@zvenigora/ng-eval-core 0.7.0 → 0.9.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -14,6 +14,75 @@ Versions 0.1.104–0.1.107 are on npm without entries here.
14
14
 
15
15
  ---
16
16
 
17
+ ## [0.9.0] - 2026-10-02
18
+
19
+ **A primitive receiver no longer skips the prototype-pollution guard —
20
+ [B1](../../docs/backlog-retired.md#b1).** A breaking minor. No exported symbol changes shape, but an
21
+ expression that read one of seven names off a string, number or boolean now throws.
22
+ `eval-signals` 0.2.1 and `eval-forms` 0.2.7 widen their peer ranges to admit it.
23
+
24
+ ### Breaking
25
+ - **A string, number or boolean receiver is refused `constructor`, `__proto__`, `prototype`,
26
+ `__defineGetter__`, `__defineSetter__`, `__lookupGetter__` and `__lookupSetter__`**, by dot or
27
+ computed access, with or without `caseInsensitive`, with the error any other blocked name
28
+ throws: `Access to dangerous property "constructor" is blocked for security reasons`. Why: the
29
+ member visitor skipped the guard entirely for a primitive receiver, so `s.constructor` handed
30
+ back the global `String` function — callable, `s.constructor("x")` — and `s.__proto__` handed
31
+ back `String.prototype`, with `n.__proto__.toFixed.call(1.5, 0)` running off it. No escalation
32
+ past that was found (every second hop to a dangerous name was already refused, and so was every
33
+ write to a built-in prototype), but nothing an expression needs from a primitive goes through
34
+ its constructor or its prototype.
35
+ - **Kept readable on a primitive:** `toString`, `valueOf`, `toLocaleString`, `hasOwnProperty`,
36
+ `isPrototypeOf` and `propertyIsEnumerable`. They are on the blocklist for objects, but on a
37
+ primitive they are ordinary reads — `s.toString()`, `n.toLocaleString()`,
38
+ `s.hasOwnProperty("length")` work as before, as do `s.length` and `s.toUpperCase()`.
39
+ - **A case variant is not refused.** The names are matched exactly, and a primitive receiver is
40
+ never case-corrected, so under `caseInsensitive` `s.CONSTRUCTOR` is a missing property and
41
+ evaluates to `undefined`, as it did in 0.8.0.
42
+
43
+ ---
44
+
45
+ ## [0.8.0] - 2026-10-02
46
+
47
+ **`EvalContext.getKey` resolves through `get`'s own chain —
48
+ [A4](../../docs/backlog-retired.md#a4) and [A10](../../docs/backlog-retired.md#a10).** A breaking
49
+ minor. No exported symbol changes shape — the `.d.ts` differs from 0.7.0 in documentation comments
50
+ only — but `getKey`'s answers change, and with them the `key` of identifier and member read events
51
+ and the key the write visitors assign to under `caseInsensitive`. `eval-signals` 0.2.0 and
52
+ `eval-forms` 0.2.6 widen their peer ranges to admit it.
53
+
54
+ ### Breaking
55
+ - **`getKey` answers what `get` resolved, and nothing when `get` finds nothing.** It used to walk
56
+ its own copy of the resolution order, which disagreed with `get` in four ways; each is now
57
+ `get`'s answer:
58
+ - **A namespace is reported as the scope declares it.** With a prior scope namespaced `dog`,
59
+ `Dog.Says()` under `caseInsensitive` reports read keys `['dog', 'says']`, where it reported
60
+ `['Dog', 'says']`.
61
+ - **An unbound name gets no key.** With a plain-object scope pushed, `getKey` answered every
62
+ key as present, so a name nothing holds came back as itself; and a key held only in a source
63
+ `get` does not read — a wrong-case key in a case-sensitive `Registry`, a key inside a
64
+ non-global prior scope — came back spelled as that source holds it. All of these are now
65
+ `undefined`. An identifier read still reports the name as written when `getKey` answers
66
+ nothing.
67
+ - **A key only a lookup resolves is reported as written**, where `getKey` returned `undefined`
68
+ because it never consulted lookups. Under `caseInsensitive` the write visitors resolve their
69
+ target through `getKey`, so an assignment to such a key now writes, or is refused, under its
70
+ name: a `@zvenigora/ng-eval-signals` context's write error names `'COUNT'` rather than
71
+ `'undefined'`.
72
+ - **A spelling comes from the source that supplied the value.** Under `caseInsensitive` with a
73
+ case-sensitive `Registry` holding `a`, `getKey('A')` returned `'a'` while `get('A')` found
74
+ nothing there and went on to later sources.
75
+ - **A member of the context itself resolves through `getKey`'s answer.** Under `caseInsensitive`,
76
+ `This.COUNT` for a key only a lookup resolves, and `This.Dog` for a namespace `dog`, evaluated
77
+ to `undefined`: the member visitor read `get(getKey(key))`, and `getKey` answered nothing. Both
78
+ now resolve.
79
+
80
+ ### Changed
81
+ - `getKey`'s JSDoc states its answers, and two sentences in `getFromScopes` / `scopeHolding` that
82
+ described `getKey`'s drift are corrected. Those are the whole `.d.ts` difference from 0.7.0.
83
+
84
+ ---
85
+
17
86
  ## [0.7.0] - 2026-10-01
18
87
 
19
88
  **Errors keep their identity, `thisArg` is applied, and three silent outcomes now throw —