Document archives written without some tables' data (sc-23462) - #312
Document archives written without some tables' data (sc-23462)#312simozzy wants to merge 3 commits into
Conversation
Covers what a user now sees when a project contains a table with no data behind it: the archive is still written, the skipped tables are named, and a restore re-creates them empty. Adds the July release entry. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Deploying with
|
| Status | Name | Latest Commit | Preview URL | Updated (UTC) |
|---|---|---|---|---|
| ✅ Deployment successful! View logs |
plaidcloud-docs | 689bf8f | Commit Preview URL Branch Preview URL |
Aug 02 2026, 07:54 PM |
The page claimed a restore re-creates those tables "with their shape but no rows". It doesn't: the archive's schema comes from reflecting the physical table, which is precisely what is missing, so the entry restores with no schema behind it and is listed at the end of the restore instead. Also notes that a scheduled archive can only record skipped tables in the log — there is nobody on screen to tell. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Adversarial review recordA cross-repo review pass over this PR together with plaid #6848 and PlaidClient #1940 caught a real defect in the first version of this page, which is worth recording because the wrong sentence was the plausible one. It claimed a restore "re-creates those tables with their shape but no rows". It doesn't. The archive's table schema comes from Second correction in the same commit: the original paragraph promised "the export finishes with a message naming them" without qualification, sitting directly above the Archiving Schedule section. A scheduled or queue-driven archive has nobody on screen to tell, so it records skipped tables in the log only. That caveat is now a Also fixed before pushing: the caveat was first written as The release-note entry was corrected in step ("a restore lists them" rather than "re-creates those tables empty"). Nothing else in the pass flagged this PR: the behaviour described matches what the backend actually returns and what the client actually shows. |
…ames skipped tables (sc-23462) Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
Obsolote |
* docs: cover the last four days of merged customer-facing work Supersedes #311, #312, #313 and #317, which were each written against an August page that did not exist yet or a July page the work did not ship in. All four were green and stuck; their content is carried here, rebased and corrected, plus the five items that had no docs at all. Release placement corrected: sc-23460 and sc-23477 both merged on 2 Aug, after the beta-202607 promotion on 1 Aug, so they ship in the August release, not July. #311 and #313 filed them under July. Guides: - ai-coding-agents/getting-started — Keeping Results Small (row caps, fields=, count_only) and Column-Oriented Results (sc-23424, sc-23423). - ai-coding-agents/troubleshooting — a short list is not a complete one, and what a `_fmt: cols` result is (sc-23424, sc-23423). - ai-coding-agents/tracing-allocations — period-column detection and the explicit-column check, forward what-ifs are the exception, and combination cuts where enabled (sc-23477, sc-23429). - ai-assistant/using-ai-assistant — diagrams always render as a picture, and quantities are asked for rather than invented (sc-23476). - projects/archive-a-project — tables archived without their data, and a failed export now reporting failure (sc-23462, sc-23448). August What's New: the four above plus the leaner tool catalogue (sc-23426), the Lookup/Join output-column save fix (sc-23460), email and SMS sending (sc-23469), change detection across an edit and warehouse name search (sc-23424). Deliberately not documented: sc-23423's own measurement detail, and the combination-cut default, which stays off — the guide says so rather than describing an off-by-default search as available. Verified: astro build clean over 1530 pages, and every anchor and page link added here resolves in the built output. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * docs: quote the measured saving, not the flattering one sc-23423's live find-traffic figures are ~29% bytes / ~15% tokens. The first draft said "about a third" for both pages, which reads off the 34% table_find case rather than the mix — the exact overstatement CLAUDE.md warns against. Says ~30% of characters and ~15% of reading budget now, and explains why the two differ instead of letting the larger one stand for both. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> --------- Co-authored-by: simozzy <simon@plaidcloud.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
For the record: this content was not dropped — it was carried into #323 (merged), rebased onto current |
1 participant
Docs half of sc-23462 — plaid #6848, PlaidClient #1940.
A project export used to fail outright, producing no file at all, when a table listed in the project had no data behind it. It now archives everything else, names the tables it left out, and records them so a restore re-creates those tables empty rather than presenting them as complete. The export window also names the project it is archiving.
## Fixedentry.🤖 Generated with Claude Code