iiinova — where your digital self keeps evolving

Oliver Medhurst @honk.foo · 25/09/2026

this also introduces incremental compilation which can speedup builds by up to 30x versus last release!

warm compilation
react 17.45s → 0.56s (31x)
svelte 7.23s → 0.29s (25x)
preact 5.22s → 0.19s (27x)
solid 9.90s → 0.33s (30x)

Oliver Medhurst @honk.foo · 25/09/2026

the latest version of Porffor features ESM support! it not only removes needing a bundler beforehand but also makes compilation much faster as we can split C across modules

compilation time, selfhosted
react 17.31s → 4.16s (4.2x)
svelte 7.30s → 2.21s (3.3x)
preact 5.17s → 1.63s (3.2x)
solid 9.95s → 2.80s (3.6x)

4574

Oliver Medhurst @honk.foo · 18/09/2026

for fun: TSC compiled to native with Porffor! 25MB, starts in <5ms

honk@gosling tsc % bat -p demo.ts
type Config = { port: number };
const cfg: Config = { port: "8080" };
honk@gosling tsc % tsc --noEmit demo.ts
demo.ts:2:23 - error TS2322: Type 'string' is not assignable to type 'number'.

2 const cfg: Config = { port: "8080" };
~~~~

demo.ts:1:17
1 type Config = { port: number };
~~~~
The expected type comes from property 'port' which is declared here on type 'Config'

Found 1 error in demo.ts:2

honk@gosling tsc % du -h out/release/tsc
25M out/release/tsc
honk@gosling tsc % ./out/release/tsc --noEmit demo.ts
demo.ts:2:23 - error TS2322: Type 'string' is not assignable to type 'number'.

2 const cfg: Config = { port: "8080" };
~~~~

demo.ts:1:17
1 type Config = { port: number };
~~~~
The expected type comes from property 'port' which is declared here on type 'Config'

Found 1 error in demo.ts:2

7946

Oliver Medhurst @honk.foo · 15/07/2026

now state of the art btw because why not :)

A table:

method median error mean error words ≤300ms off ↓
autolrc + line anchors 32 ms 70 ms 97.2%
autolrc 34 ms 190 ms 95.8%
dafx (2025) 41 ms 216 ms 95.2%
kglw (2025) — 200 ms 94%
gupta et al. (2020) 50 ms 220 ms 94%
huang et al. (2022) — 230 ms 94%
hx-ih (2023) 43 ms 160 ms 93%
mstre-net (2021) — 310 ms 93%
spotify (2023) — 150 ms 92%
vaglio et al. (2020) — 370 ms 92%
hx-d (2023) 43 ms 420 ms 89%
stoller et al. (2019) — 820 ms 85%

0150

Oliver Medhurst @honk.foo · 08/07/2026

uses ~1mb memory btw ;)

080

Oliver Medhurst @honk.foo · 08/07/2026

Porffor is now self-hosted, meaning it compiles itself with itself! This is made possible by Porffor's new rewrite, reducing its lines of code by >50% and making C output up to 5x more efficient! Here you can see the stats before and after the rewrite for Porffor itself:

Porffor selfhost rewrite graphs, before vs after.

bundle lines 40.4K→24.7K (1.6x smaller)
bundle size 2.4MB→2.0MB (1.2x smaller)

C lines 1.6M→289.0K (5.5x smaller)
C size 70.4MB→12.3MB (5.7x smaller)

debug compile time 5.0s→1.2s (4.1x faster)
release compile time 143.2s→37.1s (3.9x faster)

debug binary size 52.4MB→13.6MB (3.8x smaller)
release binary size 14.1MB→4.6MB (3.1x smaller).

48910

Oliver Medhurst @honk.foo · 16/06/2026

but nan boxing wins for in-memory stuff! hence hybrid ;)

Scenario Nanbox Struct Winner
Sequential write numbers 0.031199 s 0.061291 s Nanbox, 1.96x faster
Sequential read numbers 0.125784 s 0.148851 s Nanbox, 1.18x faster
Sequential read + write numbers 0.030615 s 0.061146 s Nanbox, 2.00x faster
Sequential write mixed values 0.073787 s 0.108784 s Nanbox, 1.47x faster
Sequential read mixed values 0.151721 s 0.156975 s Nanbox, 1.03x faster
Type scan mixed values 0.039022 s 0.057213 s Nanbox, 1.47x faster
Copy array 0.045452 s 0.093518 s Nanbox, 2.06x faster
Random read numbers 0.774604 s 0.908102 s Nanbox, 1.17x faster
Random read mixed values 0.934132 s 1.299783 s Nanbox, 1.39x faster
Random read + write numbers 0.811805 s 0.914708 s Nanbox, 1.13x faster

020

Oliver Medhurst @honk.foo · 11/06/2026

Map and Set are now up to over 300x faster :)

scenario (n=20k) old wasm new wasm Δ old native new native Δ
insert string keys 933ms 8ms 117× 442ms 4ms 111×
lookup string keys 9153ms* 29ms 316× 3141ms* 22ms 143×
int keys 3532ms* 41ms 86× 1412ms* 26ms 54×
object keys 3524ms* 17ms 207× 1445ms* 10ms 145×
set dedup 476ms* 6ms 79× 244ms* 3ms 81×
delete/add churn 4051ms* 11ms 368× 2030ms* 8ms 254×

2280

Oliver Medhurst @honk.foo · 10/06/2026

(very!) early preview for a new experimental threading Porffor api. it enables intentionally low level concurrency to (try to) enable the most efficient communication feasible. here are 1 million threads alive simultaneously using only ~16gb. 16kb/thread, <1us start avg.

A graph with RSS as Y axis and threads as X axis. It increases linearly bottom left to top right. Footer shows: "1.00M threads · 15.9 GB · 16 KB/thread · 0.7 µs/start"

0432

Oliver Medhurst @honk.foo · 10/06/2026

cleanup grind continues

[5.48s] compiled selfhosted/porf
bundle 37.44K lines (-0.10%) 2.17MB (-0.08%)
c 1.639M lines (-0.16%) 72.84MB (-0.16%)
binary 53.75MB (-0.11%)

080

Oliver Medhurst @honk.foo · 09/06/2026

optimized selfhosted porffor binary is now <20mb and only takes ~1 minute to compile (>90% clang). (% change is vs debug binary) still buggy though. also still too big ;)

1210

Oliver Medhurst @honk.foo · 09/06/2026

self compilation is now ~15% faster after some optimizations for 2c itself, and emits a smaller (unoptimized debug!) binary

[5.39s] compiled selfhosted/porf
bundle 37.51K lines (+0.15%) 2.18MB (+0.09%)
c 1.644M lines (-0.58%) 73.04MB (-0.47%)
binary 53.86MB (-0.70%)

0180

Oliver Medhurst @honk.foo · 09/06/2026

after more 2c optimizations

[6.18s] compiled selfhosted/porf
bundle 37.46K lines (+0.04%) 2.18MB (-0.02%)
c 1.653M lines (-17.40%) 73.38MB (-17.76%)
binary 54.24MB (-12.32%)

090

Oliver Medhurst @honk.foo · 09/06/2026

more (vs same snapshot)

[6.67s] compiled selfhosted/porf
bundle 37.31K lines (+0.07%) 2.18MB (+0.08%)
c 2.024M lines (-8.16%) 90.01MB (-5.90%)
binary 62.27MB (-7.90%)

050

Oliver Medhurst @honk.foo · 09/06/2026

cleaning up code directly correlates to smaller (unoptimized!) selfhosted binary size 😇 (the ~release binary size is much much smaller)

A terminal screenshot:

[7.07s] compiled selfhosted/porf
bundle 37.42K lines (-0.21K) 2.18MB (-3.2KB)
c 2.204M lines (-6.06K) 95.66MB (-0.26MB)
binary 67.61MB (-0.2MB)

1140

Oliver Medhurst @honk.foo · 08/06/2026

elysia uses >2x less memory with ~15% lower latency and higher throughput vs ~a week ago :)

./elysia_new
Latency #1 median 0.377 ms, p90 0.525 ms, p99 0.909 ms
Throughput (req/s) #1 overall 118411.4, p10 110169.6, p1 69386.1

Memory usage #1 peak 7.80 MB, median 6.91 MB
CPU time / req #1 median 8.26 µs, p90 8.51 µs, p99 8.61 µs
Binary size #2 6.07 MB
Startup #1 7.630 ms

./elysia_old
Latency #2 median 0.443 ms, p90 0.528 ms, p99 0.942 ms
Throughput (req/s) #2 overall 104255.5, p10 95517.0, p1 67389.8

Memory usage #2 peak 16.50 MB, median 16.48 MB
CPU time / req #2 median 9.46 µs, p90 9.60 µs, p99 9.62 µs
Binary size #1 5.80 MB
Startup #2 21.27 ms

1390

Oliver Medhurst @honk.foo · 07/06/2026

fun data from optimizing how builtin apis are embedded (they are precompiled). numbers are still unoptimized/not final! oh and btw porffor can selfhost now ;)

a markdown table:

precompiled.js size
json (before): 3.05MB
huffman: 1.30MB
change: -57.2%
array interning: 1.17MB
change: -10.5%
final diff: -61.7%

selfhost bundle size
json (before): 4.15MB
huffman: 2.40MB
change: -42.0%
array interning: 2.29MB
change: -4.9%
final diff: -44.8%

selfhost binary size
json (before): 34.28MB
huffman: 33.22MB
change: -3.1%
array interning: 28.12MB
change: -15.4%
final diff: -18.0%

selfhost compiling globalThis time
json (before): 725.5ms
huffman: 601.5ms
change: -17.1%
array interning: 574.3ms
change: -4.5%
final diff: -20.8%

selfhost compiling globalThis rss
json (before): 233.7MB
huffman: 144.1MB
change: -38.3%
array interning: 141.4MB
change: -1.9%
final diff: -39.5%

1240

Oliver Medhurst @honk.foo · 05/06/2026

hono uses ~10% less memory with ~5% lower latency vs a few days ago, thanks to new gc and other allocation optimizations

./hono_new
Latency #1 median 0.417 ms, p90 0.451 ms, p99 0.867 ms
App #1 median 0.003 ms, p90 0.003 ms, p99 0.004 ms
Throughput (req/s) #1 overall 115719.0, p10 109517.8, p1 82189.2
Memory usage #1 peak 3.58 MB, median 3.56 MB
CPU time / req #1 median 8.58 µs, p90 8.61 µs, p99 8.62 µs
Binary size #2 731.1 KB
Startup #1 7.322 ms

./hono_old
Latency #2 median 0.433 ms, p90 0.471 ms, p99 0.898 ms
App #1 median 0.003 ms, p90 0.003 ms, p99 0.004 ms
Throughput (req/s) #2 overall 111291.0, p10 105300.5, p1 75054.0
Memory usage #2 peak 3.91 MB, median 3.89 MB
CPU time / req #2 median 8.90 µs, p90 9.11 µs, p99 9.23 µs
Binary size #1 650.4 KB
Startup #2 7.486 ms

0551

Oliver Medhurst @honk.foo · 05/06/2026

Porffor's optimization phase is now 5-10x faster for large files

selfhost old new diff
---------------- ---------- ---------- --------------
parse 70 70 same
codegen 1591 1576 1.01x faster
opt 4219 612 6.90x faster
assemble 269 269 1.00x slower
2c 2116 1626 1.30x faster
total 8449 4330 1.95x faster

2290

Oliver Medhurst @honk.foo · 04/06/2026

forgot I added some debug logs um

A terminal screenshot with "compareStrings ToString b type=7" repeated.. a lot

1140

Oliver Medhurst @honk.foo · 04/06/2026

if you want a more traditional/detailed profiler I got you too ;)

0180

Oliver Medhurst @honk.foo · 04/06/2026

new experiment

$A terminal screenshot showing: hot: 5 lines in scheduler_schedule account for 27% of runtime --> bench/richards.js:180 | 178 | Scheduler.prototype.schedule = function scheduler_schedule() { 179 | this.currentTcb = this.list; 180 | while (this.currentTcb != null) { | ^^^^^^^^^^^^^^^ 3% of runtime 181 | if (this.currentTcb.isHeldOrSuspended()) { | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 10% of runtime 182 | this.currentTcb = this.currentTcb.link; | ^^^^^^^^^^^^^^^^^^^^ 3% of runtime 183 | } else { 184 | this.currentId = this.currentTcb.id; | ^^^^^^^^^^^^^^^^^^ 4% of runtime 185 | this.currentTcb = this.currentTcb.run(); | ^^^^^^^^^^^^^^^^^^^^^ 6% of runtime 186 | } 187 | } | = note: object lookup accounts for 81% of these lines' time$

2323

Oliver Medhurst @honk.foo · 15/05/2026

i guess everyone is making code review tooling now. idk if I'll release this yet

A screenshot of a code review app. It shows a diff center stage with a file sidebar and a header saying unstaged changes with a line diff viewable. There is also a refresh button and a sparkles button.

3190

Oliver Medhurst @honk.foo · 14/05/2026

rewrote 2c and now it emits ~2x less lines of c :)

170

Oliver Medhurst @honk.foo · 14/05/2026

learning to get less attached to old code. you can just delete things

2760

Oliver Medhurst @honk.foo · 08/05/2026

enjoy 2 new blog posts! - Porffor now has closures: surprisingly annoying for a compiler. goose.icu/porffor-clos... - Profile-guided dead code elimination: a novel approach for DCE in dynamic languages. goose.icu/profile-guid...