FrappeTestCase on this branch rolls back per class, not per test, so
sibling tests leak state. test_new_warehouse_can_inherit_inventory_account
left an explicit account on the root group, which made later ambiguous
fixtures resolve through the root: the insert validation stopped raising
and the unresolved warehouse stayed in the map.
The fixture helper now clears group warehouse accounts so every call
re-establishes ambiguity. The fallback test also clears the account of
the warehouse it picks, since a leftover explicit account skips the
single-account fallback it asserts.
Move the insert-time check from before_insert to validate. before_insert
runs before set_new_name, so the validation message rendered the
warehouse name as None. validate runs after naming and only applies to
new documents via is_new().
Resolve inheritance through the parent's lft/rgt bounds instead of the
request-cached warehouse account map. The cached map can be stale within
a request (a parent created moments earlier is missing from it), which
made get_warehouse_account trigger a full nested-set rebuild_tree and
could falsely reject a child whose parent carries a valid account.
rebuild_tree enables auto_commit_on_many_writes, which must not run
inside a document insert.
v15 adaptation of the develop fix: the timer flow here runs through
add_time_log via make_time_log, so gate that instead of the
start_timer/complete_job_card methods that only exist on develop, and
hide Start Job while transfer is pending. Corrective job cards and
job cards without items (transfer against Work Order) stay exempt,
matching validate_transfer_qty on submit.
`time_diff_in_hours` returns hours, so `time_in_mins` needs `* 60`, not
`/ 60`. Matches `Job Card.validate_time_log_row`.
No behaviour change: the `doc.save()` on the next line runs Job Card's
`validate`, which recomputes `time_in_mins` correctly before the row is
written. This only stops the expression from reading as a bug.
(cherry picked from commit 422a9161dd)
* test: cover both repost branches and the no-repost case
* fix: queue repost for entries backdated by a concurrent submit
---------
(cherry picked from commit 399ff463cc)
Co-authored-by: nareshkannasln <nareshkannashanmugam@gmail.com>
update_qc_reference() writes the QI link and bumps the reference
document's modified timestamp via raw db writes, which emit no realtime
event. A reference form (Purchase Receipt, Delivery Note, Stock Entry,
Job Card) still open in the browser keeps the old timestamp and fails
the timestamp conflict check on the next save/submit, forcing a manual
refresh after every QI submit/cancel/delete.
Calling notify_update() on the reference publishes the standard
doc_update event, so an open, unedited form silently reloads and syncs
its timestamp. get_lazy_doc skips child table loading since
notify_update only needs the parent row.
(cherry picked from commit 647452c95b)
covers the case where the percentages are correct but the accumulated
sum is 100.00000000000001. two rows can never drift, since the second
reconstructs exactly as 100 - first, so the case needs three rows.
(cherry picked from commit 4afba94d1c)
# Conflicts:
# erpnext/selling/doctype/sales_order/test_sales_order.py
the total of allocated_percentage was compared to 100 with exact float
equality, so a correct allocation could be rejected when the sum drifts
in binary floating point (10.0 + 58.02 + 31.98 -> 100.00000000000001).
round the total to the field precision before comparing, in both
SellingController.calculate_contribution and Customer.validate.
(cherry picked from commit f7b2775829)
