release: cut v0.45.0

CHANGELOG.md

@@ -4,6 +4,53 @@ All notable changes to EigenScript are documented here.
 
 ## [Unreleased]
 
+## [0.45.0] - 2026-10-07
+
+### Breaking changes
+
+- **A distinct `bool` type (#1637).** Comparisons and `not` return `true`/`false` of type `bool`
+  instead of the numbers `1`/`0`. Truthiness is unchanged (`false` is falsy, `true` truthy).
+  Arithmetic on a bool raises, and so does `==`/`!=` between a bool and a number: migrate
+  `(pred of x) == 1` to `if pred of x:` or `== true`.
+  `list_contains`/`list_index_of` raise on bool vs number the same way. Any
+  builtin given a bool where it does not take one raises, even under
+  `EIGS_STRICT=0`: as its argument, at a position of its argument list, or
+  as an element of a list it reads as numbers. The slots that take a bool are
+  listed in `tests/bool_fuzz_anyvalue.txt`. A bool slice bound, index, `range`
+  bound or `at` line raises. Deep `==` stops at the first unequal pair, and a
+  bool/number pair raises when the walk reaches it. Trace tapes are format v6;
+  a v5 tape is refused, and so is a record of a kind its builtin cannot
+  return. Embedding hosts declare a recorded name's kinds with
+  `eigs_trace_declare_kind`. `db_execute`/`db_query_value`/`db_query_json` take a bool
+  parameter and bind it as an SQL boolean, so a bool read from a boolean
+  column can be written back. `json_build` writes a bool value as JSON
+  `true`/`false`, and `what is` a bool answers the bool. The embedding API's
+  `eigs_value_as_num` answers NaN for a bool (0.0 for any other non-number).
+- **Wrong-typed numbers that used to be read as garbage now raise (#1637).** The
+  bool work routed every C number read through a checking accessor, which
+  also changes some non-bool cases:
+  - a string `at` line (`what is x at "s"`) raises; it answered null;
+  - a non-number net timeout (`net_accept`/`net_dial`/`net_recv`), a
+    non-number `http_serve` port, and a non-number code byte in a
+    `vm_run_bytecode`/`sandbox_run` descriptor raise; they were read as
+    garbage or as 0 (a non-number `param_count` placeholder such as `[]`
+    still means 0 parameters; only a bool there raises);
+  - the tensor mutators `numerical_grad*`/`sgd_update*` raise on a non-number
+    cell or row in every strict mode, `EIGS_STRICT=0` included: v0.44.0 read
+    a non-number cell as 0 (and `numerical_grad_rows`/`_cols` wrote into it
+    in place) and skipped a null row;
+  - a non-number `screen_put` color and a non-number, non-bool
+    `write_bytes` append flag raise in every strict mode (neither had a
+    documented `EIGS_STRICT=0` answer);
+  - a non-number cell `gather` selects raises under `EIGS_STRICT` (it was
+    0); `EIGS_STRICT=0` keeps the 0.
+  `EIGS_STRICT=0` keeps every answer it documents for non-bool values:
+  `str_from_bytes`, `inflate`/`deflate` and the other byte-list builtins
+  still read a non-number element as 0 there (a bool raises in every mode).
+  Embedders that include `eigenscript.h`: a Value's number is `VAL_NUM_RAW(v)`
+  and a slot's is `SLOT_NUM_RAW(s)` (the members were renamed so an unchecked
+  read does not compile).
+
 ## [0.44.0] - 2026-10-05
 
 - HTTP routes and the static root are frozen once `http_serve` starts (#1140). A `code` route's worker points at the spawning server, so `http_route`, `http_route_authed` or `http_static` called from one wrote the live route table while every other worker read it lock-free: one request could register a route for the whole process, and `route_count++` with no fence let a concurrent lookup `strcmp` a half-filled slot (ThreadSanitizer: 10 reports on a /mutate-vs-/ hammer before, 0 after). The three builtins now raise "must register before http_serve" once serving, as `http_response_header` already did. An uncaught error in a `code` route's source now answers 500 with the generic body `{"error": "internal error"}` instead of `200 null`; the message goes only to the server's stderr, since it can carry paths or values a client must not see.

VERSION

@@ -1 +1 @@
-0.44.0
+0.45.0