Repository navigation
Changed
-
Visual Studio solution versioned with the repository (
Kubuno.Forum.slnx, generated by vskubuno, the naming of
every Kubuno repository): the server is the projectKubuno.Forum.Server, the frontendKubuno.Forum.Web, in
the solution folders Server, Libraries and Frontend, with the shared launch profile "Forum (Kubuno Core + browser)".
The crate and npm names do not change. -
The forum now runs on PostgreSQL, MySQL/MariaDB or SQLite. The engine is an
administrator choice read at run time ([database] engine), and a single
binary connects to whichever is named — no rebuild. The module was moved off
its PostgreSQL-only stack onto the sharedkubuno-dbfoundation: one enum pool
in place ofPgPool, engine-agnostic bind parameters, and the fragments the
three engines spell differently (ON CONFLICT/ON DUPLICATE KEY,RETURNING,
= ANY, casts, upserts) produced by the dialect layer. Every insert now mints
its primary key in the process (noRETURNING, which MySQL lacks), reads
integer results at a portable width, and keeps all SQL schema-qualified with
strictly increasing, never-reused placeholders. -
Full-text search rebuilt to work on every engine. Post bodies and topic
titles are reduced to Snowball French stems and deaccented at write time
(stored inposts.body_norm/topics.title_norm), and a query is put through
the same reduction and matched with a portableLIKE, ranked title over body —
replacing the PostgreSQL-onlytsvector/websearch_to_tsquery/
ts_rank_cd/ts_headlinepipeline. Search results keep their highlighted
snippet. Typo tolerance that relied onpg_trgmis dropped; inflected and
accented queries still match.
Security
-
Attachments and images from Drive in posts no longer rely on the access-token cookie the web client
used to keep readable by page scripts; they use short-lived signed tickets. Attachment links now
download the file. Requires a Kubuno core that issues signed tickets (POST /api/v1/auth/tickets) and@kubuno/sdkwith the signed-URL helpers. -
Security fixes from the shared database layer (kubuno-db 0.9.0). The
database password can no longer appear in a log through the debug output of
the database settings. -
Duplicate-report protection preserved on every engine. The "one open
report per member per post" guard, previously a PostgreSQL partial unique
index (which MySQL cannot express), is now enforced in the service before the
insert, so re-reporting a post you have already flagged is still refused until
the first report is handled. -
Database driver updated past an unfixable advisory. The previous line
pulled in an RSA implementation vulnerable to a timing side-channel
(RUSTSEC-2023-0071) for which no fix will ever exist. The new line does not
depend on it at all, and it refuses any SQL string built at run time unless it
has been vouched for. Most queries are compile-time constants; the few
assembled at run time (dynamic list filters, the per-engine dialect fragments)
go through the sharedkubuno-dbbuilder, which vouches for the text and
guarantees placeholders are used once each in ascending order. -
Input validation library updated. The version in use carried
RUSTSEC-2024-0421 through its domain-name parser, which accepted Punycode
labels that decode to plain ASCII — a mismatch an attacker can use to make two
different names look like one.
Fixed
-
Oracle MySQL installs use the migrations written for them. The module now runs its
mysql-oraclemigration set when the database is Oracle MySQL, instead of the generic MySQL set. -
The forum installs on Oracle MySQL 8. Its schema migration gave a
TEXTcolumn a literal default, which MariaDB accepts but MySQL 8 refuses
(error 1101), so the module could not start on MySQL. On Oracle MySQL a variant
of that migration (migrations/mysql-oracle/) now declares the same default as
an expression; MariaDB keeps running the original file. Both are recorded under
the original checksum, so a database that already applied it is unaffected. -
Database migrations keep the same checksum on every OS. The repository now
pins line endings to LF (.gitattributes), so a checkout on Windows no longer
turns SQL migrations, scripts, manifests or sources into CRLF. A database
migrated by a Linux build is therefore no longer refused by a Windows or macOS
build of the same version because its migration checksums differ.