| Age | Commit message (Collapse) | Author |
|---|
|
|
|
Editors' TinyMCE output can contain unclosed void elements like <br>,
which is valid HTML5 but invalid XML -- three different places assumed
the stricter rule and broke or silently misbehaved on the looser one.
The Atom feed's <content type="xhtml"> block required real, well-formed
XML structure but was handed a raw, unescaped body string; switched to
type="html" with CGI.escapeHTML, matching how title/summary already
handle the same content. rewrite_links_in_body used libxml's strict XML
parser to rewrite internal links to be locale-prefixed, which raised on
exactly this class of malformed markup -- silently, since the whole
method was wrapped in rescue; nil, meaning the link rewrite (not the
save) quietly failed with no error anywhere. Replaced with Nokogiri's
lenient HTML parser, which repairs malformed void elements rather than
rejecting them; also drops the bare rescue now that the actual failure
mode it was guarding against shouldn't occur, and fixes two adjacent
bugs found while in this method: a typo'd /sytem/uploads/ regex that
could never match, and a missing https:// exclusion alongside the
existing http:// one.
Also addresses stale flash messaging surfaced while testing the above:
update's save confirmation was being clobbered by edit's own "locked
and ready" notice on the very next request, since nothing distinguished
a fresh lock acquisition from a redirect back after saving. The save
confirmation now names the next step (publish from Status) and flags a
stale translation if one exists, using Page#outdated_translations?,
already present but previously unused by any controller.
|
|
Remove dead pages#index and occurrences resource; fix menu_items#new
menu_items#new called menu_item_params, which requires a submitted
menu_item key that can never be present on a fresh GET -- new only
ever needs a blank record to render against.
pages#index and the entire occurrences resource were unmodified Rails
scaffold generator output, unlinked from any nav and unreachable except
by direct URL. pages#index had no controller action at all (@pages was
never assigned); occurrences#index queried every row with no scoping or
pagination and its layout referenced a scaffold.css asset that no
longer exists in the pipeline. Neither is a real editorial surface --
occurrences are auto-generated from an event's RRULE and were never
meant to be browsed as a flat list. Removed rather than repaired.
pages#preview and pages#sort_images remain, now as explicit routes
rather than under a full resources :pages block.
|
|
Fix button height mismatch, extend table styling to assets/users/events
a.action_button (bordered) and the state_changing/destructive/computation
pills (borderless) rendered at different heights, since each pill variant
set its own padding at higher specificity than the shared rule meant to
equalize them. Padding and border-radius now live only in one shared
rule; also restores the #page_editor prefix on a.action_button, needed
to outrank the #page_editor a wavy-underline rule.
table.assets_table, table.user_table, and table.events_table now share
table.node_table's border-collapse/header/hover treatment. tr
min-height is dropped -- browsers don't honor height on <tr> -- in
favor of cell padding, kept at zero for node_table (spacing already
comes from its h4/p content) and set to 8px for the plain-text tables.
|
|
Adds a fourth button category, .computation -- teal, matching Preview's
existing color -- for actions that compute a view without changing state,
the first case in the app that didn't fit reversible/state-changing/
destructive. Diff revisions now starts disabled and only enables once two
distinct revisions are selected, matching the disabled-state pattern
already used for Unlock + Back.
The header row is wrapped in a real <thead> (previously flat inside the
table alongside data rows) so a sticky diff bar can sit above the fold on
long revision lists -- selecting a comparison pair no longer requires
scrolling back up to find the button, or back down to confirm what's
selected, both surfaced live via one shared JS function.
The two most recent revisions are pre-selected by default, covering the
most common comparison (current against previous) with no extra clicks.
Deliberately scoped to numbered revisions only -- comparing draft/head or
autosave/draft was raised and set aside, since neither exists as a row in
this table by design and extending the comparison to them is a separate
feature, not a default on this one.
Also drops the dead "Edit" subnav link, redundant with nodes#show's
Status section covering the same action with state-aware labeling this
static link could never have.
|
|
Retires content_for :subnavigation on nodes#show entirely -- Preview
was fully redundant with Links' own preview URLs in every state, and
Edit is now a permanent, always-rendered action inside the new Status
section rather than a link floating at the top, consistent with the
"action lives next to the info" pattern People already established
for Unlock. Status surfaces head/draft/autosave plainly and gates
Edit/Publish/Destroy/Discard on lock ownership specifically, not mere
lock presence -- @node.locked? alone would have blocked the lock
owner's own session, caught by the click-test and fixed here rather
than shipped.
nodes#edit's action bar is rebuilt to sit outside form_for (both
button_to calls render their own nested form, invalid HTML the
browser was silently stripping) with Save wired back in via
form="..." rather than needing to live inside the form tag at all.
Also brings the locked-and-ready-to-edit flash message, and the
visual polish from this session's click-test: consistent button
heights across the bordered and pill-shaped variants sharing a row,
spacing between Status's data and its actions, and error_messages
styling reusing the destructive-red vocabulary already established
elsewhere.
|
|
revert! requires the lock to already be held, correct for nodes#edit
where the editor holds it throughout -- but both exclusive-case
buttons on nodes#show fire from a node that starts out unlocked,
which raised LockedByAnotherUser incorrectly. lock_for_editing! first
makes both call sites safe: a harmless re-stamp when already the
owner, a real but momentary acquisition otherwise, released again by
revert! itself once nothing is left to protect.
|
|
A blank title failed presence and length validation simultaneously,
showing two redundant messages for one problem; length now skips
whenever slug is already blank, matching presence's own skip
condition. Separately, create's failure path never computed
@selected_kind/@parent_id/@parent_name the way new does, so
re-rendering after a validation error silently lost which kind and
parent had been chosen -- for the auto-tagging/auto-templating kinds,
that meant a corrected resubmission could silently produce a plain
generic node instead. required on the title field closes the common
case without a round trip; the server-side fix remains the actual
guarantee, since required is trivially bypassed.
|
|
|
|
|
|
revert! discards exactly the topmost non-empty layer -- autosave if
present, else draft -- and reveals whatever's beneath it, releasing the
lock only once nothing is left to protect. Guards against destroying a
brand-new, never-published node's only draft, which would violate the
(head | draft) invariant every other method here assumes holds; the
view's Destroy/Discard button is gated the same way.
nodes#edit now calls lock_for_editing! instead of find_or_create_draft,
and always displays autosave || draft || head, resurrecting an
abandoned session's unsaved content by default with an explicit flash
explaining what's shown and how to get back to the last saved version.
The view drops content_for :subnavigation entirely: Show becomes
"Unlock + Back", Preview stays a plain link, metadata's own <details>
already replaced the old toggle, and Publish moves off this page for
good, per the earlier decision to manage the publish lifecycle
entirely from nodes#show. Save Draft and Save + Unlock + Exit appear
both above and below the form, given the body field alone runs 600px
on desktop.
|
|
Two real bugs surfaced by the full test suite, not by the new tests
written for this feature -- both were latent the moment lock_for_editing!
and save_draft! replaced find_or_create_draft, and only visible once an
existing test exercised the exact path each one lived in.
lock_for_editing! never stamped user/editor onto a draft that already
existed when the lock was acquired. find_or_create_draft used to do this
as part of acquiring the lock; splitting lock acquisition from draft
creation dropped it entirely, since it looked like draft-creation logic
rather than locking logic. Restored as an explicit step: claim authorship
only if none is set yet, editorship unconditionally, matching the old
behavior exactly.
save_draft!'s "no draft yet" branch set user/editor before calling
clone_attributes_from, whose first line is an unconditional self.reload
-- silently discarding both, since neither had been persisted yet.
clone_attributes_from also always copies published_at from its source
without the ||= guard used for template_name, and the source here is
always the autosave, whose published_at is never anything but nil --
meaning every single promotion, not just the first, was quietly
resetting a published page's published_at, which publish_draft!'s own
||= Time.now would then treat as never-published and re-stamp. Fixed by
running clone_attributes_from first on both branches, then applying
user/editor/published_at afterward, exactly as the existing-draft branch
already happened to do by accident.
Four controller tests updated to insert a real put :update between
get :edit and assertions that used to be true immediately after
visiting edit -- deferred draft creation means edit alone no longer
produces one, which is the intended consequence of this whole
redesign, not something these tests were meant to catch.
|
|
|
|
Fallout from the earlier nodes#show heading change, which replaced the
Title row with an <h1> -- unrelated to the autosave work in the
preceding commits.
|
|
Previously piggybacked on the main form's action and relied on a
.js.erb response to update #flash; now posts to the dedicated
autosave route with an explicit _method=put override, and reacts to
a 423 by stopping the timer and showing a persistent banner rather
than letting the next Save discover the conflict on its own.
|
|
Route Save and autosave through the new Node methods
update now promotes the current autosave into the draft via
save_draft! rather than writing submitted params directly; autosave
gets its own PUT member route so PATCH update can mean "deliberate
save" specifically, matching every other custom action on this
resource. Both now require the lock to already be held rather than
acquiring it themselves, which is a deliberate narrowing: through the
real UI this is always true, since neither can fire before edit has
loaded. Two existing tests called update directly with no prior
edit, which the old code tolerated by acquiring the lock as a side
effect; updated to call edit first instead of loosening the
guarantee back open.
NodesController#edit itself is unchanged and still calls the old
find_or_create_draft, so drafts are still created eagerly on entering
edit for now -- switching that over is the next step, not this one.
|
|
Introduces autosave_id as a third, unversioned layer above draft/head,
with lock_for_editing!, autosave!, and save_draft! as the new entry
points. Also fixes a real bug in wipe_draft!: its "no draft" branch
unconditionally released the lock, which was safe when "no draft" only
ever meant "nothing is happening" — no longer true now that a lock can
exist with only an autosave beneath it. lock_for_editing! deliberately
does not call wipe_draft! at all, for the same reason: an intruder
calling it while a lock was genuinely held would otherwise silently
steal it via wipe_draft!'s own unlock side effect, caught by the new
two-user lock test.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Sorts rules by hierarchy (base -> layout -> navigation -> buttons ->
tables -> forms -> components), consolidates three duplicated media
query breakpoints into one each, and adds section comments. Bundled
with this reorganization, since the file was rewritten wholesale:
- form.button_to.state_changing: new tinted-pill variant (blue) for
Publish/Restore/Unlock/Revoke, matching .destructive's pattern
- form.button_to.destructive: now a tinted pill at rest instead of
plain colored text, consistent padding at rest and on hover
- Scoped wavy-underline link-visibility fix for #page_editor,
table.node_table, table.assets_table, table.events_table,
table.user_table, .add_child_links, and the dashboard draft list
- #flash decoupled from the page's structural nav-to-content spacing
(.admin_content_spacer), which the flash div was silently providing
as a side effect whenever it rendered
|
|
|
|
|
|
parent_match Procs on CccConventions::NODE_KINDS, matched against
unique_path, decide which "add child" kinds show on a given node. Fixes
nodes#new not honoring a pre-selected kind (radio group and parent-field
visibility both defaulted to "generic" unconditionally).
|
|
Root-caused this session: appending a child to any node never widened
that parent's own rgt boundary, on the pinned revision (Gemfile tracked
main directly, chasing a too-conservative gemspec constraint - not, as
first assumed, a deliberate pin to avoid a known bug). Reproduced
cleanly on a single ordinary create with no concurrency and no bulk
operation involved, confirmed via the gem's own SetValidator, then
confirmed as the root cause of nodes_controller_test.rb's 3 long-standing
"pre-existing" failures - not three separate mysteries, one bug.
admin_controller's sitemap needed its own real conversion, not just a
drop-in: awesome_nested_set's lft column implicitly provided correct
depth-first tree order for free, which the old code combined with a
separate class-level each_with_level iterator. Both replaced by one
method, self_and_descendants_ordered_with_level, computing an ordered
[node, level] list in a single query-then-walk pass - checked against
the actual view template first (admin/index.html.erb) rather than
assumed, since it relies on list order alone to render correct visual
nesting.
lft/rgt/depth columns intentionally left in schema, unused - dropping
them is a separate, deliberately deferred migration once this is proven
running for a while, not bundled with the behavior change.
|
|
New :children option ("direct" or "all") on the [aggregate] shortcode,
composable with the existing :tags filter - both apply as independent
successive .where clauses, ANDing together automatically, no special
casing needed. content_helper.rb passes the current @page.node through
as :node whenever :children is requested, since Page.aggregate has no
way to resolve "which node" from the shortcode's bare option strings on
its own. Guarded so nothing changes for any existing [aggregate]
shortcode that never uses children= - :node stays nil, the new branch
never fires.
"all" resolves via node.descendants.pluck(:id) rather than embedding the
descendants relation directly as a subquery. Not proven strictly
necessary - extensive debugging this session initially pointed at a
subquery/table-alias collision, but every actual failure traced instead
to accumulated test-node debris left in the dev DB by earlier
interrupted runs, confirmed by re-running against a cleaned database.
Kept anyway as a one-round-trip-cheaper, more defensive default
regardless.
|
|
staged_slug existed to protect a live public URL from changing until
an editor explicitly publishes - correct once a node has a head, but
nodes_controller#update wrote to it unconditionally, so a brand-new
node being renamed before its first-ever publish showed a stale slug
in previews (e.g. nodes#new's resulting-path preview) that silently
diverged from what would actually go live.
Node#staged_slug= now applies directly to slug when head is blank -
nothing is live yet, so there's nothing to protect. Once head is
present, defers to staged_slug exactly as before. Verified both paths
directly: a never-published node's slug now updates immediately, and
an already-published node's rename still stays deferred until the
next publish, unchanged from existing behavior.
|
|
|
|
nodes#show's flat <table> of label/value pairs grew unreadable once
Admin Preview and Public Preview were added alongside Public Link -
switched to the div.node_description/div.node_content pattern already
used by nodes#edit, grouped into People/Dates/Links/Revisions/Tags/
Events/Children sections rather than one undifferentiated list.
People/Dates/Links use a flex-based item layout, not a second <table> -
multiple independent tables sharing one outer width but different
column counts produced misaligned columns with no shared grid. Flex
items size to their own content with a shared minimum instead.
Revisions and Children are both collapsed via <details>, no JS needed -
previously nodes#show didn't list a node's children at all, blocking
the ability to find/navigate erfa and chaostreff pages without already
knowing they're now standalone nodes. Revision list items link to
node_revision_path(@node, page) - the actual per-revision diff/restore
view, not the plain index.
nodes#new converted to the same pattern for visual consistency, plus a
few things surfaced along the way: submit buttons were unstyled
site-wide (a bare input[type=submit] with no border, fill, or hover
state - Create on this exact page was easy to miss entirely), fixed
with a bordered/bold treatment reusing the existing form.button_to
visual language rather than introducing a fourth button style. Title
regained the bold weight it had before the table rewrite dropped it
silently. Tag list grouping and full link/button semantic taxonomy
(show vs edit/add vs publish/revoke vs destroy, applied consistently
across every link on the page, not just buttons) are known follow-ups,
not attempted here.
|
|
A token's page could stop being node.head_id (superseded by a newer
draft) while still being published_at.present? - the old check only
compared against the current head, so a superseded page fell through
to direct rendering instead of redirecting, serving a stale frozen
snapshot indefinitely instead of the current live content.
Also handles scheduled publishing correctly: a page can be head_id but
not yet public? (published_at in the future) - that case must still
render directly, not redirect into a 404 on the not-yet-live public
URL.
|
|
Lets editors share a draft with people outside the admin system, via a
random per-page token rather than an enumerable id - /preview/<token>.
shared_previews#show is intentionally unauthenticated. Redirects to the
real public URL once the page is published. Surfaced on nodes#show
(Admin Preview + Public Preview, next to Public Link) as generate/revoke
buttons.
|
|
New column + unique index, plus ensure_preview_token!/revoke_preview_token!
on Page. Generated lazily (only when explicitly requested) rather than
via has_secure_token's default auto-generate-on-create, so a live,
shareable secret isn't silently minted for every page ever created.
|
|
Forms were already cleaned up in an earlier commit; the column itself
was the last remnant.
Dropping the column surfaced a live reference in events#index (fixed
separately, 970f108) - this migration file itself just never got
staged until now.
|
|
Was the last remaining gap in the occurrence-orphaning fix from earlier
this session: dependent: :destroy on Event#occurrences covers deletion
through ActiveRecord, this covers anything that bypasses it (raw SQL,
Model.delete). Confirmed zero existing orphans and zero NULL event_ids
before adding the constraint (event_id stays nullable - an occurrence
without an event is valid, one with a bogus event_id is not). Also adds
the index that should have existed on this column already - Postgres
needs one for the cascade to be anything other than a full table scan
per delete, and none existed before this. deferrable: :immediate added
so legitimate application code can still defer the check within a
transaction if ever needed - unrelated to, and not a fix for, the
fixture-loading issue below.
Applying this constraint broke fixture loading entirely (config fix:
81a07bf) and exposed a live bug in events#index (fix: 970f108), both
already committed separately - this migration file itself just never
got staged until now.
|
|
test_can_remove_a_node_with_an_event previously just called node.destroy
and get :index with no assertions at all - would pass whether or not
occurrences were actually cleaned up, or even if index rendered
correctly afterward. Now confirms occurrences genuinely exist before
destroy (otherwise a passing post-destroy count of zero is meaningless -
indistinguishable from "nothing to cascade in the first place"), scopes
the count to this event specifically rather than a global Occurrence.count
that could coincidentally pass regardless of whether this cascade works,
and checks the trailing index request actually succeeds rather than just
not raising.
First real test of the occurrences.event_id FK constraint added earlier
this session, not just the application-level dependent: :destroy.
|
|
events#index still read event.custom_rrule - a live bug the column-drop
migration introduced, not just stale test data. Caught by four test
failures (three fixtures/direct attribute hashes still setting the
removed column, one UnknownAttributeError from a stale fixture loaded
before any test method runs), none of which were actually testing this
view - "should get index" passed throughout with zero Event records
present, meaning it could never have caught a per-row rendering bug.
Strengthened to create one real event first, so a future stray
reference to a dropped or renamed column fails loudly instead of
silently passing on an empty table.
|
|
Adding the occurrences.event_id FK constraint broke fixture loading
entirely (every test failing at setup) with a misleading "Foreign key
violations found in your fixture data" error. The real cause: this
Rails feature does its own proactive integrity check during fixture
insertion, independent of the constraint's own deferrability, by
directly flipping pg_constraint.convalidated and re-validating - which
needs catalog-level privilege the app's DB role correctly doesn't have.
No actual fixture data was ever invalid; the check itself couldn't run.
Scoped to test.rb only - this never affected real application behavior,
and Postgres still fully enforces the actual constraint everywhere,
in every environment, regardless of this setting.
|
|
|
|
The following changes were already announced in the last commit, but
the files forgotten. Here's them actually attached.
lib/ccc_conventions.rb: NODE_KINDS registry replaces the kind-specific
branches previously scattered across nodes_controller#create (tags),
unique_path-position check). Each kind declares its parent lookup
(a Proc - Update's "update"/"press_release" continue to genuinely
find-or-create their year-folder via Update.find_or_create_parent;
erfa/chaostreff use a plain find_by_unique_name!, since their parent
nodes are fixed and a missing one should fail loudly, not be silently
created), its tags, and (new) its default template.
New read-only admin/conventions view dumps the NODE_KINDS registry as
a reference table for whoever's confused later about how a given kind
behaves - deliberately just renders the same label/hint text used on
nodes#new rather than re-describing parent/tags/template separately,
since decomposing the parent Proc would mean either invoking it
(Update's has a real side effect - creates the year folder) or
re-stating its logic a second time in a different shape, either of
which risks drifting from the actual behavior.
|
|
|
