v1.1.1

v1.1.1 (2026-10-06)

🧹 Chores

• sync lockfile (b57ebcc)
• raise the dev dependencies, and drop .ncurc.yml (23af37a)

💬 General Changes

• turn off asyncErrorHandling, which pg has no counterpart for (016bd5f)

v1.1.1 (2026-10-06)

🧹 Chores

• sync lockfile (5db6df9)

v1.1.1 (2026-10-06)

🧹 Chores

• sync lockfile (b57ebcc)
• raise the dev dependencies, and drop .ncurc.yml (23af37a)

💬 General Changes

• turn off asyncErrorHandling, which pg has no counterpart for (016bd5f)

v1.1.1 (2026-10-06)

🧹 Chores

• sync lockfile (5db6df9)

v1.1.0

v1.1.0 (2026-09-29)

✨ Features

• choose the protocol before sending, not after being refused (ef1383f)

🐛 Bug Fixes

• measure memory in a process of its own, and count what is off the heap (250cf11)
• separate what a driver holds from what a call throws away (58276b3)
• bench: let the int4[] values be a dimension, not a choice (f6c0cf5)

📚 Documentation

• name 3.11.0, which is what the floor and the verification are now (38e7f30)
• drop the Status section (c807344)
• re-measure on postgrejs 3.11.0 (5f4b216)
• say where this benchmark and postgrejs's own differ (2c1cf5f)
• the accounting difference is closed (4ee3a7a)
• say why memory gets a process each and the timings do not (edb20ab)

📦 Build System

• generate the benchmark report instead of copying it across (f0aa0db)
• move to rman, github-actions@v3 and the shared preset (440a5ba)
• run rman build in the drizzle suite script, because npm run build is gone (189b4c2)

💬 General Changes

• three bytea sizes, so the payload advantage has a shape (462f099)
• drop the numeric scenarios (8b465e0)
• read every scenario's data from a table, and let the report say pg won (011a38c)
• re-measure on the uuid decode fix, which turns that row over (7170951)
• fold peak memory into each driver's cell (0431429)
• show the memory difference as a percentage, not a multiple (7ca37f2)
• disclose that pg and postgrejs are not always on the same protocol (0a0660c)
• bind a parameter everywhere, so both drivers speak the extended protocol (1d2b4a8)
• one size per type, grouped into reading and writing, with writes (2a46195)
• count what goes out, widen the write values, and say who wins each write (6292ad1)
• re-measure on the array-literal fix, which turns that write over (62101ae)
• make peak memory what one call needs, not what a batch leaves behind (d06aae3)
• say what the peak is, and report what each client keeps (7ee0aaa)
• the large writes hold a buffer, and it comes back (66219c8)
• say what the driver actually sends: text, not binary (c13b7dc)
• measure what a run needs, not just what a call adds (8a0a92e)
• move to postgrejs 3.12.0 and re-measure on it (b6ac46c)
• count the garbage, and say where counting it stops working (58b0ff2)
• say that the high-water is partly about the process around us (f57acfe)
• read the peak at the end of the call instead of hoping to sample it (ec40cb2)
• measure what a call allocates, because a per-call peak cannot be (8687195)
• say that the memory number is a rate, and what one call is (a085bb0)
• hold the same 5000 float8s in two shapes, and say each scenario's shape (a3b0e92)
• cut the method and reading sections, and group their rows by memory (5d97a36)
• re-run on postgrejs 3.12.1, which carries both fixes reported from here (6bd2b9b)
• the peer range resolves to 3.12.1 now, and both suites were re-run on it (2e9736e)

v1.1.0 (2026-09-29)

🧹 Chores

• sync lockfile (b57ebcc)

v1.1.0 (2026-09-29)

✨ Features

• choose the protocol before sending, not after being refused (ef1383f)

🐛 Bug Fixes

• measure memory in a process of its own, and count what is off the heap (250cf11)
• separate what a driver holds from what a call throws away (58276b3)
• bench: let the int4[] values be a dimension, not a choice (f6c0cf5)

📚 Documentation

• name 3.11.0, which is what the floor and the verification are now (38e7f30)
• drop the Status section (c807344)
• re-measure on postgrejs 3.11.0 (5f4b216)
• say where this benchmark and postgrejs's own differ (2c1cf5f)
• the accounting difference is closed (4ee3a7a)
• say why memory gets a process each and the timings do not (edb20ab)

📦 Build System

• generate the benchmark report instead of copying it across (f0aa0db)
• move to rman, github-actions@v3 and the shared preset (440a5ba)
• run rman build in the drizzle suite script, because npm run build is gone (189b4c2)

💬 General Changes

• three bytea sizes, so the payload advantage has a shape (462f099)
• drop the numeric scenarios (8b465e0)
• read every scenario's data from a table, and let the report say pg won (011a38c)
• re-measure on the uuid decode fix, which turns that row over (7170951)
• fold peak memory into each driver's cell (0431429)
• show the memory difference as a percentage, not a multiple (7ca37f2)
• disclose that pg and postgrejs are not always on the same protocol (0a0660c)
• bind a parameter everywhere, so both drivers speak the extended protocol (1d2b4a8)
• one size per type, grouped into reading and writing, with writes (2a46195)
• count what goes out, widen the write values, and say who wins each write (6292ad1)
• re-measure on the array-literal fix, which turns that write over (62101ae)
• make peak memory what one call needs, not what a batch leaves behind (d06aae3)
• say what the peak is, and report what each client keeps (7ee0aaa)
• the large writes hold a buffer, and it comes back (66219c8)
• say what the driver actually sends: text, not binary (c13b7dc)
• measure what a run needs, not just what a call adds (8a0a92e)
• move to postgrejs 3.12.0 and re-measure on it (b6ac46c)
• count the garbage, and say where counting it stops working (58b0ff2)
• say that the high-water is partly about the process around us (f57acfe)
• read the peak at the end of the call instead of hoping to sample it (ec40cb2)
• measure what a call allocates, because a per-call peak cannot be (8687195)
• say that the memory number is a rate, and what one call is (a085bb0)
• hold the same 5000 float8s in two shapes, and say each scenario's shape (a3b0e92)
• cut the method and reading sections, and group their rows by memory (5d97a36)
• re-run on postgrejs 3.12.1, which carries both fixes reported from here (6bd2b9b)
• the peer range resolves to 3.12.1 now, and both suites were re-run on it (2e9736e)

v1.1.0 (2026-09-29)

🧹 Chores

• sync lockfile (b57ebcc)

v1.0.4

v1.0.2

v1.0.1

v1.0.0