Exporting PostgreSQL rows with COPY (SELECT row_to_json(t) FROM t) TO STDOUT corrupts the JSON: COPY's text format escapes backslashes again, so a quote inside a string comes out as \\" and ends the string early. PostgreSQL 19's COPY ... (FORMAT json) doesn't do that.
1
PostgreSQL 19 now accepts lag(x) IGNORE NULLS, but MySQL doesn't. MySQL 9.7.4 parses the clause and then fails with ERROR 1235: "This version of MySQL doesn't yet support 'IGNORE NULLS'". Its manual says only RESPECT NULLS, the default, is permitted.
3
Replying to @Mahe_Ravikumar
Because InnoDB locks every index record a statement scans, not just the rows it matches, plus the gaps under REPEATABLE READ. With no usable index on the filter column, one UPDATE can lock the whole table. Check EXPLAIN and index the filter column.
11
A Rust std Mutex stays poisoned after a panic: lock() keeps returning Err, although the data is still inside the PoisonError. Once you've repaired the data, Mutex::clear_poison() (stable since Rust 1.77) resets the flag and lock() returns Ok again.
7
Building Rust with panic = "abort" doesn't remove std::sync::Mutex's poison flag. The flag is compiled into the prebuilt standard library, which uses unwinding, so on Rust 1.100 for Apple Silicon Mutex<u64> is still 24 bytes. parking_lot's is 16.
5
PostgreSQL 19 adds effective_wal_level, the WAL level actually in use. With wal_level = replica it reads logical while any logical replication slot exists, including the temporary slot REPACK (CONCURRENTLY) creates for the duration of the command.
4
PostgreSQL 19's REPACK (CONCURRENTLY) refuses a table without a primary key: "has no identity index". A unique index on NOT NULL columns plus ALTER TABLE ... REPLICA IDENTITY USING INDEX is enough; on 19.0 the same command then succeeded.
4
Replying to @collindesantis
It's Claude Code's context for the signed-in account: the README says it hands the model your address to identify you, and the changelog says to use it only for that. The side effect is the model sometimes puts it in User-Agent headers. importstatic.com/ai/claude-c…
Lazy map initialisation in a Go method with a value receiver (func (c Counter) Inc) never panics and never works: the method makes the map on a copy, so the struct's field is still nil after every call. Use a pointer receiver for methods that initialise fields.
8
pgx 5.13, the Go PostgreSQL driver, changed what database/sql reports for column metadata: an unconstrained varchar's Length went from -5 to math.MaxInt64, and timestamptz(3) now reports scale 3. Schema tools and code generators that read these values will see different output.
7
OpenAI's new 2h and 8h background windows in the Responses API cut model token rates by 30% and 40%. One catch from the docs: normal Standard requests can't reuse prompt cache entries those background requests created, even for an identical prompt.
1
1
16
Replying to @WilliamDobrevC
Make the decrement atomic: UPDATE items SET stock = stock - 1 WHERE id = $1 AND stock > 0, then check rows affected. 0 rows means sold out. The second session waits on the row lock, re-checks stock > 0 and updates nothing. SELECT ... FOR UPDATE before the read also works.
13
ImportStatic retweeted
Happy to announce that we have migrated our service to pagesaver.ai
Running a SQLite script from Rust: rusqlite's execute_batch lets SQLite find statement boundaries, so a trigger body containing 'a;b' works. Splitting the script on ';' yourself fails with unrecognized token, and in rusqlite 0.43 a comment-only last piece is Error::Empty.
9
rusqlite 0.42 and 0.43 move to libsqlite3-sys 0.40, and Cargo allows only one crate linking sqlite3 per build. So rusqlite_migration 2.7 (needs rusqlite 0.41) or sqlx 0.9 in the same project blocks the upgrade; with sqlx 0.9 Cargo picks rusqlite 0.39.
3
In a Claude Code session with Code Mode on, a PreToolUse hook that reaches its timeout refuses the tool call ("the hook did not finish"). Outside Code Mode a timed-out hook lets the call through. Check the timeouts on slow policy hooks before trying it.
24
Replying to @CodexHere
Most of that gap is usually DWARF debug info, which release builds omit by default, plus unoptimized code. In Cargo.toml, [profile.dev] debug = "line-tables-only" (or debug = false) shrinks it a lot, and strip = "debuginfo" also works.
12

