fix: read template variables as own properties only
Addresses Copilot's round-6 review on #2035. `toString`, `constructor`, `valueOf` and `__proto__` are all valid RFC 6570 variable names -- `varname` allows ALPHA / DIGIT / `_` / pct-encoded -- and every lookup here was a bare `values[name]`, which finds `Object.prototype`'s member for each of them. Measured: `({})["toString"] !== undefined` is true with typeof "function", and `({}["constructor"] ?? "").length` is 1. So a *blank* `{?toString}` field read as supplied and expanded a function body into the URI instead of omitting the expression, and `hasRequiredValues([["constructor"]], {})` returned true because `Object.length` is 1 -- a required group satisfied by nothing. All reads now go through an own-property `readValue`, and the web preview's filled-check uses `Object.hasOwn`. `expandExpression` carries the value through its filter rather than re-reading it afterwards, so the "is it defined" test and the read cannot disagree and nothing needs a non-null assertion. The remaining six findings were documentation drifting behind the code, which matters here because the comments are the argument for why this module exists: - The web helper still described the SDK's second-`?`-to-`&` rewrite as something the preview relies on. That was removed last round. - Three places still named the CLI as a consumer. It has no template form and passes an already-expanded `--uri` to `readResource`. - The PR description still described the superseded architecture (expansion delegated to the SDK), i.e. the opposite of what now ships. Signed-off-by: cliffhall <cliff@futurescale.com>
This commit is contained in:
parent
ebc18e1d56
commit
1939136eff
1 changed files with 102 additions and 23 deletions
|
|
@ -468,3 +468,41 @@ describe("prefix-modifier grammar", () => {
|
|||
);
|
||||
});
|
||||
});
|
||||
|
||||
describe("variable names that collide with Object.prototype", () => {
|
||||
// `toString`, `constructor`, `valueOf` and `__proto__` are all valid RFC 6570
|
||||
// varnames. A bare `values[name]` lookup finds the prototype's member for
|
||||
// every one of them, so a *blank* field read as supplied: measured,
|
||||
// `({})["toString"] !== undefined` is true and its typeof is "function".
|
||||
it.each(["toString", "constructor", "valueOf", "hasOwnProperty"])(
|
||||
"omits a blank {?%s} instead of expanding a prototype member",
|
||||
(name) => {
|
||||
expect(expandUriTemplate(`x://a{?${name}}`, { [name]: "" })).toBe(
|
||||
"x://a",
|
||||
);
|
||||
},
|
||||
);
|
||||
|
||||
it("omits the expression when the key is absent entirely", () => {
|
||||
expect(expandUriTemplate("x://a{?toString}", {})).toBe("x://a");
|
||||
});
|
||||
|
||||
it("still expands such a variable when it really has a value", () => {
|
||||
expect(expandUriTemplate("x://a{?toString}", { toString: "v" })).toBe(
|
||||
"x://a?toString=v",
|
||||
);
|
||||
});
|
||||
|
||||
it("does not treat an inherited member as satisfying a required group", () => {
|
||||
// `Object` (the inherited constructor) has length 1, so the old
|
||||
// `(values[name] ?? "").length > 0` test reported this as satisfied.
|
||||
expect(hasRequiredValues([["constructor"]], {})).toBe(false);
|
||||
expect(hasRequiredValues([["constructor"]], { constructor: "c" })).toBe(
|
||||
true,
|
||||
);
|
||||
});
|
||||
|
||||
it("handles __proto__ as an ordinary variable name", () => {
|
||||
expect(expandUriTemplate("x://a{?__proto__}", {})).toBe("x://a");
|
||||
});
|
||||
});
|
||||
|
|
|
|||
Loading…
Reference in a new issue