Bug 1950096 [wpt PR 50907] - [functions] Let parameter defaults reference other parameters, a=testonly

Automatic update from web-platform-tests
[functions] Let parameter defaults reference other parameters

Per "evaluate a custom function" [1], the default value of a parameter
may "see" other parameters, e.g. @function --f(--x, --y:var(--x)).
Before this CL, that var(--x) would resolve in the outer scope,
but now it's supposed to resolve against the actual argument value
passed for --x. At the same time, default values should not have
access to the actual *locals* of the function body, meaning that
they resolve in some gray area *between* the outer scope and inner
scope.

In this CL, that "gray area" is implemented by using a disposable
FunctionContext with the unresolved defaults set as the unresolved
locals. This is convenient, because we get all the needed "dependency
handling" (including cycle detection) for free.

Since various things now need to parse against a type (arguments,
defaults, and the 'result' descriptor), and because these various
things differ in how they are represented (some CSSVariableData,
some StringView), this CL aligns all of those things to use
CSSVariableData, and adjusts ResolveFunctionExpression accordingly.
This would be needed anyway, to actually transport various tainting
information, though that is not explicit handled/tested in this CL.

Finally, note that there is a drive-by fix in this CL: the "already
applied" branch of ApplyLocalVariables now actually skips the work
if it's already applied. This should ideally have a performance test,
which is tracked as a follow-up task (Issue 397164440).

[1] https://drafts.csswg.org/css-mixins-1/#evaluate-a-custom-function

Fixed: 397459622
Bug: 325504770, 397164440
Change-Id: Ic705209d25f4bee6149f2397bda78e1224b16901
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/6276516
Commit-Queue: Anders Hartvoll Ruud <[email protected]>
Reviewed-by: Steinar H Gunderson <[email protected]>
Cr-Commit-Position: refs/heads/main@{#1423855}

--

wpt-commits: b5045d03a5a6963f8af68d403e24bb325f2ce2e6
wpt-pr: 50907

Author: andruud

testing/web-platform/tests/css/css-mixins/dashed-function-cycles.tentative.html

@@ -307,6 +307,31 @@
   </style>
 </template>
 
+<template data-name="Function in a cycle with its own default">
+  <style>
+    @function --f(--x, --y: --f(13px)) {
+      result: 10px;
+    }
+    #target {
+      --tmp: --f(42px);
+      --actual: var(--tmp, PASS);
+      --expected: PASS;
+    }
+  </style>
+</template>
+
+<template data-name="Cyclic defaults">
+  <style>
+    @function --f(--x, --y: var(--z), --z: var(--y)) {
+      result: var(--x, FAIL) var(--y, PASS-y) var(--z, PASS-z);
+    }
+    #target {
+      --actual: --f(42px);
+      --expected: 42px PASS-y PASS-z;
+    }
+  </style>
+</template>
+
 <script>
   test_all_templates();
 </script>

testing/web-platform/tests/css/css-mixins/dashed-function-eval.html

@@ -305,6 +305,63 @@
   </style>
 </template>
 
+<template data-name="Default referencing another parameter">
+  <style>
+    @function --f(--x, --y: var(--x)) {
+      result: var(--x) var(--y);
+    }
+    #target {
+      --x: FAIL;
+      --y: FAIL;
+      --actual: --f(5px);
+      --expected: 5px 5px;
+    }
+  </style>
+</template>
+
+<template data-name="Default referencing another parameter, local interference">
+  <style>
+    @function --f(--x, --y: var(--x)) {
+      --x: 17px;
+      result: var(--x) var(--y);
+    }
+    #target {
+      --x: FAIL;
+      --y: FAIL;
+      --actual: --f(5px);
+      --expected: 17px 5px;
+    }
+  </style>
+</template>
+
+<template data-name="Default referencing another defaulted parameter">
+  <style>
+    @function --f(--x: 5px, --y: var(--x)) {
+      result: var(--x) var(--y);
+    }
+    #target {
+      --x: FAIL;
+      --y: FAIL;
+      --actual: --f();
+      --expected: 5px 5px;
+    }
+  </style>
+</template>
+
+<template data-name="Typed default with reference">
+  <style>
+    @function --f(--x: 5px, --y <length>: calc(var(--x) + 1px)) {
+      result: var(--x) var(--y);
+    }
+    #target {
+      --x: FAIL;
+      --y: FAIL;
+      --actual: --f();
+      --expected: 5px 6px;
+    }
+  </style>
+</template>
+
 <!-- Locals -->
 
 <template data-name="Unused local">
@@ -420,11 +477,6 @@
   </style>
 </template>
 
-<!--
-  TODO(andruud): Add more tests for dynamic scoping when Issue 10954
-  resolves: https://github.com/w3c/csswg-drafts/issues/10954
--->
-
 <template data-name="Substitute local from outer scope">
   <style>
     @function --f() {
@@ -623,6 +675,51 @@
   </style>
 </template>
 
+<template data-name="Inner function call should see resolved outer locals">
+  <style>
+    @function --a() {
+      --x: --b();
+      --y: var(--px);
+      result: var(--x);
+    }
+
+    @function --b() {
+      result: var(--y, FAIL);
+    }
+    #target {
+      --px: 10px;
+      --actual: --a();
+      --expected: 10px;
+    }
+  </style>
+</template>
+
+<!--
+     This the same test as the one above, but the *values* of --x and --y
+     are flipped, as are the references to those vars. If there is a bug
+     related to this behavior, it may be masked by "lucky" ordering of items
+     in a hash backing. Testing both ways ensures that at least one test
+     fails.
+-->
+<template data-name="Inner function call should see resolved outer locals (reverse)">
+  <style>
+    @function --a() {
+      --x: var(--px);
+      --y: --b();
+      result: var(--y);
+    }
+
+    @function --b() {
+      result: var(--x, FAIL);
+    }
+    #target {
+      --px: 10px;
+      --actual: --a();
+      --expected: 10px;
+    }
+  </style>
+</template>
+
 <!-- Shadowing -->
 
 <template data-name="Parameter shadows custom property">