2 Commits

Include renames

Author	SHA1	Message	Date
steveandClaude Opus 4.8	5eedcaf945	Refuse deleting a plant that has seed lots ... seed_lots.plant_id is ON DELETE CASCADE, so deleting a plant would have silently destroyed the purchase records attached to it — vendor, cost, germination rate, gone with no warning. Plantings were already guarded with ErrPlantInUse because their FK is RESTRICT; lots needed the same guard for the opposite reason, since their FK would have obliged rather than refused. Refusing lets the owner delete the lots deliberately if that is really what they meant, which is the same shape as the existing clones-and-edits path. Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]> Claude-Session: https://claude.ai/code/session_01H3zbym8Doka2d7D48maSgZ	2026-07-21 01:29:24 -04:00
steveandClaude Opus 4.8	81481ede2f	Seed provenance and inventory: source links and seed lots (#50) ... "German Red Garlic" now links back to the Johnny's page it came from, and pansy knows how much is left. Two modelling calls, both of which look like omissions unless stated. The plant catalog stays FLAT — no species/variety hierarchy. A row in your catalog just is the thing you plant; the built-in "Garlic" is a generic starting point you clone. A hierarchy buys taxonomy nobody asked for and complicates every join. Inventory belongs to a PURCHASE, not to a variety. You might buy the same garlic from two vendors in two years at two prices and two germination rates; quantity columns on plants would collapse all of that into one number and lose it. So lots get their own table, and owner_id sits on the lot rather than being inherited from the plant — you can buy generic built-in Garlic from Johnny's and the record is still yours alone. Lots are never shared along with a garden. `remaining` is derived, never stored: quantity minus the summed effective count of the active plantings attributed to the lot. The alternative — decrementing a column when you plant — drifts the moment anything edits or removes a planting behind its back, and once it has drifted there's no way to tell what the true number was. Removing a plop gives the seed back with nothing having to remember to. The usage query returns raw radius/count/spacing rather than a SUM, because the effective-count formula lives in the service and reproducing it in SQL would guarantee the two drift apart. source_url is validated as http(s) with a host, on both plants and lots. It renders as a clickable link, so that check is load-bearing rather than tidiness: without it a stored javascript: URL becomes script execution the moment someone clicks it. Schemeless and relative URLs are refused too — they'd resolve against pansy's own origin, which is never what a vendor link means. A planting's lot must be the actor's AND a lot of the plant being planted. Attributing a garlic planting to a tomato lot would silently corrupt every remaining count downstream, so it's refused rather than stored, and re-checked on update in case a plant swap invalidated a previously-valid attribution. Deleting a lot leaves its plantings alone with a null link: the plants really were in the ground, whatever happened to the record of the packet. Closes #50 Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]> Claude-Session: https://claude.ai/code/session_01H3zbym8Doka2d7D48maSgZ	2026-07-21 01:20:50 -04:00