copy changed gets a patch bump.
Assisted-by: AI
Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…mbers A review of the spotify, chatgpt, github, amazon and gmail mappings found three that misled the owner: - amazon older_pages_deferred_page_budget said later runs collect the rest, but each run restarts the year at page one and keeps no resume offset, so an unchanged large year repeats the same pages. heb's copy of the same code already said the tail stays unproven. Both now get their own sentence. - gmail_progress_invalid_instant and _invalid_time_range come from the requested time range, so the owner can fix them: fix_settings, not unexpected_data. - spotify_recent_context_boundary_unverified is emitted even after a good read and retrying cannot help, so it no longer says to run again. The generator now refuses a manifest with two reason_display_messages members. JSON.parse keeps the last one but the splice edited the first, so --check could pass while the effective copy was stale. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…t deliver A review of every connector's code-to-sentence mapping found two patterns where the shared sentence told the owner something false. Ten page- and run-budget codes (shopify, google-calendar, google-contacts, oura, doordash, uber, tiktok, x-browser) mapped to continues_later, which says later runs collect the rest. These connectors save no continuation cursor when the budget stops them, so the next run rereads the same range. They now map to a new limit_repeats sentence that says older data may stay missing. Generic codes such as stream_collection_failed also fire for failures marked retry_on_connector_upgrade, so "Run again later" alone was wrong for them. incomplete_retry now adds that a connector update may be needed if the problem keeps happening. Single-cause codes move to a closer sentence: browser sign-in handoffs and interruptions to a new sign_in_unfinished, claude-code local reads to a new local_file_unreadable, anthropic export parse failures to unexpected_data, and about a dozen codes get their own sentence where the owner's next step is specific (a Steam privacy setting, gmcli limit settings, a Muse bot check, an ical subscription address). Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
jellyfin_error covered rejected sign-ins, bad settings, replies the connector cannot read and transient faults, and jellyfin_session_failed mapped every first-sign-in failure to 'sign in again'. Classify by cause: jellyfin_auth_failed for rejected credentials, the existing base-url codes and jellyfin_settings_rejected for settings, jellyfin_response_unreadable for replies that need a connector update, jellyfin_error for transport faults. First-sign-in failures stay non-retryable: the stored password was already sent, and a redispatch would resend it. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
ics_fetch_failed covered HTTP 401/403/404 and invalid URLs as well as network faults. Rejected access now reports auth_failure, an address the server refuses or cannot use reports ics_subscription_address_invalid, 429 reports rate_limited, and only transient faults keep ics_fetch_failed. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…ead failures The per-chat catch in fetchAndParseGmcliMessages reported every error as gmcli_query_failed, so a helper that lost its pairing or went missing mid-run told the owner to run again. Reuse classifyGmcliFetchError there. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
HTTP 401/403 now reports auth_failed, unsafe redirects report unsafe_redirect_refused, and a missing Location, oversized body or redirect loop reports carddav_response_unusable. Only HTTP-status transient failures keep carddav_request_failed. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…tp errors http_error carried both retryable statuses and non-retryable 4xx. A 4xx that is not 401/403/404 now reports anthropic_live_request_rejected, which only a connector update fixes. 401 already throws the sign-in error and 403/404 already report not_available or a null plan, so credentials never reached http_error. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
A non-auth 4xx (for example 404) was classified unrecognized_response, which is upgrade-only, but reported icloud_http_error, whose copy tells the owner to run again. Give it icloud_request_rejected so the reason matches the recovery. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
muse_stream_incomplete carried retryable failures and the page-cap and unstorable-text failures that only a connector update fixes. The non-retryable run failure now reports muse_stream_needs_update. Runs under the runtime still reach the owner through stream_collection_failed. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…e reporter The no-reporter path threw instinct_stream_incomplete for every failure, hiding classified codes and the upgrade-only invalid-shape and duplicate-key failures. Reuse the error's own code, report instinct_source_unreadable for upgrade-only failures, and keep instinct_stream_incomplete for what a retry can clear. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…an update failStream threw whatsapp_web_stream_incomplete for text that cannot be stored, which no retry fixes. The non-retryable failure now reports whatsapp_web_stream_needs_update. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…load failStream threw character_ai_stream_incomplete for unkeyed chat-list rows, which no retry fixes. The non-retryable failure now reports the existing character_ai_invalid_payload code. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
failStream threw perplexity_stream_incomplete both for a library that changed while it was read (a retry clears it) and for list rows the parser cannot place (only a connector update helps). The non-retryable failure now reports perplexity_stream_needs_update. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…dance_missing credit_card_export_unverified also covered source_structure_changed, where a retry is pointless until the selectors are repaired. A structural outcome now reports export_affordance_missing for credit cards too. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
runOptionalStream reported optional_stream_failed, copy that says a run again may help, for a SlackApiAuthError that needs refreshed credentials. Report the existing slack_auth_failed code for that case. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…ibrary metadata Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…n scanner Map each new branch to a plain literal so the reason-code completeness scanner can resolve it, and drop the instinct classified-code passthrough that needed a computed code. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
A run stops at 200 Activities pages or 200 trip-detail fetches. Before this change it saved no position, so every run started again at the newest trip, re-hydrated the same first 200 trips, and never reached older history on a large account. The trips cursor now records the newest trip already read, the Activities page tokens where the walk stopped, the trip IDs still waiting for detail, and the IDs already hydrated. The next run reads new trips down to the saved head, then continues the saved pages within the same page budget. The position is tagged with the run's since, until and resources; a different scope or full_refresh starts over. When the history ends and no detail is pending, the backfill is cleared and the stream completes. While it is open, the budget skips now say a later run continues (retry_by_runtime, retryable). A rejected or repeated page token fails the stream instead of restarting from the newest trip. Activities ordering and token lifetime across sessions are not verified against a live account. The hydrated-ID lists grow with trip count. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…un stopped A run stops at about 1,000 messages, 25 servers or 100 requests. A server whose walk was cut short kept no position, so the next run searched it again from the newest message. A server with more history than one run's budget never got past it, and every run re-read the same head. Each server's cursor now keeps a scope tag and a backfill: a list of ID ranges still to read, each with the oldest message ID consumed (before_id) and, for a head segment, the old newest ID to stop at (stop_id). The next run reads new messages down to the saved head, then continues the ranges with Discord search's max_id parameter. The page script's search allow-list now accepts a numeric max_id. A head larger than the budget becomes its own range, so neither position is lost. A changed since, until or resources, or full_refresh, drops the saved ranges. When every range is finished, backfill is null and the server is complete. The search offset ceiling (about 10,000 results) used to count as the end of a server. The walk now restarts at offset zero with max_id set to the oldest message read and continues to the real end. If Discord returns messages at or above the max_id it was given, the walk stops as source_unreadable instead of working around it. The run-limit and interrupted messages in the manifest no longer say the next run starts again from the newest message. Not verified against live Discord: that search honours max_id. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
A run stops after 90 seconds or its scroll ceiling. Before this change it saved only record fingerprints, so the next run started again on the newest orders and an account with more orders than one run could read never got its older history. When orders come from the order-history API, the cursor now keeps the API's "after" token where the walk stopped and the newest order already read, tagged with the run's since, until and resources. The next run reads new orders down to that head, then continues from the saved token. A head that does not finish within the run is kept as its own pending segment ahead of the older one. If the API refuses a saved token, the run drops it and starts from the newest order. A different scope or full_refresh starts over. At the end of the list the position is cleared and the stream completes. Orders read by scrolling the page (the Apollo cache or DOM path) cannot seek to a saved position, so that path keeps its earlier behaviour and its earlier recovery hint. Not verified against a live shop: how long an "after" token stays valid across sessions. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
… full one
A run stops at 100 pages. Before this change a capped walk kept its
starting day as the cursor, so the next run read the same pages again
and never got past them. The recovery hint said only a connector
upgrade could help.
Oura's pages have no documented order, so a day seen mid-walk is not a
safe place to resume. The walk now reads fixed 90-day windows
(start_date to end_date). A window counts only when all its pages have
been read; the saved frontier is then that window's end. If the page
budget stops a run inside a window, the frontier stays at the window's
start and the next run reads the whole window again. Records keep
their keys, so the overlap changes nothing. Sleep and its daily_sleep
helper must both finish a window before sleep records are emitted.
The frontier is tagged with the run's since, until and resources. A
different scope or full_refresh starts from the scope's since. With no
saved day and no since, the walk starts at 2015-01-01. The last window
ends tomorrow (UTC); when it is read, the frontier is cleared and
last_day records the newest day, as before. State from earlier versions ({ last_day }) loads as
the start. While the walk is unfinished, the budget skip now says a
later run continues (retry_by_runtime, retryable).
Assisted-by: AI
Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
… lines The cause splits shifted 114 allowed await-in-loop sites in 14 connectors, so the conformance check listed them as stale and new. Each row now points at its new line. Three new test loops get their own rows: anthropic-live rejected statuses, google-messages gmcli failure kinds and muse upgrade-only failures. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…Ds once backfill completes A trip whose GetTrip or GetReceipt call fails every time stayed in the pending list forever, and each run reported trips_deferred_detail_budget even when the detail budget was not the reason it stopped. The cursor now counts failed runs per trip in detail_failure_counts. After three failed runs the trip leaves the pending list without being marked hydrated, and the budget skip is reported only when there were more candidates than MAX_DETAIL_FETCHES. The hydrated trip and receipt ID arrays grew by about 78 bytes per trip for the life of the connection. They are only needed while a backfill is open, to avoid refetching details across overlapping pages, so they are now cleared when the backfill completes. New tests cover each permanent-failure case, the cleared arrays, a capped head burst over an existing backfill, and reordered resources keeping the saved page position. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…get on excluded messages When a run had an until bound, the head query still started at the newest message. It spent the request budget on messages the scope excluded, so a large server never reached the requested range, and each run added another saved range. The head and refresh queries now start at the snowflake for the until bound (max_id), and saved ranges with identical bounds are merged. Three smaller fixes come with it: - Resources are sorted and deduplicated before the scope is saved, so the same request in a different order keeps the saved backfill. - A search hit that can never be parsed no longer pins a capped walk. The walk checkpoints past it and reports discord_records_unreadable, as the base connector did on a full walk. - A widened scope keeps separate head and older intervals around the coverage it already has, so a run cut off in the new head still reaches the older floor. New tests cover the until bound over three runs, duplicate ranges, resource order, a permanently malformed hit, and adjacent snowflakes that collapse to the same millisecond. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
Each view (posts, replies, likes, bookmarks) stops after about 100 new posts per run. Before this change the next run only added posts newer than the last run's head, so anything older than the first run's cap was never read. STATE now keeps, per view, the newest post seen (the head) and the oldest post reached (the backfill position), tied to the requested scope. A later run first reads new posts down to the saved head, then scrolls to the saved position without counting those posts against the run's cap, and continues into older history from there. A head burst larger than one run keeps its own position, so it does not cut off the older backfill. Changed resources, a changed since or until, and full_refresh start again from the current head. Legacy STATE without backfill positions still loads. A post that appears in two views (an original that X also lists among replies) is still charged once per view, as before, so a first run reads the same number of posts as the base connector. The posts, likes and bookmarks descriptions now say that later runs continue into older history instead of saying it is never read. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…m a rejected saved token Before resume, every run read all order-history pages (up to the budget) and the fingerprint gate emitted orders whose content had changed, such as fulfilled to delivered. With resume, a run stopped at the newest order it already had, so a later change to an older order was never seen. The backfill position now continues as a rolling sweep: after reading new orders, a run spends its remaining pages walking older pages from the saved position. When the sweep reaches the last page, fingerprints of orders no longer listed are pruned, and the next sweep starts again from the newest page. When the whole history fits in one run, every page is read each run, as before. If Shopify rejected a saved page token on the last request the run was allowed, no checkpoint was written, so every later run sent the same token and stopped. The run now saves a reset position even when it has no request left. The run deadline is also passed into the reader, so a rejection that arrives late does not start a new request after the deadline. The saved scope is normalized before comparison: missing fields equal null, resources are sorted and deduplicated, and key order does not matter. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…ame-day ranges A 90-day window that needed more pages than one run allows never finished: the run read up to the page limit, emitted nothing for that window, saved the same frontier, and the next run did the same. The frontier now stores window_days. A window that stops before it finishes is halved on the next run, down to one day, and the size returns to 90 days after a window completes. If a single day still needs more pages than one run allows, the frontier stays and the skip no longer claims retry_by_runtime, because running again cannot help. A time range inside one day (since and until on the same date) made no request and reported the stream complete. Such a range now requests that day. The saved scope is normalized before comparison: missing fields equal null, resources are sorted and deduplicated, and key order does not matter. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…tail is missing The previous fix retired a trip after three failed detail fetches, but it counted every failure. Three runs of rate limits or network errors retired a trip for good: the next head pass stops at that trip, so it was never fetched again. Fetch errors are now typed. Only a 4xx other than 408 and 429, or a response the parser rejects, counts toward retirement. A 5xx or network error keeps the trip pending without a count. A 429 stops detail fetching at once, saves the remaining IDs and coverage, and then rethrows uber_rate_limited, so the runtime sees the rate limit. A trip whose detail failed left nothing in the output to say so: with receipts requested, a failed GetReceipt left the stream looking complete. Each requested stream with attempted trips still missing detail now gets a trip_details_unavailable SKIP_RESULT. Its hint is retry_by_runtime while any of those trips are still pending, and retry_on_connector_upgrade once they are all retired. New tests cover 429, 408, 503 and network failures on both detail calls, the rate-limit stop, the detail gap with receipts only and with both streams, a retired receipt, and the Activities page ceiling shared by head and continuation pages. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…lk past a broken message A changed scope reset the saved ranges but kept newest_id from the old scope. A walk with a narrower until could then save a newest_id above its own until. When a later run widened until again, it treated that head as already collected and skipped the messages between the two bounds. A scope change now drops the old head, a newest_id above the scope's until is not saved, and old coverage heads are used again only after they are read inside the bounds. A walk that reached the end of the list after a permanently malformed message returned no cursor, so every run replayed the same range and asked the runtime to retry. Such a walk now saves a completed cursor for the readable messages and reports only discord_records_unreadable, which asks for a connector upgrade, not a runtime retry. Identical or overlapping saved ranges are now merged when the cursor is loaded, before any range is walked. Before, they were removed only when the cursor was saved, after one of them had already moved, so the other re-emitted its messages. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…roll limit A resumed run scrolls back to the saved backfill position (the anchor) and then reads older posts. When the anchor was found on the last scroll the run allowed, nothing older was read, the cursor did not move, and every later run did the same while asking the runtime to retry. A view now gets one more scroll after it finds the anchor, within the run's total scroll limit. A seek that cannot reach the anchor reports a non-retry outcome. Finishing a large pending head on likes or bookmarks saved the wrong newest IDs: the older pending IDs were written before the newer ones and the list was cut to 50, so the next run treated the newest 100 posts as new and emitted them again. The pending IDs now go after the newer arrivals. The posts, likes and bookmarks descriptions and the refresh rationale now say that the posts stream reads two lists of about 100 new posts each, that later runs reach older posts only 60 scroll steps deep, and that scrolling back past saved posts counts toward X's daily reading allowance. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
The scope-change test changed resources from null to [] but never between two lists of the same length. A comparator that checked only the list length would keep a saved frontier from resources ["a"] when the run asks for ["b"], and no test caught that. The new case saves ["a"] and runs with ["b"]; it fails against a length-only comparator. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
The scope-check test added five lines above five allowlisted loops. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
A trip or receipt whose detail failed three runs was dropped from pending, and its failure count lived inside `backfill`, which the next complete run removed. Run 4 then emitted no `trip_details_unavailable` skip, so the stream looked complete. STATE now keeps up to 100 retired IDs per stream in `retired_trip_ids`. Every later run reports them as a non-retryable gap and does not fetch them again. full_refresh or a scope change clears the list. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
The saved head keeps the 50 newest post ids, sorted by id. An old pinned post sorts below them and was dropped, so every later unchanged run emitted the pinned post again. The run now records the pinned post it read or skipped as known, and the saved head keeps that id after the 50 newest. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
Search pages advanced by `offset` under a fixed `max_id`. A message deleted between two page requests shifted later results by one, so a surviving message was skipped, and the end-of-results check then saved the server as complete. Main has the same race. After each page, the next request now uses the oldest readable ID seen as an exclusive `max_id` at offset 0, plus the count of unreadable groups after that ID, which were already reported. A page with no readable hit has no ID to page by and still advances by offset, so the 9,975 offset ceiling check is unchanged for that case. Tests whose fakes ignored `max_id` or matched `offset=25` now honor `max_id` like the real endpoint and match on the second page instead. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…wser fixes Rows are matched per file in sorted order with the same column. The uber retirement loop moved into a per-stream test, so its column changed from 3 to 4. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
These files were format-clean on main. Allowlist rows move with the formatted lines. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
UberHttpError used a parameter property, which erasableSyntaxOnly rejects. The x_browser tests read retryable from a hint that can be a string, and cast object resources straight to StreamScope. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
Main published new versions of anthropic, apple_health, fitbit, netflix_export, spotify, strava, strava_browser and whatsapp. This branch had bumped the same connectors to the same numbers with different contents, so each now moves one past main. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
github-actions
Bot
added
evidence: stale
Both connectors saved their resume scope in STATE as a JSON string, so
the cursor held escaped quotes. The PageShim harness checks that
discord_browser STATE holds only IDs and keys, and that check failed.
STATE now saves the scope as a plain {since, until, resources} object.
A saved scope in the older string form gives the same comparison key,
so STATE written by earlier builds of this branch still resumes.
A changed scope and full_refresh still discard the saved position.
The x_browser PageShim test expected STATE without backfill_scope and
backfill. Both fields are needed: without `complete: true` a later run
treats the cursor as one whose backfill never started and walks the
whole list again, and the scope is what detects a changed range. The
expectation now includes them. The two later subtests that re-emitted
the pinned post did so only because the first subtest failed before it
saved STATE, so they ran with none; they now fail with that cause.
Assisted-by: AI
Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
#610 and #626 both bumped discord_browser, oura, shopify, uber and x_browser from main, to equal numbers for x_browser, discord_browser and uber. Each now moves one past the higher of the two. Owner-facing copy comes from reason-copy.json, so the hand-edited sentences from #626 give way to it. With resume in place, the budget stops for oura, shopify, uber, x_browser and discord_browser continue on the next run, so they map to continues_later instead of limit_repeats. Uber's new trip_details_unavailable maps to skipped_items. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
Main published spotify 0.1.53 and linkedin 0.3.43. Spotify here moves to 0.1.54; linkedin stays at 0.3.44, which is already past main. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
…ines #628 moved LinkedIn's sequential loops and added two, without updating the allowlist, so the conformance check fails on main. Six rows move with their code; the section loop's first await is now the skills walk at 957:14. The two new loops are skills pagination, where each page's start offset depends on the prior page, and the skills-walk test cases, which share one fetch shim. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
Main published spotify 0.1.54 with different content, so the merged spotify moves to 0.1.55. LinkedIn allowlist rows follow main's #629. Assisted-by: AI Signed-off-by: Tim Nunamaker <tnunamak@gmail.com>
Main released Spotify 0.1.52 to 0.1.56. Spotify moves to 0.1.57, because this branch stamps its reason copy into the manifest. Main's two new skipped-entry codes keep their specific wording as custom messages, and spotify_stream_outcome_unproven maps to incomplete_retry. The index and library metadata are regenerated. Signed-off-by: Tim Nunamaker <tnunamak@gmail.com> Assisted-by: AI
Main's Spotify 0.1.56 throws spotify_playlist_revision_unproven. The connector catches it to drop the playlist streams, but the reason-copy test requires copy for every thrown code. It maps to incomplete_retry: Spotify did not confirm the playlist revision, so a later run can try again. Signed-off-by: Tim Nunamaker <tnunamak@gmail.com> Assisted-by: AI
This branch has not been deployed
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.
Draft. Integration branch for connector work that is ready but held for one review. Each PR below merges here when its own checks and review pass, and this PR merges the set to
mainat once. Do not merge yet.reason-copy.jsonand stamped into each manifestpage_init_script, Instagram posts, and other Mobile portsMerging #626 after #610 moved its page-budget codes from
limit_repeatstocontinues_laterinreason-copy.json, because those runs now resume. Ubertrip_details_unavailablemaps toskipped_items. Where #610 and #626 bumped the same connector to the same version, each now sits one past the higher number.Merging main brought in Spotify 0.1.52 to 0.1.56, so Spotify is now 0.1.57 here. Main's two new skipped-entry codes keep their own wording as custom copy.
spotify_stream_outcome_unprovenandspotify_playlist_revision_unprovenmap toincomplete_retry.To do
Assisted-by: AI