All
Hardware
Maps
ML & AI
Music
Social
Tools
Writing
Apps & Games
Deciduous
Decision graph CLI for tracing how software decisions evolve
Deciduous Archaeology Demo
Demonstrations of decision archaeology on React and stacked git workflows
Phish Explorer
Jam analytics dashboard for Phish 3.0 era
Local LLM on MacBook
4-bit quantization, safetensors, and Bumblebee + EMLX for Apple Silicon
A Bot that posts like me
Porting the posting bot to Elixir using local LLM work
Receipt Printer Software Suite
A complete software suite for thermal receipt printers
Role Call
TV writer overlap explorer
Fill Your Sky
Interactive map of 418+ Bluesky communities with 545K+ people
Code Mirror
A live code mirror experiment
Pocket Pnin
A local LLM running on my iPhone, coming to the App Store for free
NYC Census Maps
Interactive census and PLUTO data visualization for New York City
MTA Bus Tracker
Real-time MTA bus and train tracking on an interactive map
Concert GIF Maker
Extract GIFs from concert videos with a retro Mac interface
Send a VERY direct message, to my receipt printer
A social project where friends send photos that print on my receipt printer
Archive TV
A real over-the-air TV channel from magnetic media archives
Losselot
Neural network loss function explorer
Todoinksies
A personal todo app
Ormery
An ORM written in Temper
Collage Maker
Upload photos and arrange them into grid collages
GenStage Tutorial 2025
A modern GenStage tutorial for the Elixir ecosystem
Temper Rust Bug
Found a bug in the Temper compiler and built a demo repo
300+ Years of Tree Law
A blog post that became its own LiveView application
HEEx in Other Languages
Experiments porting Phoenix HEEx templates to Rust, Lua, C#, Java, Python, and JS
Live Draft LSP
Live-stream blog drafts from Zed to Phoenix via a custom LSP
Bluesky Hoover Apps
Various apps that vacuum up and process the Bluesky firehose
Bobby Posts Bot
A bot that posts like me, in Python
Photo Booth Receipt Printer
A portable photo booth that prints on receipt paper
Nathan For Us
A Nathan For You social network with video search and GIF creation
Browser History Roast MCP
An MCP server that roasts you based on your browser history
GenStage Tutorial (Original)
The original GenStage tutorial
Bluesky Firehose Toys
Real-time firehose visualizations: emoji streams, jetstream comparisons, and more
+14-3 bglusman/zeroclawed/feat/security-profiles Apr 10 14:24
6c3199d ci: align job names with required status check contexts
Ruleset requires contexts named fmt-and-clippy, test, and build, but
the workflow was reporting them as 'Format & Clippy', 'Test Suite (...)',
and 'Release Build'. Drop the display names so jobs report under their
keys, and add a 'test' aggregator that fans in from the matrix so a
single 'test' context reports.
+700-204 notactuallytreyanastasio/deciduous/main Apr 08 11:28
c06a5da chore: update deciduous tooling to v0.14.0
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+188-51 notactuallytreyanastasio/deciduous/main Apr 08 11:26
236c8c5 release: v0.14.0 - observation title + description support
Observations now have first-class title + description semantics:
- CLI warns when creating observations without -d
- `deciduous nodes` shows description inline for observations
- Web viewer detail panel shows description prominently below title
- All templates, commands, and instructions updated
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+49-12 notactuallytreyanastasio/elixir/add-hole-operator Apr 07 01:04
f24f307 Rewrite system prompt with detailed Expert context guidance
The system prompt now explains:
- Every context section Claude receives and how to use it
- That callers tell you the expected return shape
- That sibling modules show available APIs — use them, don't invent
- That functions must actually exist (no hallucinating APIs)
- To write inline logic when needed functionality doesn't exist
- Session continuity across multiple holes in one compile
- That generated code gets written to source for developer review
This directly addresses Claude hallucinating nonexistent functions
like Blog.Visitors.known_ip?/1 by explicitly instructing it to
only call functions visible in the Expert context.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+63-17 notactuallytreyanastasio/elixir/add-hole-operator Apr 07 00:55
2bc621b Add automatic retry on compile errors with error feedback to Claude
When generated code fails to expand/compile, the error is captured
and sent back to Claude: "The previous code had this compile error:
[error]. The code that failed: [code]. Please fix it."
Uses the existing session continuity (--continue) so Claude has full
context from the original intent + its previous attempt. Retries up
to situation_max_retries times (default: 3) before giving up.
Output shows the attempt number and error at each retry:
│ Compile error (attempt 1/4):
│ undefined function foo/1
│ Asking Claude to fix...
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+20-1 notactuallytreyanastasio/elixir/add-hole-operator Apr 07 00:52
af3498c Strip markdown fences from Claude responses before parsing
Claude sometimes wraps code in ```elixir ... ``` fences despite
system prompt instructions. The parser now strips these before
attempting to parse the response as Elixir code.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+17-4 notactuallytreyanastasio/elixir/add-hole-operator Apr 06 23:58
6996fc6 Filter Kernel and SpecialForms from situation prompt imports
These are auto-imported in every module and just add noise to the
prompt. Claude already knows about Kernel functions.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+32-1 notactuallytreyanastasio/elixir/add-hole-operator Apr 06 23:45
a63639d Add situation_verbose for debug logging of prompts and responses
When situation_verbose: true is set, the compiler prints the full
system prompt, user prompt (with all Expert context), command details,
and raw Claude response for each hole. Off by default.
Set via mix.exs:
elixirc_options: [situation_command: "claude", situation_verbose: true]
Or at runtime:
Code.put_compiler_option(:situation_verbose, true)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+120-38 notactuallytreyanastasio/elixir/add-hole-operator Apr 06 23:15
ce1f489 Enrich Expert LS context and add Claude session continuity
Expert context enrichment — when Expert LS engine is running, each hole
now receives:
- Callers of the current function (what callers expect)
- Function definition/spec (the stated contract)
- Module attributes (@moduledoc, @type, etc)
- Struct definitions in the module
- Sibling modules in the same namespace
- Related ExUnit tests
All queries use safe_expert_call which returns [] on failure, so
compilation never breaks if Expert isn't running.
Claude session continuity — the first hole in a compile run starts a
fresh Claude session. Subsequent holes use --continue to resume the
same session, so Claude naturally accumulates context from previously
filled holes. One Claude "brain" across the entire compilation.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+15719-180 notactuallytreyanastasio/blimp/main Mar 30 18:59
fc29d3c fix: pipes lesson and predict validator
Pipes with higher-order fns (filter/map) broken in pipe evaluator.
Changed lesson to sort/reverse/length which work. Predict lessons
no longer show Correct when the eval itself errors.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ba0e10f feat: Blimp lexer written in Blimp -- first self-hosting step
Added char_at, char_code, from_char_code builtins for char-level
string scanning. The lexer (lib/lexer.blimp) tokenizes: integers,
floats, strings, atoms, identifiers, upper identifiers, keywords,
all operators (+, -, *, /, ==, !=, <-, |>, ::, .., ..., ->),
comments, and whitespace. 6/10 tests pass (multiline and keyword
matching need the string case-match fix).
370 lines of pure Blimp. The compiler is starting to eat itself.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
3b7a55b feat: rewrite tutorial -- views by lesson 6, live preview pane
20 lessons restructured: values (3) -> variables/functions (2) ->
views (5, with live rendering!) -> actors (3) -> map/filter/pipes (3)
-> putting it all together (4). View lessons show rendered output
in a light preview pane. Tab inserts spaces, Cmd+Enter submits,
progress saved to localStorage.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
17e3527 feat: interactive Blimp tutorial with runtime eval/test
blimp_eval(code) and blimp_test(code) builtins for runtime
code evaluation. 12-lesson tutorial: values, functions, actors,
builtins. Three step types: predict output, fill blanks, write code.
Dark theme, mobile-friendly (viewport meta, 44px tap targets,
responsive breakpoint at 480px).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
8b2911d ship: rebuild WASM (182KB), update tutorial + manifesto
WASM rebuilt with all fixes: and/or operators, pipes with
filter/map/reduce, Any type checks, <-- async sends, auto-scheduler,
cons operator, self-hosted compiler.
Manifesto pipe example now uses filter |> map |> sort |> reverse.
Tutorial reference pane adds and/or, <-, <-- operators.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
6ba6da7 fix: Functions lesson no longer loops -- merged into one check
The 'then' hint showed 'Now try: double(21)' but the next
eval matched the same lesson's check (double still in env),
creating an infinite loop. Merged into one lesson that checks
for double(21) == 42.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
785a030 feat: self-hosted parser in Blimp -- 11/11 tests pass
Recursive descent parser (699 lines) that produces map-based AST:
- Actor definitions with dotted names, state, handlers, tests
- Function definitions with typed params and return types
- Expressions: arithmetic, comparisons, function calls, message
sends, list literals, case expressions, spawn, assignment
- Operator precedence: comparison > addition > multiplication
Combined with lexer (429 lines): 1128 lines of pure Blimp that
can tokenize and parse Blimp source code into an AST.
Also fixed: string literal lexing uses char code 34 instead of
escaped quote in case branch, scan_string uses char codes.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ef9b0c2 feat: self-hosted evaluator in Blimp -- 10/11 tests pass
Tree-walking evaluator (480 lines) that executes map-based AST:
- Literals: int, float, string, bool, nil, atom
- Arithmetic: +, -, *, / with correct precedence
- Comparisons: ==, !=, <, >, <=, >=
- Variables: assignment and lookup via Env actor
- Functions: def, call, params, closures over env
- Builtins: 25+ delegated to host (print, concat, length, etc.)
- Lists: literal construction and operations
- Case expressions: pattern matching with wildcard
- Actors: definition, spawn, message send (1 test remaining)
Combined self-hosted compiler: lexer (429) + parser (699) +
evaluator (480) = 1608 lines of pure Blimp.
Also fixed: checker allows Any-typed vars as message send targets.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
e62b8d4 fix: strip quotes from WASM string results in validators
WASM eval returns strings as "hello" (with quotes) but validators
compared against hello (without). Added unquote() helper. Fixed
Greeter lesson validator.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
4c80350 feat: Blimp Chat -- first real web app, fully working
Multi-user chat room with:
- ChatRoom actor: messages, users, join/leave/send
- Cookie-based sessions with auto-generated tokens
- Login page, chat page with sidebar (online users) + message list
- Message send via POST, redirect-after-POST pattern
- Dark theme, mobile-friendly, auto-refresh every 3s
- Multiple simultaneous users see each other's messages
Fixed: sequential become statements (only last takes effect),
slice_list using elem() instead of head/tail in case branches,
Any-typed actor ref passing through function chains.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
c57724f fix: chat room fully working -- the 'bug' was a corrupted binary
The Zig evaluator is fine. The elem()/head()/tail() calls all work
correctly at any call depth. The crash was caused by a corrupted
binary from sed-based debug print removal that left the builtins.zig
in an inconsistent state.
Clean rebuild fixes everything. The full chat flow works:
- Login (POST /login, cookie, session)
- Chat rendering (messages from actor state via head/tail)
- Message sending (POST /send)
- Multi-user (Bob sees Alice's messages)
- WebSocket upgrade + broadcast
Used head/tail recursion instead of elem/idx for rendering
(simpler and avoids the index-based patterns).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
57205b0 fix: form falls back to HTTP POST when WebSocket not connected
e.preventDefault() only fires when WS is ready. Otherwise the
form submits normally via POST. Messages always send now.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
65988cd feat: tutorial IS the REPL -- guided lessons in the REPL stream
Complete redesign: the tutorial is the REPL itself with lesson
guidance. Lessons appear as styled text in the REPL output, you
type at the blimp> prompt, results show inline, state sidebar
updates live, views render in the view pane.
18 lessons: values -> strings -> atoms -> booleans -> variables ->
lists -> functions -> views -> actors -> messages -> map/filter ->
free play. /lessons restarts, /skip advances.
Same REPL as index.html: history, multi-line (shift+enter),
view rendering, state sidebar, mounted actor views.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
fdc1689 feat: --self-hosted flag -- Zig bootstraps, Blimp runs
./blimp file.blimp --self-hosted
The Zig runtime loads lib/*.blimp (lexer, parser, evaluator,
codegen, completion, compiler), then executes user code through
the self-hosted pipeline. The Zig lexer/parser/evaluator become
the bootstrap layer only.
./blimp expr.blimp --self-hosted # eval via Blimp compiler
./blimp expr.blimp # eval via Zig compiler (default)
New builtins: write_bytes(path, bytes), read_file(path).
compiler.blimp provides blimp_compile(source, :eval/:wasm/:tokens/:ast).
The Zig is now the kernel: builtins + WASM bridge + CLI.
Everything above it is Blimp.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
fcd258c fix: remove Google Fonts import that blocks page rendering
3841a17 feat: auto-scheduling -- no manual schedule() needed
<-- auto-drains mailbox immediately after enqueue. Reduction
counter also triggers drain every 4000 evals. No explicit
schedule() call required.
w <-- :compute(5) # enqueued AND processed
print(w <- :get) # 25 -- already done
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
949a13a feat: WebSocket broadcast chat -- zero Any types, typed actors
Full rewrite with proper types:
- ChatRoom, Sessions, Connections, HTTPServer all actor-typed
- No Any types anywhere -- every param is ChatRoom, Sessions, etc.
- WebSocket: upgrade handshake, ws_loop, broadcast to all clients
- Connections actor tracks open WS file descriptors
- broadcast() pushes rendered HTML to ALL connected clients
- JSON builder with proper escaping (no string concat hacks)
- Client JS: WebSocket with reconnect, form intercept, DOM update
WS upgrade confirmed working (101 + state push).
HTTP login + cookie confirmed working (302 + Set-Cookie).
Remaining: render_messages crashes on map lookup from actor state
(likely a scoping issue with lookup on maps from become).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
b6a40a0 feat: BOOTSTRAP -- Blimp compiles Blimp end-to-end, 56/56 tests
The self-hosted compiler compiles programs with: functions with
return type annotations, && and || boolean operators, case
expressions, nested function calls, list operations, string
builtins, comparisons, and multi-line programs.
Fixed: self-hosted lexer now tokenizes &&. Self-hosted parser
handles or/and/|| /&& in the precedence chain. Self-hosted
evaluator executes && and ||. Parser case-branch scoping fixed
(assignments in case branches don't leak to outer scope).
Self-hosted Blimp: 2186 lines across 5 components.
Zig bootstrap: 19101 lines. Compression ratio: 8.7x.
429 lib/lexer.blimp
752 lib/parser.blimp
479 lib/eval.blimp
69 lib/complete.blimp
457 lib/tests (lexer+parser+eval+bootstrap+complete)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
396f630 fix: enforce type annotations on ALL parameters, no exceptions
Lambda params now require types: fn(x: Int) not fn(x).
Checker traverses all function call arguments to catch lambdas
passed to map/filter/reduce/each. Complete audit:
on :msg(x: Int) -- enforced (existing)
def foo(x: Int) -- enforced (previous commit)
state x: Int :: 0 -- enforced (existing)
fn(x: Int) do end -- enforced (this commit)
The only untyped bindings are assignment (x = 42, inferred)
and for loops (for x in list, inferred from iterable).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
f9ac5f7 feat: autocomplete shows automatically as you type
Completions appear after 2+ characters, updating on every
keystroke. No need to hit Tab first. Exact matches are
suppressed. Menu stays open while typing and updates live.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
54bff04 fix: slice(s, start, len) uses length not end index -- 10/10 lexer tests pass
The slice builtin treated arg 3 as end index but the lexer
(and the docs) use it as length. Fixed to start+len semantics.
This was the root cause of all 4 failing lexer tests -- keyword
matching, newlines, comments, and actor definitions all used
slice to extract substrings.
Also: nil-safe comparisons in evaluator, nil-comparable in
checker, char_code returns nil for nil input.
The self-hosted Blimp lexer (370 lines) now passes all 10 tests.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
1f7478c feat: the Blimp manifesto -- full language vision with live code
Long-form page covering every premise: values, functions, pipes,
actors, become, message passing, dot notation supervision, views
as message responses, the web framework in Blimp, the Hole operator,
the test runner, runtime internals (arenas, scheduling, LLVM),
WASM in the browser, the canvas, and the self-hosting dream.
Every code example has a Run button that evals in WASM. Actor
examples have the canvas showing hexagons and message rays.
Written in the voice: casual, direct, confident about taste.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
09b06e5 feat: secure string composition -- SafeHtml, SafeSql, context automaton
Implements the accumulator pattern from the paper:
- SafeHtmlAccumulator: tracks HTML parse context through fixed parts,
applies context-sensitive escaping for each interpolation
- SafeSqlBuilder: parameterizes user input ($1, $2...), never interpolates
- HTML context automaton: pcdata/tag/attr_name/attr_value/url states
- Context-sensitive escapers: html_escape, html_attr_escape, url_escape
- Escaper selection based on current parse context (href -> url_escape)
- Reified safe types: %{__safe_html__: true, content: "..."}
22/34 tests pass. Security-critical tests all pass:
- XSS prevention: <script> tags escaped in pcdata
- URL injection: javascript: URLs escaped in href attributes
- SQL injection: user input parameterized, never interpolated
- Safe text passes through unescaped
Context automaton tests have a scoping issue (12 failures).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
a7c6b52 fix: actor definition no longer auto-spawns
actor Foo do ... end registers the template but does NOT spawn.
First message send (Foo <- :msg) auto-spawns the singleton
and replaces the atom binding with an actor_ref. spawn Foo
creates additional instances.
Canvas only shows actors that are actually alive, not templates.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
0001b87 fix: chat deploys cleanly -- WS fallback, no server crashes
- WS upgrade returns 400 (disabled for now, Linux WS builtins
need debugging separately)
- JS only attempts WS connection on chat page, not login page
- Auto-refresh fallback every 2s when WS unavailable
- Server no longer crashes on WS upgrade attempts
Tested on production: login, session, chat, message send all work.
Server stays up across multiple requests.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
63dac0f feat: actor scheduler -- mailbox, reduction counter, async sends
The scheduler infrastructure:
- Mailbox: per-actor FIFO message queue (mailbox.zig)
- Scheduler: run queue with round-robin processing (scheduler.zig)
- Reduction counter: every eval() costs 1 reduction (4M default)
- ActorEntry gets mailbox + reductions fields
New builtins:
- send_async(actor, :msg, args...) -- enqueue without blocking
- schedule(ticks) -- process queued messages in round-robin
The scheduler enables concurrent actor execution:
send_async(w1, :compute, 5)
send_async(w2, :compute, 10)
schedule(100) # both workers process their messages
Sync sends (<-) still work as before for backward compat.
All existing tests pass.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
9526197 feat: property-based testing with generators
New syntax inside actors:
property "name" do
given x: gen_integer(0, 100)
given xs: gen_list(10, 0, 50)
assert_eq(reverse(reverse(xs)), xs)
end
Each property runs 100 times with fresh random inputs.
Generators:
gen_integer(min, max) random int in range
gen_string(max_len) random printable ASCII
gen_boolean() random true/false
gen_list(max_len) random list of ints
gen_list(max_len, lo, hi) random list with range
gen_one_of([a, b, c]) random element from list
Parser: property/given keywords, PropertyDef/GivenStmt AST nodes
Evaluator: property runner (100 iterations), given binds generated
value in scope per iteration
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
eaa72de feat: Code/Output tabs on every lesson
Two tabs: Code (editor + Run button) and Output (canvas with
rendered result). Run auto-switches to Output tab. Tap back
to Code to edit. Visible on every lesson type.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
4ea148c fix: Any type accepted in runtime type checks
checkType("Any") now returns true for all values. Previously
def foo(a: Any) followed by a <- :msg crashed because the
runtime type checker compared ActorRef.type_name with "Any"
and found no match.
All 4 bugs from the audit are now fixed:
1. and/or as expression operators -- fixed
2. Pipes with map/filter/reduce -- fixed
3. Type checker scope leak -- fixed (was the and/or bug)
4. Any-typed params in evaluator -- fixed (this commit)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
9c6c3a0 feat: WebSocket working in production -- graceful dead fd handling
Fixed: ws_write_frame and tcp_write return :error instead of
crashing on dead file descriptors. broadcast_fds auto-removes
dead connections. Server no longer crashes when browsers
disconnect.
WS confirmed working on Linux production server:
- Upgrade handshake (101 Switching Protocols)
- Initial state push (JSON with rendered messages + users)
- Dead fd cleanup on broadcast
- Server stays active after WS disconnect
Live at http://5.161.181.91:8080
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
c51bbe3 feat: Tab autocomplete in the tutorial REPL
Tab triggers completion dropdown from blimp.complete(prefix).
Shows function names with signatures, variables, actors, builtins.
Arrow keys navigate, Enter/Tab accepts, Escape dismisses.
Dropdown positioned above the input, scrollable, styled to match.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
16017bb feat: self-hosted Blimp compiler -- 32/32 tests pass
Fixed three bugs:
1. Cons operator [h | tail] dropped the tail (evalList ignored
the tail field). Now correctly concatenates head + tail.
2. Any-typed params couldn't send messages (checker rejected them
as non-actor types). Fixed resolveActorName to accept Any/hole.
3. Test runner first-pass errors now logged instead of silent.
The self-hosted compiler (1608 lines of pure Blimp) passes all
32 tests: lexer (10), parser (11), evaluator (11).
Blimp can tokenize, parse, and evaluate itself.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
d1a3c48 feat: Blimp stdlib -- 185 lines replacing Zig builtins
Pure Blimp implementations of:
- Math: abs, max, min
- Lists: sum, range, flat, uniq, list_contains, zip, set_at
- Strings: upcase, downcase (char-by-char transformation)
- Views: text, heading, bold, italic, code_inline, divider,
stack_view, row_view (as pure map constructors)
empty?, nil?, not stay as Zig primitives (can't be self-referential).
51 tests (44 unit + 7 property). All green.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
19c0f35 feat: reference cheat sheet pane next to state sidebar
Right column has State/Reference tabs. Reference pane covers
the full stdlib: math, strings, lists, higher-order (map/filter/
reduce), maps, views, actors, control flow, utility. Compact
format with green function names and gray descriptions.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
01ccc98 feat: bootstrap loop closed -- 44/44 tests, Blimp compiles Blimp
blimp_self_eval("source") runs the full self-hosted pipeline:
source string -> Blimp lexer -> Blimp parser -> Blimp evaluator -> result
12 bootstrap tests: integers, arithmetic with precedence, strings,
variables, function def/call, nested calls, builtins, comparisons,
case expressions, list ops, multi-line programs. All pass.
Full suite: lexer (10) + parser (11) + evaluator (11) + bootstrap (12) = 44
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
3f4844c fix: rebuild WASM with all new builtins, gate native-only code
Native-only functions (TCP, fork, WS, file IO) are stubbed on WASM
via comptime dispatch. Hole operator returns nil on WASM. Assert
builtins skip stderr on WASM. WASM binary: 178KB (was 131KB).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
612f934 feat: WASM-powered interactive Blimp tutorial at /tutorial/
14 lessons running entirely client-side via blimp.wasm (130KB).
Predict output, fill blanks, write code -- all evaluated in-browser.
Curriculum: values, variables, functions, actors, builtins, pipes.
Dark theme, mobile-friendly, progress bar.
Deploy: rsync docs/ to server, lives at blimp.bobbby.online/tutorial/
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ad0a059 fix: autocomplete now includes all view primitives + new builtins
Added to completion engine: stack, row, grid, text, heading, bold,
italic, code, code_block, blockquote, divider, list, link, image,
button, input, textarea, select, form, mount_root, char_at,
char_code, from_char_code, set_at, to_atom, actor_name, assert,
assert_eq, assert_ne, refute, blimp_eval, blimp_test.
Added keywords: test, and, or, not.
'sta' now shows both 'state' and 'stack(children...) -> ViewNode'.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
b85fa45 test: 75 property tests across the entire compiler -- 7,500 random cases
Properties cover:
- Arithmetic: commutativity, associativity, identity, zero, negation
- Comparison: complementarity, reflexivity, negation
- Boolean: identity, absorption, De Morgan's laws
- Lists: reverse/reverse, sort idempotent, append length, cons,
sum linearity, uniq idempotent, range length, elem/head
- Strings: concat identity/length, slice identity, char roundtrip,
to_string/to_int roundtrip, upcase/downcase
- Maps: put/lookup, overwrite, keys/values length, merge identity
- Higher-order: map length, filter length, filter complement,
map identity, reduce=sum, map compose, filter idempotent
- Self-hosted compiler: lexer tokenizes random ints, self-eval of
random arithmetic (add/sub/mul), comparisons, variables
- Generators: range bounds, length bounds, type correctness
- Actors: cumulative state, reset, spawn independence
Total suite: 698 tests (623 unit + 75 property) = 15,123 test cases
(623 + 75*100). All green in 5.6 seconds.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
0b96330 fix: no auto-spawn on message send -- must explicitly spawn
actor Foo do ... end registers a template. Foo <- :msg is now
an error: "Message send target is not an actor." You must:
f = spawn Foo
f <- :msg
Fixed manifesto examples, dot_notation example. All tests pass.
This is the correct semantic: define is not spawn.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
fca8a7c feat: completion engine in Blimp -- replaces complete.zig logic
69 lines of Blimp replaces the completion matching logic.
Keywords and builtin names derived from lists, prefix matching
via starts_with. 8/8 tests pass, 52/52 total.
Self-hosted Blimp: 1674 lines (lexer 429 + parser 699 +
evaluator 477 + completion 69). Tests: 432 lines, 52 passing.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
bc6b4fb fix: nil-safe comparisons, type checker allows nil in ==/</>
char_code returns nil for nil input instead of crashing.
Comparisons (< > <= >=) return false when either side is nil.
Type checker treats nil as comparable with any type.
Lexer nil guards in is_digit, is_alpha, is_alnum, scan functions.
Lexer tests: 6/10 pass (multiline tests need further debugging
in the type checker's interaction with case-match bodies).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
26adbc0 feat: <-- operator for async message sends
actor <- :msg is sync (blocks until reply)
actor <-- :msg is async (enqueues in mailbox, returns :queued)
New token: async_send (<--), parsed as MessageSend with is_async.
Lexer checks 3-char <-- before 2-char <-. Evaluator enqueues in
mailbox instead of executing inline. schedule(N) drains mailboxes.
w <-- :compute(5) # queued, returns :queued
schedule(100) # processes all queued messages
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
e1a14b2 feat: THE canvas -- generative hex actors, message rays, REPL blob
Added the real BlimpCanvas from web/canvas.js to the tutorial.
Actors render as generative-filled hexagons with SHA-hashed
patterns. Message sends shoot animated rays from the REPL blob
with trailing heads and impact rings. Variables show as squares.
State changes flash white. Auto-layout in hex grid.
This is the compile-time artifact canvas that's always been
part of the Blimp REPL -- now in the tutorial.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
6924581 fix: add debug output for WASM load failures
5e6441d Merge branch 'worktree-agent-a29a62a7'
# Conflicts:
# chunks/lang/web/server.blimp
0d5030f fix: and/or as expression operators, pipes with map/filter/reduce
Bug 1: 'and' and 'or' now work as expression operators alongside
&& and ||. Added kw_and/kw_or to token table and parser.
x > 0 and x < 10 # works now
Bug 2: Pipes with higher-order functions (map/filter/reduce/each)
now work. Pipe evaluator routes through evalFuncCall instead of
only checking the builtin registry.
[1,2,3] |> filter(_, fn(x: Int) do x > 1 end) # works now
[1,2,3] |> map(_, fn(x: Int) do x * 10 end) # works now
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
a8f8dd9 deploy: Blimp Chat live at http://5.161.181.91:8080
Cross-compiled Blimp binary for Linux x86_64, deployed to Hetzner.
Running as systemd service (blimp-chat.service), 780KB memory.
Port 8080 opened on firewall.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
50afcf7 feat: multi-actor view composition with mount()
Actors own their state via become, render via :render handler.
mount() sends :render + wraps in data-actor boundary for routing.
JS DOM walk finds nearest actor boundary on button click.
New builtins: mount_root, actor_name, to_atom
Demo: Counter + Greeter composed in one page
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2c7cbee feat: canvas output pane on every lesson
Every lesson shows an Output canvas that displays the running
system's result: values for expressions, rendered views for view
code, errors in red, print output in blue. Canvas always visible
with "Hit Run to see output" placeholder. Consistent Run button
across all lesson types.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
bfeabd0 test: 623 tests -- comprehensive coverage of the entire language
Builtins: 231 tests -- every string/list/map/math/logic/HOF builtin
with edge cases (empty, nil, zero, out-of-bounds)
Operators: 142 tests -- all arithmetic, comparison, logical, pipe,
send, concat operators with type combinations
Language: 125 tests -- actors, become, guards, case, for, def, fn,
closures, cons lists, maps, interpolation, scoping,
dot notation, actor interaction, edge cases
Lexer: 36 tests -- all token types and edge cases
Parser: 34 tests -- all AST node types
Evaluator: 31 tests -- all eval paths
Completion: 8 tests
Bootstrap: 16 tests -- full self-hosted pipeline
4014 lines of tests. 1.5 seconds to run. All green.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
e0396f2 feat: ExUnit-inspired test runner for Blimp
New syntax: test "name" do ... end inside actor bodies.
Suite actor state fields are re-evaluated per test for isolation
(spawn Counter gives each test a fresh instance).
Assertions: assert, assert_eq, assert_ne, refute with colored
left/right output on failure.
Two run modes:
./blimp file.blimp --test (single file)
./blimp test test/ (discover *_test.blimp files)
Green dots for pass, red F for fail, failure summary, timing.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
b48d91d fix: reuse REPL artifact for canvas -- same styles, same rendering
Output tab now mirrors the REPL exactly:
- REPL output lines (=> result, prints in blue, errors in red)
- View pane with blimp-* CSS classes from index.html
- State pane showing vars (green names) and actors (with fields)
- Message log (arrows with target + message)
No custom SVG. Same artifact as the compiled REPL.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
7bebcdc fix: autocomplete derives from builtin registry, not hardcoded list
Completions now iterate eval.builtins.entries.items directly.
New builtins automatically appear in autocomplete without
maintaining a parallel list. Higher-order builtins (map/filter/
reduce/each) added separately since they're in evalFuncCall
not the registry.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
da4afa2 feat: forms, websockets, concurrency builtins + string match fix
Three parallel agents' work merged as one (worktree isolation failed):
- Forms: input/textarea/select/option/form view primitives, list
flattening in makeViewNode, todo list demo replacing counter
- WebSockets: ws_accept_key (SHA-1+Base64), ws_read_frame,
ws_write_frame, view_diff tree diffing, client JS with fallback
- Concurrency: fork/waitpid/exit, tcp_set_nonblocking/tcp_poll,
poll-based concurrent_server.blimp demo
- Bug fix: string pattern matching in case expressions (lexer
stores "POST" with quotes, runtime values are unquoted)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ca21d7e feat: dot notation for actor names in message sends and spawn
Shop.Checkout <- :add(10) now parses, type-checks, and evaluates.
Parser accepts upper_identifier after dot, eval/checker flatten
DotAccess chains into dotted name strings for registry lookup.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ca61a07 feat: real-time broadcast -- all WS clients get updates instantly
When any user sends a message or joins/leaves, ALL connected
WebSocket clients receive the updated state immediately.
Architecture:
- HTTP handles login/send/page loads (blocking, one at a time)
- WS upgrade registers the fd in Connections actor, keeps it open
- After any state change (join/send/leave), broadcast() pushes
rendered HTML to every registered WS fd
- Client JS replaces DOM content on WS message
The accept loop handles both HTTP and WS upgrade requests.
WS fds stay open after upgrade for push-only communication.
Messages go via HTTP POST form, updates arrive via WS push.
Open two browser tabs, type in one, see it in the other instantly.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
bfe8565 fix: reject untyped function parameters at compile time
def double(n) do ... end is now a type error:
"function parameter 'n' is missing a type annotation"
Must be: def double(n: Int) do ... end
Checker now validates def_stmt params the same way it
validates handler params. WASM rebuilt.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
c884d57 feat: WebSocket chat -- no more page reloads
The chat room now uses WebSockets for real-time updates:
- Initial page load is HTTP (serves HTML + WS client JS)
- Client opens WebSocket to /ws
- Messages sent as JSON frames over WS
- Server pushes rendered HTML updates instantly
- Client replaces DOM content without page reload
- Exponential backoff reconnect on disconnect
Server-side: accept_loop detects Upgrade header, does WS
handshake (SHA-1 + Base64), enters ws_chat_loop that reads
frames and pushes state. JSON escaping for safe embedding.
No more setTimeout(reload, 3000). Instant.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ef1e202 debug: traced elem() index bug in recursive TCP handler chain
The render_messages function works standalone but crashes when
called from the HTTP accept loop. Traced to elem(messages, idx)
receiving idx=1 when it should be idx=0.
The issue appears in slice_idx recursive calls:
slice_idx(items, start, stop, idx + 1, append(acc, elem(items, idx)))
The elem(items, idx) should use idx=0 but gets idx=1. This only
happens when called from 4+ levels deep in the TCP accept loop.
Likely a scope/environment issue in the Zig evaluator's handling
of recursive function calls with expression arguments.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
12b9018 feat: WASM codegen -- Blimp compiles to WebAssembly
The self-hosted compiler now has a WASM backend (lib/codegen.blimp,
225 lines). Compiles Blimp integer expressions to valid .wasm
modules that run in any WASM runtime.
compile_to_wasm("2 + 3") -> [0,97,115,109,...] (40 bytes)
# Node.js: main() returns 5
# Browser: main() returns 5
Supported: integer literals, +, -, *, /, ==, !=, <, >, <=, >=,
&&, ||, unary negation, booleans, nested expressions.
Output: valid WASM module exporting "main" -> i32.
Bytecode: i32.const, i32.add/sub/mul/div_s, i32.eq/ne/lt_s/gt_s/
le_s/ge_s, i32.and/or.
New builtins: write_bytes(path, byte_list), read_file(path).
23 tests (18 unit + 5 property/500 random compilations). All green.
The full pipeline: Blimp source -> self-hosted lexer -> self-hosted
parser -> self-hosted codegen -> valid .wasm bytes -> runs in browser.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
3bbdd09 feat: real system canvas showing actors, state, messages, views
Output tab now shows the full running system:
- Actor boxes (purple name, green state fields)
- Message log (arrows with target + message name)
- Rendered views in a light preview pane
- Return value
- Variable pills
All pulled from WASM getState() which exposes the full
actor registry, state fields, and message log.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
99b1feb feat: cookie-based session identity for Blimp HTTP server
Pure Blimp string parsing — no Zig changes. Each browser gets its
own session via Set-Cookie: blimp_session=<token>; HttpOnly; SameSite=Lax.
Token from random()+now(), Sessions actor stores per-session state.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
d364115 Merge branch 'worktree-agent-a34cbc3d'
# Conflicts:
# chunks/lang/src/builtins.zig
# chunks/lang/web/server.blimp
7b23f3a feat: visual actor canvas with boxes, state fields, and message rays
Output tab now shows a real visual diagram:
- Actor nodes as purple-bordered boxes with state fields in green
- Message rays as animated dashed yellow lines from "you" to actors
- :message labels on each ray
- Staggered ray animation (0.2s delay per message)
- Diagram auto-layouts actors in a row with SVG overlay
- State text and view pane below the diagram
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
c77779e feat: SVG system canvas with actor nodes, message rays, state
Output tab now renders a live SVG diagram:
- Actor nodes as purple-bordered rounded rects with state fields
- Message rays (yellow animated lines) from REPL to target actors
- "you" node as message origin
- Ray pulse animation on each message send
- Variables as pills, print output, rendered views below the graph
- Canvas paint happens before validate to avoid state clobber
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
d2b1898 test: comprehensive coverage -- 125/125 tests pass
Lexer: 36 tests (was 10) -- all tokens, operators, delimiters,
keywords, comments, line/col tracking, edge cases
Parser: 34 tests (was 11) -- all literals, all operators,
unary, parens, func calls, defs with return types,
actors, handlers, dotted names, message sends,
lists, case, spawn, multi-statement, empty input
Evaluator: 31 tests (was 11) -- all literals including nil/bool/atom,
all arithmetic, all comparisons, && and ||, variables,
function def/call/nested, builtins, lists, case with
wildcard and no-match, spawn
Completion: 8 tests
Bootstrap: 16 tests -- full pipeline source->lex->parse->eval
Also fixed: lexer handles underscore via char code (95) to
avoid conflict with Blimp's wildcard pattern.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
6261950 fix: add type annotations to function params in tutorial lessons
Blimp requires typed parameters: def double(n: Int), not def double(n).
Fixed lessons 5 (double) and 18 (profile_card).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+485-0 notactuallytreyanastasio/lumines-clone/main Mar 28 22:49
738e81b update claude with elixir stuff
+196-11 notactuallytreyanastasio/lumines-clone/main Mar 28 22:45
40b2e58 Merge pull request #2 from notactuallytreyanastasio/liveview-engine
Liveview engine
+196-11 notactuallytreyanastasio/lumines-clone/liveview-engine Mar 28 22:43
eb63143 add decidous db
810967c docs: add zero-to-running README and update CLAUDE.md with architecture notes
README covers installing Erlang/Elixir/Postgres from scratch on
macOS, Ubuntu, and Windows, plus setup, controls, and project structure.
CLAUDE.md updated with architecture decisions, style rules, and notes
on what strategies worked during parallel development.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+1768-0 notactuallytreyanastasio/lumines-clone/main Mar 28 22:41
26a8986 Merge pull request #1 from notactuallytreyanastasio/liveview-engine
Lumines game engine + LiveView UI
8bb02d1 feat: implement Lumines engine in Phoenix LiveView with shared CSS
- 7 engine modules: Board (sparse map), Piece, Gravity, Scanner, Sweep, Scoring, Game
- LiveView at /game with keyboard input and dual timers
- Integrated shared CSS design system (lumines-* classes)
- 72 ExUnit tests, all passing
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+442-44 notactuallytreyanastasio/lumines-clone/liveview-engine Mar 28 22:37
7a1b355 Address PR feedback + add property-based testing
- board.ex: replace nested ifs with multi-clause functions using guards
- board.ex: use `not is_nil()` instead of `!= nil`
- game.ex, piece.ex, sweep.ex, scoring.ex: convert to Ecto embedded_schema
- Add 27 property-based tests using stream_data covering:
- Board: put/get roundtrip, nil clearing, out-of-bounds, occupied? consistency
- Piece: 4-rotation identity, color preservation, 2x2 area, hard_drop validity
- Gravity: cell count preservation, color preservation, idempotency, no floating cells
- Scanner: only marks valid squares, preserves non-square cells
- Sweep: bounds, 16-tick wraparound
- Scoring: monotonic score, combo tracking, reset preserves score
- Game: valid init, game_over handling, input safety
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+9346-78 notactuallytreyanastasio/blimp/main Mar 27 23:00
b4ae6b6 finish up some more typing concerns
33ccd90 WIP
fb5466b more stuff
9c5c74d more chugging along with the hole operator
33b3690 Hole operator: situation _ # directive shells out to Claude
- lexer.zig: capture comment text in last_comment (not reset on newlines)
so parser can read directives like `_ # Hole: fill this in`
- parser.zig: hole branches now store directive text from the comment
body contains a hole node with directive instead of being empty
- eval.zig: evalHole() called when situation hits a directive-only _ branch
- builds prompt: directive + subject value + visible bindings + syntax notes
- shells out to `claude -p <prompt>` as subprocess
- strips markdown fences from response
- parses response with parseFilePublic, evaluates all nodes, returns last
- graceful nil fallback if claude not found or parse fails
- Works at REPL time and file evaluation time (both contexts)
Semantics: situation = incomplete code with Holes, case = complete dispatch
20d8541 Merge blimp-core/view-dsl: View DSL actors render to DOM, interactive tutorial, mobile playground
Resolved conflict in builtins.zig by keeping both math/util builtins from main
and view primitive registrations from view-dsl. Took view-dsl WASM binaries.
5238c23 docs: web framework handoff guide for next session
a23d476 Hole: auto re-exec after patch + diff output
After patching the source file, execv() replaces the current process with
a fresh run of the patched file — zero user interaction required.
Diff output shows exactly what changed:
╔═══ Hole filled: file.blimp:42 ══════
║ - _ # Hole: directive text
║ + _ ->
║ + generated line 1
║ + generated line 2
╚═════════════════════════════════════
The full cycle: hit hole → call Claude → print diff → patch file → re-exec.
Second run uses the real implementation, no hole, no Claude call.
f0435a1 Typed lists and record types: [Item] and %{key: Type, key: Type}
types.zig: add record_type variant to the Type union
- RecordField struct: {name, ty}
- recordField(name) method for O(n) lookup by field name
- record_type and map are mutually compatible (subtyping both directions)
since records are just typed maps at runtime
- parseTypeName now handles %{key: Type, ...} format (record)
alongside the existing %{K => V} format (homogeneous map)
- parseRecordField helper splits "key: Type" into name + type
parser.zig: %{key: Type, ...} accepted as a type annotation
- detects homogeneous (%{K => V}) vs record (%{key: T}) by presence of =>
- handles multi-line record type annotations with newlines between fields
- trailing comma allowed
checker.zig: lookup() on record type returns specific field type
- if key is a string literal, resolves the declared field type
- lookup(profile: %{name: String}, "name") infers String return
- falls back to Any for non-literal keys or unknown fields
4a847ab Fix docs and hole prompt: lambdas/filter/reduce/map all exist and work
f9f865f web server working: escape sequences, auto-spawn actors, file evaluation
- main.zig: file mode now evaluates (was only printing AST); --ast flag for old behavior
- eval.zig: string literals process \r \n \t \\ escape sequences
- eval.zig: actor definitions auto-spawn a singleton + bind actor_ref in env
- value.zig: add ViewNode to typeName() exhaustive switch
- builtins.zig: ArrayListUnmanaged for Zig 0.15 compat in to_html
- .tool-versions: zig 0.15.2
- Makefile: build commands for macOS 26 (zig build runner broken)
- server.blimp: case (not situation) for complete dispatch
b0bef0e web framework: TCP/HTTP builtins in Zig, server.blimp in Blimp
- tcp_listen/accept/read/write/close builtins (POSIX sockets, native only)
- to_html builtin: renders view_node trees to HTML strings
- Handles all 17 view primitives
- row gets data-row attr for flexbox direction
- button sends: attr becomes data-sends + onclick handler
- HTML-escapes string content
- web/server.blimp: HTTP/1.1 server written entirely in Blimp
- HTTP.Server actor: tcp_listen loop, accept, read, dispatch, respond
- Sessions actor: in-memory session store (map of id -> state)
- Pure render_app(data) -> view_node
- Pure handle_event(data, msg) -> data
- html_page wraps view HTML in a full document with styles + client JS
- Client JS: blimpSend() submits button clicks as form POSTs
- PRG pattern: POST handler -> redirect -> GET
a496725 Fix type checker issues, remove --no-type-check escape hatch
types.zig: List → [Any], Map → %{Any=>Any}, ViewNode → Any
Previously these fell through to actor type, causing false errors
checker.zig: length() accepts list, map, tuple, string, nil — any collection
parser.zig: multi-line map literals — skipNewlines() after % and after {
allows %{\n key: val\n} syntax to parse correctly
case/situation as expression on RHS of assignment (x = case foo do...)
builtins.zig: set_at(list, index, value) — list element replacement
main.zig: remove --no-type-check flag entirely, type checker is mandatory
examples: add missing type annotations, use set_at instead of replace_at
683dfb7 examples: three nontrivial Hole operator demos
rate_limiter.blimp — adaptive token bucket with missing backoff strategy,
burst abuse detection, and priority request overdraft logic
content_pipeline.blimp — staged moderation pipeline: sanitize → classify →
score → publish/reject. Classification heuristics and scoring edge cases
are holes; the pipeline wiring and decision dispatch are implemented.
game_rules.blimp — tic-tac-toe engine with missing win detection, score
calculation, illegal move handling, and AI opponent move selection.
Good example of domain-specific reasoning Claude can fill in from context.
8acd934 Hole source patching: living codebase — holes fill themselves permanently
- lexer.zig: preserve last_comment across newline tokens so parser can read
directive text from the line AFTER the hole token is consumed
- parser.zig: save `_` token's line/col BEFORE advancing (correct loc for patch)
new parseHandlerBodyPublic() allows parsing handler-body statements in holes
- eval.zig: evalHole() now patches the source file after filling in
- finds the hole line by its stored loc
- detects original indentation from that line
- builds replacement: `_ ->\n <generated lines>`
- writes patched source back to disk
- subsequent runs hit the real implementation, not the Hole
- main.zig: set evaluator.source_path from resolved file path
--no-type-check flag to skip type checker for exploratory code
+3119-21 notactuallytreyanastasio/blimp/blimp-core/view-dsl Mar 27 15:52
7a33020 View DSL: actors render to DOM, interactive tutorial, mobile playground
- view_node JSON serialization in wasm_api.zig (blimp_has_view, blimp_get_view_ptr/len)
- JS DOM renderer maps view primitives to HTML elements
- Buttons send messages back to actors via <- syntax, triggering re-render
- Parser fix: skip newlines inside () argument lists for multi-line view calls
- dom_playground.html: mobile playground with live sim loop, 5-actor coffee shop demo
- tutorial.html: 14-section interactive tutorial from zero to coffee shop, all code blocks runnable
- docs/lang_design/view-engine.md: capabilities document
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Vibe Coding Rescue Missions
this mortal coil
are the buttons in the post editor just for show?
heya
it is katproto
Under-learning > Over-learning
Asking. Hacker News to Write For me
Anyone can post
Rat Kings Are Good Actually
Untitled
327 Years Of Tree Law In The Usa
A Genstage Tutorial And Reflection
Nathan Fielder
Whats My Schtick
Lets Write Letters
A Quick Typewriter Set Letter Project
I Pulled The Paul Tickets
A Fun Experiment
A Hilarious Shared Keyboard Key
A Me Museum
On Making A Link Blog
Building This Blog
Online (0)