Closes#9092.
`restore-version-from-plugin` accepted `_reject` as a dead parameter and
its stream had no `rx/catch`, so errors raised during the restore flow
(failed `rp/cmd! :restore-file-snapshot`, persistence timeouts, or
exceptions inside the watch body) silently swallowed instead of
rejecting the plugin-facing promise at `file.cljs:81`. Plugin code
that did `await version.restore()` would hang indefinitely on any
failure.
Wire `reject` through and wrap the emission with the same `rx/catch`
pattern already used by `create-version-from-plugins` in this file.
- Rename `_reject` to `reject` in the function signature
- Wrap the `rx/concat` body with `rx/catch` that calls `(reject error)`
and returns `rx/empty` on error, mirroring `create-version-from-plugins`
- Add a CHANGES.md entry under the 2.17.0 Unreleased bugs-fixed section
Signed-off-by: Andrey Antukh <niwi@niwi.nz>
Co-authored-by: Andrey Antukh <niwi@niwi.nz>
The malli schema for the LDAP provider params (`schema:params` in
`backend/src/app/auth/ldap.clj`) declared the bind-password slot as
`:bind-passwor` (missing trailing `d`). The runtime code in the same
file uses `:bind-password` everywhere — `prepare-params` reads
`(:bind-password cfg)` on line 21 and `try-connectivity` reads
`(:bind-password cfg)` on line 89. Effects of the typo:
1. The schema slot for `:bind-password` is missing, so a wrong type
(e.g. a number or vector instead of a string) for the actual key
slips through `check-params` unvalidated. Malli `[:map ...]` is
open by default, so the genuine `:bind-password` key is silently
accepted as an unknown extra key.
2. Anyone reading the schema (operator, future contributor, or
tooling generating docs) sees a non-existent `:bind-passwor`
parameter and could legitimately set that key — schema would
accept it, runtime would never read it, LDAP bind would silently
fail with a confusing "no password" error.
Cross-checked against the pre-malli `clojure.spec` shape removed in
commit 88fb5e7ab (2024-10-29, "♻️ Update integrant to latest
version", which carried the spec→malli migration). The deleted spec
defined `(s/def ::bind-password ::us/string)` correctly — the typo
was introduced when re-typing the keys into the new malli vector-of-
tuples form.
Add a CHANGES.md entry under the 2.17.0 Unreleased 🐛 Bugs fixed
section.
One-character fix.
Signed-off-by: Andrey Antukh <niwi@niwi.nz>
Co-authored-by: Andrey Antukh <niwi@niwi.nz>
login-with-ldap raised a :restriction exception with the message
"ldap auth provider is not initialized" stored under :hide instead
of :hint. ex/raise (common/src/app/common/exceptions.cljc:33-34)
uses :hint as the ExceptionInfo message and the downstream error
formatters only read :hint (line 250, 312) — :hide is unread
anywhere in the codebase (0 other occurrences vs 447 for :hint).
Effect: when LDAP is misconfigured, operators saw the generic
"restriction" error message instead of the diagnostic string. The
typo has been present since the LDAP command was first introduced
by commit 14d1cb90bd (2022-06-30, "Refactor auth code") and was
carried forward through 6cdf696fc (2023-01-05, "Fix issues on ldap
provider and rpc method") without ever surfacing as a code-review
comment.
One-character fix: :hide -> :hint. Add a CHANGES.md entry under
the 2.17.0 Unreleased 🐛 Bugs fixed section.
Update agent configurations: change commiter mode to all, rename
engineer agent to "Penpot Engineer", and remove obsolete testing agent.
Add new read-only planner agent for architecture analysis and planning.
Add four new skills: bat-cat (syntax-highlighted cat clone), fd-find
(fast file finder), jq-json-processor (JSON processor), and ripgrep
(fast text search).
Add fd-find and bat packages to devenv Dockerfile.
Update .gitignore to exclude opencode package-lock and plans directory.
Signed-off-by: Andrey Antukh <niwi@niwi.nz>
Expand the Pull Requests section with detailed guidance on PR title
format, description expectations, branch naming conventions, the review
process, and a list of PRs that will not be accepted. Also clarify the
'Discuss Before Building' rule to link to GitHub Issues and Discussions
and reference Taiga stories. Update the Table of Contents with nested
links for all new subsections.
Signed-off-by: Andrey Antukh <niwi@niwi.nz>
The "Update Settings" button in Your Account > Settings and Notifications
was always enabled, even when the form had no changes, and clicking it
emitted a success notification despite no data being modified.
Disable the submit button when the current form data equals its initial
state, so it activates only when there are actual changes to persist.
Signed-off-by: moorsecopers99 <patellscott18@gmail.com>
Signed-off-by: Andrey Antukh <niwi@niwi.nz>
Co-authored-by: Andrey Antukh <niwi@niwi.nz>
The submenu opened by hovering Help & Learning in the user account
menu rendered with a vertical offset, making it appear visually
disconnected from its parent row and aligned instead with the
Community
Signed-off-by: Juan Flores <112629487+juan-flores077@users.noreply.github.com>
Closes#9108.
The `case` expression in `get-info` (`backend/src/app/auth/oidc.clj`)
dispatched on `:token` and `:userinfo` keywords, but the provider map's
`:user-info-source` value is a string — both from config (the malli
schema in `app.config` pins it to one of `"token"`, `"userinfo"`,
`"auto"`) and from the hard-coded Google / GitHub provider maps (which
already write `"userinfo"`). Strings never equal keywords in Clojure
`case`, so every call fell through to the auto-fallback that prefers
ID-token claims and only hits the UserInfo endpoint when claims are
empty. The net effect: setting `PENPOT_OIDC_USER_INFO_SOURCE=userinfo`
did nothing, and OIDC flows whose IdP requires the UserInfo endpoint
(so claims come back empty/partial) failed with "incomplete user info".
- Extract a pure helper `select-user-info-source` that maps the raw
config string to a dispatch keyword (`:token`, `:userinfo`, `:auto`),
falling back to `:auto` for unknown / missing / accidentally-keyword
values
- Rewrite `get-info`'s `case` to dispatch on the helper's output so
the arms unambiguously match the normalised keyword
- Add vitest-style deftests in `auth_oidc_test.clj` pinning the three
valid strings, the nil / "auto" / unknown fallback, and the reverse
regression (a keyword input must not slip through as if it were the
matching string)
- Add a CHANGES.md entry under the 2.17.0 Unreleased `🐛 Bugs fixed`
section linking back to #9108
Signed-off-by: Andrey Antukh <niwi@niwi.nz>
Co-authored-by: Andrey Antukh <niwi@niwi.nz>
* ✨ Remove the need to navigate to page for deletion operation
* 🐛 Fix multiple selection with applied-tokens on stroke-color
* 🐛 Fix button position on page header
---------
Co-authored-by: Andrey Antukh <niwi@niwi.nz>
The plugin parser's parse-point returned a plain `{:x … :y …}` map,
but shape interaction schemas (for example schema:open-overlay-interaction)
require the attribute to be a `::gpt/point` record. `(instance? Point {:x 0 :y 0})`
is false, so validation silently rejected plugin `addInteraction` calls
that passed `manualPositionLocation`; only a console warning was produced.
Change parse-point to return a `gpt/point` record via `gpt/point`.
All three call sites (parser.cljs:open-overlay, plugins/page.cljs,
plugins/comments.cljs) continue to work because Point records support
the same `:x`/`:y` access plain maps do.
Add a unit test that covers nil input and verifies the returned value
satisfies `gpt/point?`.
Github #8409
Signed-off-by: FairyPigDev <luislee3108@gmail.com>
Signed-off-by: Andrey Antukh <niwi@niwi.nz>
Co-authored-by: Andrey Antukh <niwi@niwi.nz>
When a text element has a line-height coming from a design token, the value
may be a number (e.g. 1.5) and fails frontend data validation expecting a
string. Normalize line-height before creating the typography style so the
operation succeeds without throwing an assertion error.
Signed-off-by: juan-flores077 <toptalent399@gmail.com>
The viewer-side `obfuscate-email` helper used by `anonymize-member` when
building share-link bundles called `clojure.string/split` on the raw
email input and then on the extracted domain. Two failure modes:
1. When the stored email had no `@` (legacy data, LDAP-sourced UIDs, direct
DB inserts, or fixtures that bypassed `::sm/email`), destructuring
left `domain` bound to `nil` and the follow-up `(str/split nil "." 2)`
raised `NullPointerException`. Because `obfuscate-email` runs inside
`get-view-only-bundle`, the exception aborted the whole RPC response
for share-link viewers, not just the field.
2. When the stored email used a single-label domain (`alice@localhost`),
`(str/split "localhost" "." 2)` returned `["localhost"]`; destructuring
bound `rest` to `nil` and the final `(str name "@****." rest)` produced
a dangling-dot output `"****@****."` (nil coerces to empty in `str`).
Guard both split calls with `(or x "")` so the chain is nil-safe, and
emit the trailing `.<tld>` segment only when `rest` is present. Add three
`deftest` groups covering the happy path, dotless domains, and malformed
inputs (nil / empty / no-`@`), plus a CHANGES.md entry under the 2.17.0
Unreleased bugs-fixed section.
Signed-off-by: Andrey Antukh <niwi@niwi.nz>
Co-authored-by: Andrey Antukh <niwi@niwi.nz>
