Field Changes & New Layout

Was — the file name

pump_1788.jpg

Every reader had to know which folder the value came from, and that folder was not written down anywhere the row could see.

Now — the complete path

img/CAP-BBA2A600/assets/pump_1788.jpg

Read the stored value as it stands. Do not prefix a folder onto it.

Everything stays in the bucket we already have. The path is what separates one account’s files from another’s, so the path shape is the access model — not a cosmetic detail. The first segment is always the tree:

The second segment says who the file belongs to, and that decides the rest of the shape:

account        <tree>/<capexId>/<domain>/<file>
account-user   <tree>/<capexId>/<branch>/<userId>/<domain>/<file>
user           <tree>/<capexId>/user/<domain>/<file>
library        img/<branch>/<file>          doc/<branch>/<domain>/<file>
global         <tree>/<branch>/<file>   belongs to nobody

Worked through, with real values out of clone1:

img/CAP-0YNN74P/assets/2BQISjpe_1782915476233CTS5HSW.jpeg          account
img/CAP-BBA2A600/user/profile/avatar_1788.png                      user
doc/CAP-NJQJGDJ/copilot-quote/OrderCAPSKOCI7V4DZZE9K.pdf           account
doc/CAP-NJQJGDJ/copilot/warranty/file_example_XLS_PRPTP48.xls      account
doc/CAP-048YNSJ/document-manager/purchase-orders/OrderCAPSK.pdf    account
doc/CAP-BBA2A600/export/1/assets/camp_export_1788.xlsx             account-user — user 1 ran it
img/model-library/pump_x200.jpg                                    library
doc/export/system/nightly_1788.xlsx                                global

A browser upload cannot be written to its final home, because the record referencing it does not exist yet. So it lands in temp, which takes the owner slot and leaves everything below it intact — a staged path already says what it will become:

img/temp/assets            →  img/<capexId>/assets
doc/temp/export/1/assets   →  doc/<capexId>/export/1/assets

Filing copies; it does not move. A failed filing therefore still reads — the record keeps the staged path, and that path still resolves. Where a module has thumbnails, the small copy sits at <same folder>/thumb/<same file name>.

A complete path does not fit where a file name did. Widths were raised before any value was written, so no statement in the backfill could overflow.

The five varchar[] columns used to hold one comma-joined string; each file is now its own element, trimmed, with the empties dropped. A value that was nothing but separators became NULL rather than an array of blanks — an empty array reads as present in JS and would defeat every .length guard. Three further columns took paths without needing a widening, because they were already text: equipment_models.images, copilot_launch_proposals.quote and copilot_requests.quote.

Existing rows were filled in with where the file is now — the legacy flat folder — never where it is going. The copier moves the object and the save then records the new path. A value that already contained a slash was left alone: it is either a complete path already or an address somewhere else, and prefixing either would break it.

Comma lists and arrays were rebuilt entry by entry in their original order — the first picture is the one shown as the preview, so a reshuffle would have changed what customers see on thousands of rows. docusign_requests.attachments also carries a fix: a CER row named pa-documents/… while the object sat in cer-documents/. Those rows are re-pointed at the folder the object is actually in.

Each row was read, rewritten in memory and written back only where a key actually changed. Order inside an array is preserved.

Columns that already stored complete paths were left exactly as they were, as was email_logs.data — a frozen copy of what the mailer was sent, which rewriting would falsify.

equipment_models.images_old — new column

A straight snapshot of images taken before the conversion. 1,607 of its non-empty rows held scraped vendor JSON rather than file names; those were cleared out of images so the backfill only ever prefixed real names. It is an archive: no entity declares it and nothing should read it.

transactions.additional_documents — two new keys

sellerInvoice and buyerInvoice, built from invoice_name plus the folder the order’s request type wrote to. Transfers raise no invoice, so those rows get neither key; FMV shares one object between both sides.

file_migration_batches · file_migration_queue — new tables

The ledger that drives moving pre-cutover objects into the new layout. One queue row per destination object, not per source: one legacy object referenced by several accounts owes each of them its own copy, because the path is the access model.

update_search_key_for_warranties() — replaced

The warranty search key now takes only the part after the last slash. Left alone, inventory/warranty/abc.pdf would have indexed the folder too, and a search for “warranty” would have matched every row that has a document.

Every value in every column and key above, counted by whether it contains a slash.

Both path shapes are live at once, on purpose

An old file keeps its legacy flat path — inventory/images/x.jpg. A new save writes the new layout — img/CAP-…/assets/x.jpg. The backfill copies each legacy object to its new home and never deletes the source, so a half-migrated table reads correctly throughout. In clone1 today: 757,830 sticker paths still legacy, 10 already moved. A value with a slash is used as it stands either way.

38 bare names remain, and they predate this

All in camp_inventory_items.images, which is derived from inventory_sticker_details.sticker_name — and that column is fully converted. So those 38 array entries point at stickers that no longer exist. The same 38 are in capClone, so they are older than this change. Cleanup, not a blocker.

capClone is not a clean “before” picture. It already carries the conversion — every value in it is a complete path too — but all of them are still in the legacy flat folders. The new img/ and doc/ layout appears only in clone1.

The before-types in the type table were read from clone2 and capCloneDev, where the widening has not run — not inferred from the migration files.