vendor a method-less text/template to restore whole-binary linker DCE (#8844)

### Why the changes in this PR are needed?

`text/template`'s field evaluator (`text/template.(*state).evalField`,
`exec.go`) calls`reflect.Value.MethodByName` with a non-constant name.
The Go linker treats a reachable non-constant`MethodByName` as a signal
to disable **method-level dead-code elimination for the whole binary**
(see `cmd/link/internal/ld/deadcode.go` and golang/go#72895). Two OPA
code paths pull stdlib `text/template` into the reachable graph of
ordinary embedders:

1. **Compiler frontend** — `ast.Compiler.Compile → … →
gojsonschema.formatErrorDescription → text/template`. Reached
unconditionally by anything that compiles Rego.
2. **`strings.render_template` builtin** (`v1/topdown/template.go`) —
registered in the topdown builtin table, reachable in anything that
links Rego evaluation.

So an embedder of OPA's compiler/eval retains its entire reachable
method surface — a large binary-size regression, hundreds of MB in the
reporter's case (#7903). Both edges must go before the linker re-enables
method-level DCE for that embedder.

### What are the changes in this PR?

Vendor a self-contained, method-less copy of `text/template` under
`internal/methodlesstemplate` and point both call sites at it. **No
external dependency** (`go.mod`/`go.sum` unchanged).

- Copied verbatim from **Go 1.25.8**: `doc.go`, `exec.go`, `funcs.go`,
`option.go`, `template.go`, plus `internal/fmtsort/sort.go`. Go's BSD
`LICENSE` is preserved in the vendored directory and every file keeps
its `The Go Authors` copyright header.
- Stdlib `text/template/parse` is reused unchanged (the parser has no
`MethodByName`/`evalField` edge, so it does not defeat DCE).
- `helper.go` (`ParseFiles`/`ParseGlob`/`ParseFS`) is dropped — the OPA
call sites only need `New`/`Parse`/`Execute`, and nothing in the kept
files references it.
- **The only edit to the copied code** is removing the `MethodByName`
branch in `exec.go`'s `evalField` (method resolution on the data value).
Everything else is byte-identical, so re-syncing to a newer Go release
is a diff-and-reapply of that single branch removal.
- `internal/gojsonschema` (commit 1) and `v1/topdown` (commit 2) import
the vendored package. The gojsonschema engine is retained in full, so
`ErrorTemplateFuncs` (its `FuncMap` extension point) keeps working —
**no public symbol is removed**.

Rego values and gojsonschema `ErrorDetails` decode to
`map[string]any`/`[]any`/scalars, which have no methods, so removing
method resolution is a provable no-op for these callers.

### Notes to assist PR review:

- **Diff review tip**:
`doc.go`/`funcs.go`/`option.go`/`template.go`/`internal/fmtsort/sort.go`
are **byte-identical** to the Go 1.25.8 originals. Only `exec.go`
differs, in exactly two hunks: the `internal/fmtsort` → vendored import
path, and the removed `MethodByName` block (replaced by a comment
explaining the DCE rationale).
- **Fidelity — render_template**: the `rendertemplate` conformance cases
(incl. `complex` range/if/vars, `simpleint` `%v`, `missingkey` →
`<undefined>`) pass **unchanged**.
- **Fidelity — gojsonschema**: same engine (method-less),
validation-error output unchanged; existing `internal/gojsonschema` and
`v1/ast` tests pass.
- **Tests**: `TestNoStdlibTextTemplateImport` in both
`internal/gojsonschema` and `v1/topdown` scans every non-test file and
asserts none import stdlib `text/template`/`html/template`. `go build
./...`, `go vet ./...` OK; `go mod tidy` is a no-op.
- **Lint**: the vendored directory is added to the golangci-lint path
exclusions, mirroring the existing `internal/gojsonschema` precedent —
the copy is verbatim stdlib, and linting it against OPA's house rules
would force divergence from upstream Go (it trips ~31 stdlib-idiom
issues) and break the diff-and-reapply re-sync.
- **Attribution**: the vendored code is Go stdlib only (BSD, `The Go
Authors`); it contains no third-party/DataDog code.

### Further comments:

- **Scope**: this restores method-level DCE for embedders of OPA's
**compiler/eval**. The standalone `opa` binary additionally links
`v1/server`, which imports `html/template` (a wrapper over
`text/template`) — a separate, independent edge left as a follow-up.
Embedders that don't link the server (the common case) get the full win
from this PR.
- Root cause: golang/go#72895. Closes #7903 for compiler/eval embedders.

---------

Signed-off-by: Dick Childress <dick.childress@icearp.net>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

.golangci.yaml

@@ -102,6 +102,7 @@ linters:
         path: _test\.go
     paths:
       - internal/gojsonschema
+      - internal/methodlesstemplate
       - node_modules
 issues:
   # don't hide issues in CI runs because they are the same type

build/regen-methodless-template.sh

@@ -0,0 +1,121 @@
+#!/usr/bin/env bash
+# Regenerate internal/methodlesstemplate from the Go stdlib text/template.
+#
+# internal/methodlesstemplate is a verbatim copy of the standard library's
+# text/template with exactly ONE behavioral edit: the method-calls-on-data
+# branch in exec.go's evalField (reflect.Value.MethodByName) is removed. A
+# reachable non-constant MethodByName forces the Go linker to disable
+# method-level dead-code elimination for the whole binary (golang/go#72895), so
+# eliding it is what lets OPA embedders shed the unused reflected method surface
+# (#7903). Rego values and gojsonschema ErrorDetails decode to
+# map[string]any/[]any/scalars, which have no methods, so the elision is a no-op.
+#
+# This script re-syncs that copy to whatever Go toolchain `go` resolves. Bump the
+# vendored version by running it under a newer toolchain, e.g.:
+#   GOTOOLCHAIN=go1.26.0 build/regen-methodless-template.sh
+#
+# It copies the stdlib files verbatim, then re-applies the two-hunk exec.go edit
+# (below) via `git apply`. helper.go and *_test.go are intentionally NOT vendored
+# (OPA only needs New/Parse/Execute), and text/template/parse is reused from the
+# stdlib via its normal import, so it is not copied here.
+#
+# If `git apply` fails, the stdlib changed the import block or the evalField
+# region this edit targets: re-derive the elision by hand and update the PATCH
+# heredoc at the bottom of this script.
+
+set -euo pipefail
+
+REPO_ROOT="$(git rev-parse --show-toplevel)"
+cd "$REPO_ROOT"
+
+if [[ "$(go list -m 2>/dev/null)" != "github.com/open-policy-agent/opa" ]]; then
+	echo "error: must run inside the open-policy-agent/opa module" >&2
+	exit 1
+fi
+
+GOROOT="$(go env GOROOT)"
+TMPL_SRC="$GOROOT/src/text/template"
+PKG="internal/methodlesstemplate"
+
+echo "Regenerating $PKG from $(go version)"
+
+# Verbatim stdlib copies. exec.go is patched below; the rest are byte-identical.
+mkdir -p "$PKG/internal/fmtsort"
+for f in doc.go exec.go funcs.go option.go template.go; do
+	cp -f "$TMPL_SRC/$f" "$PKG/$f"
+done
+cp -f "$GOROOT/src/internal/fmtsort/sort.go" "$PKG/internal/fmtsort/sort.go"
+cp -f "$GOROOT/LICENSE" "$PKG/LICENSE"
+
+# Re-apply the single intended edit to exec.go: swap the internal/fmtsort import
+# for the vendored one, and remove the MethodByName data-method branch.
+# --recount tolerates line-number drift from unrelated stdlib changes; it fails
+# only if the patched context text itself changed.
+if ! git apply --recount <<'PATCH'
+--- a/internal/methodlesstemplate/exec.go
++++ b/internal/methodlesstemplate/exec.go
+@@ -7,12 +7,13 @@
+ import (
+ 	"errors"
+ 	"fmt"
+-	"internal/fmtsort"
+ 	"io"
+ 	"reflect"
+ 	"runtime"
+ 	"strings"
+ 	"text/template/parse"
++
++	"github.com/open-policy-agent/opa/internal/methodlesstemplate/internal/fmtsort"
+ )
+
+ // maxExecDepth specifies the maximum stack depth of templates within
+@@ -689,21 +690,18 @@
+ 	typ := receiver.Type()
+ 	receiver, isNil := indirect(receiver)
+ 	if receiver.Kind() == reflect.Interface && isNil {
+-		// Calling a method on a nil interface can't work. The
+-		// MethodByName method call below would panic.
++		// Indexing into a nil interface can't work.
+ 		s.errorf("nil pointer evaluating %s.%s", typ, fieldName)
+ 		return zero
+ 	}
+
+-	// Unless it's an interface, need to get to a value of type *T to guarantee
+-	// we see all methods of T and *T.
+-	ptr := receiver
+-	if ptr.Kind() != reflect.Interface && ptr.Kind() != reflect.Pointer && ptr.CanAddr() {
+-		ptr = ptr.Addr()
+-	}
+-	if method := ptr.MethodByName(fieldName); method.IsValid() {
+-		return s.evalCall(dot, method, false, node, fieldName, args, final)
+-	}
++	// OPA-DCE (#7903): the upstream text/template resolves methods on the data
++	// value here via reflect.Value.MethodByName. A reachable non-constant
++	// MethodByName disables the Go linker's method-level dead-code elimination
++	// binary-wide (golang/go#72895), so that branch is deliberately removed.
++	// Rego values (and gojsonschema ErrorDetails) decode to
++	// map[string]any/[]any/scalars, which have no methods, so field/element
++	// resolution below is the only path OPA's callers need.
+ 	hasArgs := len(args) > 1 || !isMissing(final)
+ 	// It's not a method; must be a field of a struct or an element of a map.
+ 	switch receiver.Kind() {
+PATCH
+then
+	echo "error: could not apply the method-elision patch to exec.go." >&2
+	echo "       The stdlib import block or evalField region changed; re-derive" >&2
+	echo "       the edit by hand and update the PATCH heredoc in $0." >&2
+	exit 1
+fi
+
+gofmt -w "$PKG"
+
+# Fidelity guard: the whole point is that no reflect method-call on data survives.
+if grep -rn '\.MethodByName(' "$PKG"; then
+	echo "error: $PKG still contains a .MethodByName( call after patching." >&2
+	exit 1
+fi
+
+go build ./"$PKG"/...
+
+echo "Done. Vendored $PKG from $(go version)."
+echo "Review 'git diff $PKG' before committing; run 'make go-test' for full verification."

internal/gojsonschema/errors.go

@@ -4,7 +4,14 @@ package gojsonschema
 import (
 	"bytes"
 	"sync"
-	"text/template"
+
+	// A method-less copy of text/template (see internal/methodlesstemplate). Locale
+	// format strings expand only simple {{.field}} placeholders over ErrorDetails
+	// (map[string]any), which has no methods, so eliding method calls is a no-op
+	// here; it keeps text/template's evalField MethodByName off the reachable
+	// graph, which otherwise disables the Go linker's method-level dead-code
+	// elimination binary-wide (golang/go#72895, #7903).
+	template "github.com/open-policy-agent/opa/internal/methodlesstemplate"
 )
 
 var errorTemplates = errorTemplate{template.New("errors-new"), sync.RWMutex{}}

internal/gojsonschema/errors_dce_test.go

@@ -0,0 +1,46 @@
+package gojsonschema
+
+import (
+	"go/parser"
+	"go/token"
+	"os"
+	"strings"
+	"testing"
+)
+
+// TestNoStdlibTextTemplateImport guards the linker dead-code-elimination fix
+// (#7903): the error formatter uses a method-less template package, never stdlib
+// text/template or html/template. Their evalField reaches
+// reflect.Value.MethodByName, whose reachability disables method-level DCE for
+// the whole binary of every OPA embedder (golang/go#72895). gojsonschema is
+// reached unconditionally from ast.Compiler.Compile, so a stdlib import here
+// keeps the trigger live for every compiler embedder. Scans every non-test file
+// in the package so the guard holds even if the import moves.
+func TestNoStdlibTextTemplateImport(t *testing.T) {
+	entries, err := os.ReadDir(".")
+	if err != nil {
+		t.Fatal(err)
+	}
+	fset := token.NewFileSet()
+	scanned := 0
+	for _, e := range entries {
+		name := e.Name()
+		if e.IsDir() || !strings.HasSuffix(name, ".go") || strings.HasSuffix(name, "_test.go") {
+			continue
+		}
+		f, err := parser.ParseFile(fset, name, nil, parser.ImportsOnly)
+		if err != nil {
+			t.Fatalf("parse %s: %v", name, err)
+		}
+		scanned++
+		for _, imp := range f.Imports {
+			switch strings.Trim(imp.Path.Value, `"`) {
+			case "text/template", "html/template":
+				t.Errorf("%s imports %s — reintroduces the linker DCE-defeat trigger", name, imp.Path.Value)
+			}
+		}
+	}
+	if scanned == 0 {
+		t.Fatal("scanned no package files; test ran from the wrong directory")
+	}
+}

internal/gojsonschema/schema.go

@@ -31,7 +31,10 @@ import (
 	"errors"
 	"math/big"
 	"regexp"
-	"text/template"
+
+	// Method-less copy of text/template; see the import note in errors.go
+	// (golang/go#72895, #7903). ErrorTemplateFuncs below is its FuncMap.
+	template "github.com/open-policy-agent/opa/internal/methodlesstemplate"
 
 	"github.com/xeipuuv/gojsonreference"
 )

internal/methodlesstemplate/LICENSE

@@ -0,0 +1,27 @@
+Copyright 2009 The Go Authors.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+   * Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+   * Redistributions in binary form must reproduce the above
+copyright notice, this list of conditions and the following disclaimer
+in the documentation and/or other materials provided with the
+distribution.
+   * Neither the name of Google LLC nor the names of its
+contributors may be used to endorse or promote products derived from
+this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

internal/methodlesstemplate/README.md

@@ -0,0 +1,28 @@
+# methodlesstemplate
+
+A vendored copy of the Go standard library's `text/template`, with **one** behavioral change: the
+method-calls-on-data branch in `exec.go`'s `evalField` (`reflect.Value.MethodByName`) is removed.
+
+A reachable, non-constant `reflect.Value.MethodByName` makes the Go linker disable method-level
+dead-code elimination for the **entire binary** (golang/go#72895). OPA reaches `text/template` from
+its compiler (schema errors) and the `strings.render_template` builtin, so that one edge retains the
+full reflected method surface of every embedder — a large binary-size regression (#7903). Rego
+values and gojsonschema `ErrorDetails` decode to `map[string]any` / `[]any` / scalars, which have no
+methods, so eliding the data-method lookup is a behavioral no-op while restoring DCE.
+
+Only `exec.go` differs from the upstream stdlib. `doc.go`, `funcs.go`, `option.go`, `template.go`,
+and `internal/fmtsort/sort.go` are byte-identical to their Go release; `text/template/parse` is
+reused via its normal import. `helper.go` (ParseFiles/ParseGlob/ParseFS) is intentionally not
+vendored. Go's BSD `LICENSE` and per-file copyright headers are preserved.
+
+## Regenerating
+
+Do not hand-edit these files. To re-sync to a new Go release, run under the target toolchain:
+
+```
+GOTOOLCHAIN=go1.26.0 build/regen-methodless-template.sh
+```
+
+The script copies the stdlib files verbatim and re-applies the single method-elision edit. If the
+edit no longer applies (the stdlib changed that region), the script fails and the elision must be
+re-derived and the patch in the script updated.

internal/methodlesstemplate/doc.go

@@ -0,0 +1,502 @@
+// Copyright 2011 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+/*
+Package template implements data-driven templates for generating textual output.
+
+To generate HTML output, see [html/template], which has the same interface
+as this package but automatically secures HTML output against certain attacks.
+
+Templates are executed by applying them to a data structure. Annotations in the
+template refer to elements of the data structure (typically a field of a struct
+or a key in a map) to control execution and derive values to be displayed.
+Execution of the template walks the structure and sets the cursor, represented
+by a period '.' and called "dot", to the value at the current location in the
+structure as execution proceeds.
+
+The security model used by this package assumes that template authors are
+trusted. The package does not auto-escape output, so injecting code into
+a template can lead to arbitrary code execution if the template is executed
+by an untrusted source.
+
+The input text for a template is UTF-8-encoded text in any format.
+"Actions"--data evaluations or control structures--are delimited by
+"{{" and "}}"; all text outside actions is copied to the output unchanged.
+
+Once parsed, a template may be executed safely in parallel, although if parallel
+executions share a Writer the output may be interleaved.
+
+Here is a trivial example that prints "17 items are made of wool".
+
+	type Inventory struct {
+		Material string
+		Count    uint
+	}
+	sweaters := Inventory{"wool", 17}
+	tmpl, err := template.New("test").Parse("{{.Count}} items are made of {{.Material}}")
+	if err != nil { panic(err) }
+	err = tmpl.Execute(os.Stdout, sweaters)
+	if err != nil { panic(err) }
+
+More intricate examples appear below.
+
+Text and spaces
+
+By default, all text between actions is copied verbatim when the template is
+executed. For example, the string " items are made of " in the example above
+appears on standard output when the program is run.
+
+However, to aid in formatting template source code, if an action's left
+delimiter (by default "{{") is followed immediately by a minus sign and white
+space, all trailing white space is trimmed from the immediately preceding text.
+Similarly, if the right delimiter ("}}") is preceded by white space and a minus
+sign, all leading white space is trimmed from the immediately following text.
+In these trim markers, the white space must be present:
+"{{- 3}}" is like "{{3}}" but trims the immediately preceding text, while
+"{{-3}}" parses as an action containing the number -3.
+
+For instance, when executing the template whose source is
+
+	"{{23 -}} < {{- 45}}"
+
+the generated output would be
+
+	"23<45"
+
+For this trimming, the definition of white space characters is the same as in Go:
+space, horizontal tab, carriage return, and newline.
+
+Actions
+
+Here is the list of actions. "Arguments" and "pipelines" are evaluations of
+data, defined in detail in the corresponding sections that follow.
+
+*/
+//	{{/* a comment */}}
+//	{{- /* a comment with white space trimmed from preceding and following text */ -}}
+//		A comment; discarded. May contain newlines.
+//		Comments do not nest and must start and end at the
+//		delimiters, as shown here.
+/*
+
+	{{pipeline}}
+		The default textual representation (the same as would be
+		printed by fmt.Print) of the value of the pipeline is copied
+		to the output.
+
+	{{if pipeline}} T1 {{end}}
+		If the value of the pipeline is empty, no output is generated;
+		otherwise, T1 is executed. The empty values are false, 0, any
+		nil pointer or interface value, and any array, slice, map, or
+		string of length zero.
+		Dot is unaffected.
+
+	{{if pipeline}} T1 {{else}} T0 {{end}}
+		If the value of the pipeline is empty, T0 is executed;
+		otherwise, T1 is executed. Dot is unaffected.
+
+	{{if pipeline}} T1 {{else if pipeline}} T0 {{end}}
+		To simplify the appearance of if-else chains, the else action
+		of an if may include another if directly; the effect is exactly
+		the same as writing
+			{{if pipeline}} T1 {{else}}{{if pipeline}} T0 {{end}}{{end}}
+
+	{{range pipeline}} T1 {{end}}
+		The value of the pipeline must be an array, slice, map, iter.Seq,
+		iter.Seq2, integer or channel.
+		If the value of the pipeline has length zero, nothing is output;
+		otherwise, dot is set to the successive elements of the array,
+		slice, or map and T1 is executed. If the value is a map and the
+		keys are of basic type with a defined order, the elements will be
+		visited in sorted key order.
+
+	{{range pipeline}} T1 {{else}} T0 {{end}}
+		The value of the pipeline must be an array, slice, map, iter.Seq,
+		iter.Seq2, integer or channel.
+		If the value of the pipeline has length zero, dot is unaffected and
+		T0 is executed; otherwise, dot is set to the successive elements
+		of the array, slice, or map and T1 is executed.
+
+	{{break}}
+		The innermost {{range pipeline}} loop is ended early, stopping the
+		current iteration and bypassing all remaining iterations.
+
+	{{continue}}
+		The current iteration of the innermost {{range pipeline}} loop is
+		stopped, and the loop starts the next iteration.
+
+	{{template "name"}}
+		The template with the specified name is executed with nil data.
+
+	{{template "name" pipeline}}
+		The template with the specified name is executed with dot set
+		to the value of the pipeline.
+
+	{{block "name" pipeline}} T1 {{end}}
+		A block is shorthand for defining a template
+			{{define "name"}} T1 {{end}}
+		and then executing it in place
+			{{template "name" pipeline}}
+		The typical use is to define a set of root templates that are
+		then customized by redefining the block templates within.
+
+	{{with pipeline}} T1 {{end}}
+		If the value of the pipeline is empty, no output is generated;
+		otherwise, dot is set to the value of the pipeline and T1 is
+		executed.
+
+	{{with pipeline}} T1 {{else}} T0 {{end}}
+		If the value of the pipeline is empty, dot is unaffected and T0
+		is executed; otherwise, dot is set to the value of the pipeline
+		and T1 is executed.
+
+	{{with pipeline}} T1 {{else with pipeline}} T0 {{end}}
+		To simplify the appearance of with-else chains, the else action
+		of a with may include another with directly; the effect is exactly
+		the same as writing
+			{{with pipeline}} T1 {{else}}{{with pipeline}} T0 {{end}}{{end}}
+
+
+Arguments
+
+An argument is a simple value, denoted by one of the following.
+
+	- A boolean, string, character, integer, floating-point, imaginary
+	  or complex constant in Go syntax. These behave like Go's untyped
+	  constants. Note that, as in Go, whether a large integer constant
+	  overflows when assigned or passed to a function can depend on whether
+	  the host machine's ints are 32 or 64 bits.
+	- The keyword nil, representing an untyped Go nil.
+	- The character '.' (period):
+
+		.
+
+	  The result is the value of dot.
+	- A variable name, which is a (possibly empty) alphanumeric string
+	  preceded by a dollar sign, such as
+
+		$piOver2
+
+	  or
+
+		$
+
+	  The result is the value of the variable.
+	  Variables are described below.
+	- The name of a field of the data, which must be a struct, preceded
+	  by a period, such as
+
+		.Field
+
+	  The result is the value of the field. Field invocations may be
+	  chained:
+
+	    .Field1.Field2
+
+	  Fields can also be evaluated on variables, including chaining:
+
+	    $x.Field1.Field2
+	- The name of a key of the data, which must be a map, preceded
+	  by a period, such as
+
+		.Key
+
+	  The result is the map element value indexed by the key.
+	  Key invocations may be chained and combined with fields to any
+	  depth:
+
+	    .Field1.Key1.Field2.Key2
+
+	  Although the key must be an alphanumeric identifier, unlike with
+	  field names they do not need to start with an upper case letter.
+	  Keys can also be evaluated on variables, including chaining:
+
+	    $x.key1.key2
+	- The name of a niladic method of the data, preceded by a period,
+	  such as
+
+		.Method
+
+	  The result is the value of invoking the method with dot as the
+	  receiver, dot.Method(). Such a method must have one return value (of
+	  any type) or two return values, the second of which is an error.
+	  If it has two and the returned error is non-nil, execution terminates
+	  and an error is returned to the caller as the value of Execute.
+	  Method invocations may be chained and combined with fields and keys
+	  to any depth:
+
+	    .Field1.Key1.Method1.Field2.Key2.Method2
+
+	  Methods can also be evaluated on variables, including chaining:
+
+	    $x.Method1.Field
+	- The name of a niladic function, such as
+
+		fun
+
+	  The result is the value of invoking the function, fun(). The return
+	  types and values behave as in methods. Functions and function
+	  names are described below.
+	- A parenthesized instance of one the above, for grouping. The result
+	  may be accessed by a field or map key invocation.
+
+		print (.F1 arg1) (.F2 arg2)
+		(.StructValuedMethod "arg").Field
+
+Arguments may evaluate to any type; if they are pointers the implementation
+automatically indirects to the base type when required.
+If an evaluation yields a function value, such as a function-valued
+field of a struct, the function is not invoked automatically, but it
+can be used as a truth value for an if action and the like. To invoke
+it, use the call function, defined below.
+
+Pipelines
+
+A pipeline is a possibly chained sequence of "commands". A command is a simple
+value (argument) or a function or method call, possibly with multiple arguments:
+
+	Argument
+		The result is the value of evaluating the argument.
+	.Method [Argument...]
+		The method can be alone or the last element of a chain but,
+		unlike methods in the middle of a chain, it can take arguments.
+		The result is the value of calling the method with the
+		arguments:
+			dot.Method(Argument1, etc.)
+	functionName [Argument...]
+		The result is the value of calling the function associated
+		with the name:
+			function(Argument1, etc.)
+		Functions and function names are described below.
+
+A pipeline may be "chained" by separating a sequence of commands with pipeline
+characters '|'. In a chained pipeline, the result of each command is
+passed as the last argument of the following command. The output of the final
+command in the pipeline is the value of the pipeline.
+
+The output of a command will be either one value or two values, the second of
+which has type error. If that second value is present and evaluates to
+non-nil, execution terminates and the error is returned to the caller of
+Execute.
+
+Variables
+
+A pipeline inside an action may initialize a variable to capture the result.
+The initialization has syntax
+
+	$variable := pipeline
+
+where $variable is the name of the variable. An action that declares a
+variable produces no output.
+
+Variables previously declared can also be assigned, using the syntax
+
+	$variable = pipeline
+
+If a "range" action initializes a variable, the variable is set to the
+successive elements of the iteration. Also, a "range" may declare two
+variables, separated by a comma:
+
+	range $index, $element := pipeline
+
+in which case $index and $element are set to the successive values of the
+array/slice index or map key and element, respectively. Note that if there is
+only one variable, it is assigned the element; this is opposite to the
+convention in Go range clauses.
+
+A variable's scope extends to the "end" action of the control structure ("if",
+"with", or "range") in which it is declared, or to the end of the template if
+there is no such control structure. A template invocation does not inherit
+variables from the point of its invocation.
+
+When execution begins, $ is set to the data argument passed to Execute, that is,
+to the starting value of dot.
+
+Examples
+
+Here are some example one-line templates demonstrating pipelines and variables.
+All produce the quoted word "output":
+
+	{{"\"output\""}}
+		A string constant.
+	{{`"output"`}}
+		A raw string constant.
+	{{printf "%q" "output"}}
+		A function call.
+	{{"output" | printf "%q"}}
+		A function call whose final argument comes from the previous
+		command.
+	{{printf "%q" (print "out" "put")}}
+		A parenthesized argument.
+	{{"put" | printf "%s%s" "out" | printf "%q"}}
+		A more elaborate call.
+	{{"output" | printf "%s" | printf "%q"}}
+		A longer chain.
+	{{with "output"}}{{printf "%q" .}}{{end}}
+		A with action using dot.
+	{{with $x := "output" | printf "%q"}}{{$x}}{{end}}
+		A with action that creates and uses a variable.
+	{{with $x := "output"}}{{printf "%q" $x}}{{end}}
+		A with action that uses the variable in another action.
+	{{with $x := "output"}}{{$x | printf "%q"}}{{end}}
+		The same, but pipelined.
+
+Functions
+
+During execution functions are found in two function maps: first in the
+template, then in the global function map. By default, no functions are defined
+in the template but the Funcs method can be used to add them.
+
+Predefined global functions are named as follows.
+
+	and
+		Returns the boolean AND of its arguments by returning the
+		first empty argument or the last argument. That is,
+		"and x y" behaves as "if x then y else x."
+		Evaluation proceeds through the arguments left to right
+		and returns when the result is determined.
+	call
+		Returns the result of calling the first argument, which
+		must be a function, with the remaining arguments as parameters.
+		Thus "call .X.Y 1 2" is, in Go notation, dot.X.Y(1, 2) where
+		Y is a func-valued field, map entry, or the like.
+		The first argument must be the result of an evaluation
+		that yields a value of function type (as distinct from
+		a predefined function such as print). The function must
+		return either one or two result values, the second of which
+		is of type error. If the arguments don't match the function
+		or the returned error value is non-nil, execution stops.
+	html
+		Returns the escaped HTML equivalent of the textual
+		representation of its arguments. This function is unavailable
+		in html/template, with a few exceptions.
+	index
+		Returns the result of indexing its first argument by the
+		following arguments. Thus "index x 1 2 3" is, in Go syntax,
+		x[1][2][3]. Each indexed item must be a map, slice, or array.
+	slice
+		slice returns the result of slicing its first argument by the
+		remaining arguments. Thus "slice x 1 2" is, in Go syntax, x[1:2],
+		while "slice x" is x[:], "slice x 1" is x[1:], and "slice x 1 2 3"
+		is x[1:2:3]. The first argument must be a string, slice, or array.
+	js
+		Returns the escaped JavaScript equivalent of the textual
+		representation of its arguments.
+	len
+		Returns the integer length of its argument.
+	not
+		Returns the boolean negation of its single argument.
+	or
+		Returns the boolean OR of its arguments by returning the
+		first non-empty argument or the last argument, that is,
+		"or x y" behaves as "if x then x else y".
+		Evaluation proceeds through the arguments left to right
+		and returns when the result is determined.
+	print
+		An alias for fmt.Sprint
+	printf
+		An alias for fmt.Sprintf
+	println
+		An alias for fmt.Sprintln
+	urlquery
+		Returns the escaped value of the textual representation of
+		its arguments in a form suitable for embedding in a URL query.
+		This function is unavailable in html/template, with a few
+		exceptions.
+
+The boolean functions take any zero value to be false and a non-zero
+value to be true.
+
+There is also a set of binary comparison operators defined as
+functions:
+
+	eq
+		Returns the boolean truth of arg1 == arg2
+	ne
+		Returns the boolean truth of arg1 != arg2
+	lt
+		Returns the boolean truth of arg1 < arg2
+	le
+		Returns the boolean truth of arg1 <= arg2
+	gt
+		Returns the boolean truth of arg1 > arg2
+	ge
+		Returns the boolean truth of arg1 >= arg2
+
+For simpler multi-way equality tests, eq (only) accepts two or more
+arguments and compares the second and subsequent to the first,
+returning in effect
+
+	arg1==arg2 || arg1==arg3 || arg1==arg4 ...
+
+(Unlike with || in Go, however, eq is a function call and all the
+arguments will be evaluated.)
+
+The comparison functions work on any values whose type Go defines as
+comparable. For basic types such as integers, the rules are relaxed:
+size and exact type are ignored, so any integer value, signed or unsigned,
+may be compared with any other integer value. (The arithmetic value is compared,
+not the bit pattern, so all negative integers are less than all unsigned integers.)
+However, as usual, one may not compare an int with a float32 and so on.
+
+Associated templates
+
+Each template is named by a string specified when it is created. Also, each
+template is associated with zero or more other templates that it may invoke by
+name; such associations are transitive and form a name space of templates.
+
+A template may use a template invocation to instantiate another associated
+template; see the explanation of the "template" action above. The name must be
+that of a template associated with the template that contains the invocation.
+
+Nested template definitions
+
+When parsing a template, another template may be defined and associated with the
+template being parsed. Template definitions must appear at the top level of the
+template, much like global variables in a Go program.
+
+The syntax of such definitions is to surround each template declaration with a
+"define" and "end" action.
+
+The define action names the template being created by providing a string
+constant. Here is a simple example:
+
+	{{define "T1"}}ONE{{end}}
+	{{define "T2"}}TWO{{end}}
+	{{define "T3"}}{{template "T1"}} {{template "T2"}}{{end}}
+	{{template "T3"}}
+
+This defines two templates, T1 and T2, and a third T3 that invokes the other two
+when it is executed. Finally it invokes T3. If executed this template will
+produce the text
+
+	ONE TWO
+
+By construction, a template may reside in only one association. If it's
+necessary to have a template addressable from multiple associations, the
+template definition must be parsed multiple times to create distinct *Template
+values, or must be copied with [Template.Clone] or [Template.AddParseTree].
+
+Parse may be called multiple times to assemble the various associated templates;
+see [ParseFiles], [ParseGlob], [Template.ParseFiles] and [Template.ParseGlob]
+for simple ways to parse related templates stored in files.
+
+A template may be executed directly or through [Template.ExecuteTemplate], which executes
+an associated template identified by name. To invoke our example above, we
+might write,
+
+	err := tmpl.Execute(os.Stdout, "no data needed")
+	if err != nil {
+		log.Fatalf("execution failed: %s", err)
+	}
+
+or to invoke a particular template explicitly by name,
+
+	err := tmpl.ExecuteTemplate(os.Stdout, "T2", "no data needed")
+	if err != nil {
+		log.Fatalf("execution failed: %s", err)
+	}
+
+*/
+package template

internal/methodlesstemplate/funcs.go

@@ -0,0 +1,783 @@
+// Copyright 2011 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package template
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"net/url"
+	"reflect"
+	"strings"
+	"sync"
+	"unicode"
+	"unicode/utf8"
+)
+
+// FuncMap is the type of the map defining the mapping from names to functions.
+// Each function must have either a single return value, or two return values of
+// which the second has type error. In that case, if the second (error)
+// return value evaluates to non-nil during execution, execution terminates and
+// Execute returns that error.
+//
+// Errors returned by Execute wrap the underlying error; call [errors.As] to
+// unwrap them.
+//
+// When template execution invokes a function with an argument list, that list
+// must be assignable to the function's parameter types. Functions meant to
+// apply to arguments of arbitrary type can use parameters of type interface{} or
+// of type [reflect.Value]. Similarly, functions meant to return a result of arbitrary
+// type can return interface{} or [reflect.Value].
+type FuncMap map[string]any
+
+// builtins returns the FuncMap.
+// It is not a global variable so the linker can dead code eliminate
+// more when this isn't called. See golang.org/issue/36021.
+// TODO: revert this back to a global map once golang.org/issue/2559 is fixed.
+func builtins() FuncMap {
+	return FuncMap{
+		"and":      and,
+		"call":     emptyCall,
+		"html":     HTMLEscaper,
+		"index":    index,
+		"slice":    slice,
+		"js":       JSEscaper,
+		"len":      length,
+		"not":      not,
+		"or":       or,
+		"print":    fmt.Sprint,
+		"printf":   fmt.Sprintf,
+		"println":  fmt.Sprintln,
+		"urlquery": URLQueryEscaper,
+
+		// Comparisons
+		"eq": eq, // ==
+		"ge": ge, // >=
+		"gt": gt, // >
+		"le": le, // <=
+		"lt": lt, // <
+		"ne": ne, // !=
+	}
+}
+
+var builtinFuncsOnce struct {
+	sync.Once
+	v map[string]reflect.Value
+}
+
+// builtinFuncsOnce lazily computes & caches the builtinFuncs map.
+// TODO: revert this back to a global map once golang.org/issue/2559 is fixed.
+func builtinFuncs() map[string]reflect.Value {
+	builtinFuncsOnce.Do(func() {
+		builtinFuncsOnce.v = createValueFuncs(builtins())
+	})
+	return builtinFuncsOnce.v
+}
+
+// createValueFuncs turns a FuncMap into a map[string]reflect.Value
+func createValueFuncs(funcMap FuncMap) map[string]reflect.Value {
+	m := make(map[string]reflect.Value)
+	addValueFuncs(m, funcMap)
+	return m
+}
+
+// addValueFuncs adds to values the functions in funcs, converting them to reflect.Values.
+func addValueFuncs(out map[string]reflect.Value, in FuncMap) {
+	for name, fn := range in {
+		if !goodName(name) {
+			panic(fmt.Errorf("function name %q is not a valid identifier", name))
+		}
+		v := reflect.ValueOf(fn)
+		if v.Kind() != reflect.Func {
+			panic("value for " + name + " not a function")
+		}
+		if err := goodFunc(name, v.Type()); err != nil {
+			panic(err)
+		}
+		out[name] = v
+	}
+}
+
+// addFuncs adds to values the functions in funcs. It does no checking of the input -
+// call addValueFuncs first.
+func addFuncs(out, in FuncMap) {
+	for name, fn := range in {
+		out[name] = fn
+	}
+}
+
+// goodFunc reports whether the function or method has the right result signature.
+func goodFunc(name string, typ reflect.Type) error {
+	// We allow functions with 1 result or 2 results where the second is an error.
+	switch numOut := typ.NumOut(); {
+	case numOut == 1:
+		return nil
+	case numOut == 2 && typ.Out(1) == errorType:
+		return nil
+	case numOut == 2:
+		return fmt.Errorf("invalid function signature for %s: second return value should be error; is %s", name, typ.Out(1))
+	default:
+		return fmt.Errorf("function %s has %d return values; should be 1 or 2", name, typ.NumOut())
+	}
+}
+
+// goodName reports whether the function name is a valid identifier.
+func goodName(name string) bool {
+	if name == "" {
+		return false
+	}
+	for i, r := range name {
+		switch {
+		case r == '_':
+		case i == 0 && !unicode.IsLetter(r):
+			return false
+		case !unicode.IsLetter(r) && !unicode.IsDigit(r):
+			return false
+		}
+	}
+	return true
+}
+
+// findFunction looks for a function in the template, and global map.
+func findFunction(name string, tmpl *Template) (v reflect.Value, isBuiltin, ok bool) {
+	if tmpl != nil && tmpl.common != nil {
+		tmpl.muFuncs.RLock()
+		defer tmpl.muFuncs.RUnlock()
+		if fn := tmpl.execFuncs[name]; fn.IsValid() {
+			return fn, false, true
+		}
+	}
+	if fn := builtinFuncs()[name]; fn.IsValid() {
+		return fn, true, true
+	}
+	return reflect.Value{}, false, false
+}
+
+// prepareArg checks if value can be used as an argument of type argType, and
+// converts an invalid value to appropriate zero if possible.
+func prepareArg(value reflect.Value, argType reflect.Type) (reflect.Value, error) {
+	if !value.IsValid() {
+		if !canBeNil(argType) {
+			return reflect.Value{}, fmt.Errorf("value is nil; should be of type %s", argType)
+		}
+		value = reflect.Zero(argType)
+	}
+	if value.Type().AssignableTo(argType) {
+		return value, nil
+	}
+	if intLike(value.Kind()) && intLike(argType.Kind()) && value.Type().ConvertibleTo(argType) {
+		value = value.Convert(argType)
+		return value, nil
+	}
+	return reflect.Value{}, fmt.Errorf("value has type %s; should be %s", value.Type(), argType)
+}
+
+func intLike(typ reflect.Kind) bool {
+	switch typ {
+	case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
+		return true
+	case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
+		return true
+	}
+	return false
+}
+
+// indexArg checks if a reflect.Value can be used as an index, and converts it to int if possible.
+func indexArg(index reflect.Value, cap int) (int, error) {
+	var x int64
+	switch index.Kind() {
+	case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
+		x = index.Int()
+	case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
+		x = int64(index.Uint())
+	case reflect.Invalid:
+		return 0, fmt.Errorf("cannot index slice/array with nil")
+	default:
+		return 0, fmt.Errorf("cannot index slice/array with type %s", index.Type())
+	}
+	if x < 0 || int(x) < 0 || int(x) > cap {
+		return 0, fmt.Errorf("index out of range: %d", x)
+	}
+	return int(x), nil
+}
+
+// Indexing.
+
+// index returns the result of indexing its first argument by the following
+// arguments. Thus "index x 1 2 3" is, in Go syntax, x[1][2][3]. Each
+// indexed item must be a map, slice, or array.
+func index(item reflect.Value, indexes ...reflect.Value) (reflect.Value, error) {
+	item = indirectInterface(item)
+	if !item.IsValid() {
+		return reflect.Value{}, fmt.Errorf("index of untyped nil")
+	}
+	for _, index := range indexes {
+		index = indirectInterface(index)
+		var isNil bool
+		if item, isNil = indirect(item); isNil {
+			return reflect.Value{}, fmt.Errorf("index of nil pointer")
+		}
+		switch item.Kind() {
+		case reflect.Array, reflect.Slice, reflect.String:
+			x, err := indexArg(index, item.Len())
+			if err != nil {
+				return reflect.Value{}, err
+			}
+			item = item.Index(x)
+		case reflect.Map:
+			index, err := prepareArg(index, item.Type().Key())
+			if err != nil {
+				return reflect.Value{}, err
+			}
+			if x := item.MapIndex(index); x.IsValid() {
+				item = x
+			} else {
+				item = reflect.Zero(item.Type().Elem())
+			}
+		case reflect.Invalid:
+			// the loop holds invariant: item.IsValid()
+			panic("unreachable")
+		default:
+			return reflect.Value{}, fmt.Errorf("can't index item of type %s", item.Type())
+		}
+	}
+	return item, nil
+}
+
+// Slicing.
+
+// slice returns the result of slicing its first argument by the remaining
+// arguments. Thus "slice x 1 2" is, in Go syntax, x[1:2], while "slice x"
+// is x[:], "slice x 1" is x[1:], and "slice x 1 2 3" is x[1:2:3]. The first
+// argument must be a string, slice, or array.
+func slice(item reflect.Value, indexes ...reflect.Value) (reflect.Value, error) {
+	item = indirectInterface(item)
+	if !item.IsValid() {
+		return reflect.Value{}, fmt.Errorf("slice of untyped nil")
+	}
+	if len(indexes) > 3 {
+		return reflect.Value{}, fmt.Errorf("too many slice indexes: %d", len(indexes))
+	}
+	var cap int
+	switch item.Kind() {
+	case reflect.String:
+		if len(indexes) == 3 {
+			return reflect.Value{}, fmt.Errorf("cannot 3-index slice a string")
+		}
+		cap = item.Len()
+	case reflect.Array, reflect.Slice:
+		cap = item.Cap()
+	default:
+		return reflect.Value{}, fmt.Errorf("can't slice item of type %s", item.Type())
+	}
+	// set default values for cases item[:], item[i:].
+	idx := [3]int{0, item.Len()}
+	for i, index := range indexes {
+		x, err := indexArg(index, cap)
+		if err != nil {
+			return reflect.Value{}, err
+		}
+		idx[i] = x
+	}
+	// given item[i:j], make sure i <= j.
+	if idx[0] > idx[1] {
+		return reflect.Value{}, fmt.Errorf("invalid slice index: %d > %d", idx[0], idx[1])
+	}
+	if len(indexes) < 3 {
+		return item.Slice(idx[0], idx[1]), nil
+	}
+	// given item[i:j:k], make sure i <= j <= k.
+	if idx[1] > idx[2] {
+		return reflect.Value{}, fmt.Errorf("invalid slice index: %d > %d", idx[1], idx[2])
+	}
+	return item.Slice3(idx[0], idx[1], idx[2]), nil
+}
+
+// Length
+
+// length returns the length of the item, with an error if it has no defined length.
+func length(item reflect.Value) (int, error) {
+	item, isNil := indirect(item)
+	if isNil {
+		return 0, fmt.Errorf("len of nil pointer")
+	}
+	switch item.Kind() {
+	case reflect.Array, reflect.Chan, reflect.Map, reflect.Slice, reflect.String:
+		return item.Len(), nil
+	}
+	return 0, fmt.Errorf("len of type %s", item.Type())
+}
+
+// Function invocation
+
+func emptyCall(fn reflect.Value, args ...reflect.Value) reflect.Value {
+	panic("unreachable") // implemented as a special case in evalCall
+}
+
+// call returns the result of evaluating the first argument as a function.
+// The function must return 1 result, or 2 results, the second of which is an error.
+func call(name string, fn reflect.Value, args ...reflect.Value) (reflect.Value, error) {
+	fn = indirectInterface(fn)
+	if !fn.IsValid() {
+		return reflect.Value{}, fmt.Errorf("call of nil")
+	}
+	typ := fn.Type()
+	if typ.Kind() != reflect.Func {
+		return reflect.Value{}, fmt.Errorf("non-function %s of type %s", name, typ)
+	}
+
+	if err := goodFunc(name, typ); err != nil {
+		return reflect.Value{}, err
+	}
+	numIn := typ.NumIn()
+	var dddType reflect.Type
+	if typ.IsVariadic() {
+		if len(args) < numIn-1 {
+			return reflect.Value{}, fmt.Errorf("wrong number of args for %s: got %d want at least %d", name, len(args), numIn-1)
+		}
+		dddType = typ.In(numIn - 1).Elem()
+	} else {
+		if len(args) != numIn {
+			return reflect.Value{}, fmt.Errorf("wrong number of args for %s: got %d want %d", name, len(args), numIn)
+		}
+	}
+	argv := make([]reflect.Value, len(args))
+	for i, arg := range args {
+		arg = indirectInterface(arg)
+		// Compute the expected type. Clumsy because of variadics.
+		argType := dddType
+		if !typ.IsVariadic() || i < numIn-1 {
+			argType = typ.In(i)
+		}
+
+		var err error
+		if argv[i], err = prepareArg(arg, argType); err != nil {
+			return reflect.Value{}, fmt.Errorf("arg %d: %w", i, err)
+		}
+	}
+	return safeCall(fn, argv)
+}
+
+// safeCall runs fun.Call(args), and returns the resulting value and error, if
+// any. If the call panics, the panic value is returned as an error.
+func safeCall(fun reflect.Value, args []reflect.Value) (val reflect.Value, err error) {
+	defer func() {
+		if r := recover(); r != nil {
+			if e, ok := r.(error); ok {
+				err = e
+			} else {
+				err = fmt.Errorf("%v", r)
+			}
+		}
+	}()
+	ret := fun.Call(args)
+	if len(ret) == 2 && !ret[1].IsNil() {
+		return ret[0], ret[1].Interface().(error)
+	}
+	return ret[0], nil
+}
+
+// Boolean logic.
+
+func truth(arg reflect.Value) bool {
+	t, _ := isTrue(indirectInterface(arg))
+	return t
+}
+
+// and computes the Boolean AND of its arguments, returning
+// the first false argument it encounters, or the last argument.
+func and(arg0 reflect.Value, args ...reflect.Value) reflect.Value {
+	panic("unreachable") // implemented as a special case in evalCall
+}
+
+// or computes the Boolean OR of its arguments, returning
+// the first true argument it encounters, or the last argument.
+func or(arg0 reflect.Value, args ...reflect.Value) reflect.Value {
+	panic("unreachable") // implemented as a special case in evalCall
+}
+
+// not returns the Boolean negation of its argument.
+func not(arg reflect.Value) bool {
+	return !truth(arg)
+}
+
+// Comparison.
+
+// TODO: Perhaps allow comparison between signed and unsigned integers.
+
+var (
+	errBadComparisonType = errors.New("invalid type for comparison")
+	errNoComparison      = errors.New("missing argument for comparison")
+)
+
+type kind int
+
+const (
+	invalidKind kind = iota
+	boolKind
+	complexKind
+	intKind
+	floatKind
+	stringKind
+	uintKind
+)
+
+func basicKind(v reflect.Value) (kind, error) {
+	switch v.Kind() {
+	case reflect.Bool:
+		return boolKind, nil
+	case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
+		return intKind, nil
+	case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
+		return uintKind, nil
+	case reflect.Float32, reflect.Float64:
+		return floatKind, nil
+	case reflect.Complex64, reflect.Complex128:
+		return complexKind, nil
+	case reflect.String:
+		return stringKind, nil
+	}
+	return invalidKind, errBadComparisonType
+}
+
+// isNil returns true if v is the zero reflect.Value, or nil of its type.
+func isNil(v reflect.Value) bool {
+	if !v.IsValid() {
+		return true
+	}
+	switch v.Kind() {
+	case reflect.Chan, reflect.Func, reflect.Interface, reflect.Map, reflect.Pointer, reflect.Slice:
+		return v.IsNil()
+	}
+	return false
+}
+
+// canCompare reports whether v1 and v2 are both the same kind, or one is nil.
+// Called only when dealing with nillable types, or there's about to be an error.
+func canCompare(v1, v2 reflect.Value) bool {
+	k1 := v1.Kind()
+	k2 := v2.Kind()
+	if k1 == k2 {
+		return true
+	}
+	// We know the type can be compared to nil.
+	return k1 == reflect.Invalid || k2 == reflect.Invalid
+}
+
+// eq evaluates the comparison a == b || a == c || ...
+func eq(arg1 reflect.Value, arg2 ...reflect.Value) (bool, error) {
+	arg1 = indirectInterface(arg1)
+	if len(arg2) == 0 {
+		return false, errNoComparison
+	}
+	k1, _ := basicKind(arg1)
+	for _, arg := range arg2 {
+		arg = indirectInterface(arg)
+		k2, _ := basicKind(arg)
+		truth := false
+		if k1 != k2 {
+			// Special case: Can compare integer values regardless of type's sign.
+			switch {
+			case k1 == intKind && k2 == uintKind:
+				truth = arg1.Int() >= 0 && uint64(arg1.Int()) == arg.Uint()
+			case k1 == uintKind && k2 == intKind:
+				truth = arg.Int() >= 0 && arg1.Uint() == uint64(arg.Int())
+			default:
+				if arg1.IsValid() && arg.IsValid() {
+					return false, fmt.Errorf("incompatible types for comparison: %v and %v", arg1.Type(), arg.Type())
+				}
+			}
+		} else {
+			switch k1 {
+			case boolKind:
+				truth = arg1.Bool() == arg.Bool()
+			case complexKind:
+				truth = arg1.Complex() == arg.Complex()
+			case floatKind:
+				truth = arg1.Float() == arg.Float()
+			case intKind:
+				truth = arg1.Int() == arg.Int()
+			case stringKind:
+				truth = arg1.String() == arg.String()
+			case uintKind:
+				truth = arg1.Uint() == arg.Uint()
+			default:
+				if !canCompare(arg1, arg) {
+					return false, fmt.Errorf("non-comparable types %s: %v, %s: %v", arg1, arg1.Type(), arg.Type(), arg)
+				}
+				if isNil(arg1) || isNil(arg) {
+					truth = isNil(arg) == isNil(arg1)
+				} else {
+					if !arg.Type().Comparable() {
+						return false, fmt.Errorf("non-comparable type %s: %v", arg, arg.Type())
+					}
+					truth = arg1.Interface() == arg.Interface()
+				}
+			}
+		}
+		if truth {
+			return true, nil
+		}
+	}
+	return false, nil
+}
+
+// ne evaluates the comparison a != b.
+func ne(arg1, arg2 reflect.Value) (bool, error) {
+	// != is the inverse of ==.
+	equal, err := eq(arg1, arg2)
+	return !equal, err
+}
+
+// lt evaluates the comparison a < b.
+func lt(arg1, arg2 reflect.Value) (bool, error) {
+	arg1 = indirectInterface(arg1)
+	k1, err := basicKind(arg1)
+	if err != nil {
+		return false, err
+	}
+	arg2 = indirectInterface(arg2)
+	k2, err := basicKind(arg2)
+	if err != nil {
+		return false, err
+	}
+	truth := false
+	if k1 != k2 {
+		// Special case: Can compare integer values regardless of type's sign.
+		switch {
+		case k1 == intKind && k2 == uintKind:
+			truth = arg1.Int() < 0 || uint64(arg1.Int()) < arg2.Uint()
+		case k1 == uintKind && k2 == intKind:
+			truth = arg2.Int() >= 0 && arg1.Uint() < uint64(arg2.Int())
+		default:
+			return false, fmt.Errorf("incompatible types for comparison: %v and %v", arg1.Type(), arg2.Type())
+		}
+	} else {
+		switch k1 {
+		case boolKind, complexKind:
+			return false, errBadComparisonType
+		case floatKind:
+			truth = arg1.Float() < arg2.Float()
+		case intKind:
+			truth = arg1.Int() < arg2.Int()
+		case stringKind:
+			truth = arg1.String() < arg2.String()
+		case uintKind:
+			truth = arg1.Uint() < arg2.Uint()
+		default:
+			panic("invalid kind")
+		}
+	}
+	return truth, nil
+}
+
+// le evaluates the comparison <= b.
+func le(arg1, arg2 reflect.Value) (bool, error) {
+	// <= is < or ==.
+	lessThan, err := lt(arg1, arg2)
+	if lessThan || err != nil {
+		return lessThan, err
+	}
+	return eq(arg1, arg2)
+}
+
+// gt evaluates the comparison a > b.
+func gt(arg1, arg2 reflect.Value) (bool, error) {
+	// > is the inverse of <=.
+	lessOrEqual, err := le(arg1, arg2)
+	if err != nil {
+		return false, err
+	}
+	return !lessOrEqual, nil
+}
+
+// ge evaluates the comparison a >= b.
+func ge(arg1, arg2 reflect.Value) (bool, error) {
+	// >= is the inverse of <.
+	lessThan, err := lt(arg1, arg2)
+	if err != nil {
+		return false, err
+	}
+	return !lessThan, nil
+}
+
+// HTML escaping.
+
+var (
+	htmlQuot = []byte("&#34;") // shorter than "&quot;"
+	htmlApos = []byte("&#39;") // shorter than "&apos;" and apos was not in HTML until HTML5
+	htmlAmp  = []byte("&amp;")
+	htmlLt   = []byte("&lt;")
+	htmlGt   = []byte("&gt;")
+	htmlNull = []byte("\uFFFD")
+)
+
+// HTMLEscape writes to w the escaped HTML equivalent of the plain text data b.
+func HTMLEscape(w io.Writer, b []byte) {
+	last := 0
+	for i, c := range b {
+		var html []byte
+		switch c {
+		case '\000':
+			html = htmlNull
+		case '"':
+			html = htmlQuot
+		case '\'':
+			html = htmlApos
+		case '&':
+			html = htmlAmp
+		case '<':
+			html = htmlLt
+		case '>':
+			html = htmlGt
+		default:
+			continue
+		}
+		w.Write(b[last:i])
+		w.Write(html)
+		last = i + 1
+	}
+	w.Write(b[last:])
+}
+
+// HTMLEscapeString returns the escaped HTML equivalent of the plain text data s.
+func HTMLEscapeString(s string) string {
+	// Avoid allocation if we can.
+	if !strings.ContainsAny(s, "'\"&<>\000") {
+		return s
+	}
+	var b strings.Builder
+	HTMLEscape(&b, []byte(s))
+	return b.String()
+}
+
+// HTMLEscaper returns the escaped HTML equivalent of the textual
+// representation of its arguments.
+func HTMLEscaper(args ...any) string {
+	return HTMLEscapeString(evalArgs(args))
+}
+
+// JavaScript escaping.
+
+var (
+	jsLowUni = []byte(`\u00`)
+	hex      = []byte("0123456789ABCDEF")
+
+	jsBackslash = []byte(`\\`)
+	jsApos      = []byte(`\'`)
+	jsQuot      = []byte(`\"`)
+	jsLt        = []byte(`\u003C`)
+	jsGt        = []byte(`\u003E`)
+	jsAmp       = []byte(`\u0026`)
+	jsEq        = []byte(`\u003D`)
+)
+
+// JSEscape writes to w the escaped JavaScript equivalent of the plain text data b.
+func JSEscape(w io.Writer, b []byte) {
+	last := 0
+	for i := 0; i < len(b); i++ {
+		c := b[i]
+
+		if !jsIsSpecial(rune(c)) {
+			// fast path: nothing to do
+			continue
+		}
+		w.Write(b[last:i])
+
+		if c < utf8.RuneSelf {
+			// Quotes, slashes and angle brackets get quoted.
+			// Control characters get written as \u00XX.
+			switch c {
+			case '\\':
+				w.Write(jsBackslash)
+			case '\'':
+				w.Write(jsApos)
+			case '"':
+				w.Write(jsQuot)
+			case '<':
+				w.Write(jsLt)
+			case '>':
+				w.Write(jsGt)
+			case '&':
+				w.Write(jsAmp)
+			case '=':
+				w.Write(jsEq)
+			default:
+				w.Write(jsLowUni)
+				t, b := c>>4, c&0x0f
+				w.Write(hex[t : t+1])
+				w.Write(hex[b : b+1])
+			}
+		} else {
+			// Unicode rune.
+			r, size := utf8.DecodeRune(b[i:])
+			if unicode.IsPrint(r) {
+				w.Write(b[i : i+size])
+			} else {
+				fmt.Fprintf(w, "\\u%04X", r)
+			}
+			i += size - 1
+		}
+		last = i + 1
+	}
+	w.Write(b[last:])
+}
+
+// JSEscapeString returns the escaped JavaScript equivalent of the plain text data s.
+func JSEscapeString(s string) string {
+	// Avoid allocation if we can.
+	if strings.IndexFunc(s, jsIsSpecial) < 0 {
+		return s
+	}
+	var b strings.Builder
+	JSEscape(&b, []byte(s))
+	return b.String()
+}
+
+func jsIsSpecial(r rune) bool {
+	switch r {
+	case '\\', '\'', '"', '<', '>', '&', '=':
+		return true
+	}
+	return r < ' ' || utf8.RuneSelf <= r
+}
+
+// JSEscaper returns the escaped JavaScript equivalent of the textual
+// representation of its arguments.
+func JSEscaper(args ...any) string {
+	return JSEscapeString(evalArgs(args))
+}
+
+// URLQueryEscaper returns the escaped value of the textual representation of
+// its arguments in a form suitable for embedding in a URL query.
+func URLQueryEscaper(args ...any) string {
+	return url.QueryEscape(evalArgs(args))
+}
+
+// evalArgs formats the list of arguments into a string. It is therefore equivalent to
+//
+//	fmt.Sprint(args...)
+//
+// except that each argument is indirected (if a pointer), as required,
+// using the same rules as the default string evaluation during template
+// execution.
+func evalArgs(args []any) string {
+	ok := false
+	var s string
+	// Fast path for simple common case.
+	if len(args) == 1 {
+		s, ok = args[0].(string)
+	}
+	if !ok {
+		for i, arg := range args {
+			a, ok := printableValue(reflect.ValueOf(arg))
+			if ok {
+				args[i] = a
+			} // else let fmt do its thing
+		}
+		s = fmt.Sprint(args...)
+	}
+	return s
+}

internal/methodlesstemplate/internal/fmtsort/sort.go

@@ -0,0 +1,154 @@
+// Copyright 2018 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package fmtsort provides a general stable ordering mechanism
+// for maps, on behalf of the fmt and text/template packages.
+// It is not guaranteed to be efficient and works only for types
+// that are valid map keys.
+package fmtsort
+
+import (
+	"cmp"
+	"reflect"
+	"slices"
+)
+
+// Note: Throughout this package we avoid calling reflect.Value.Interface as
+// it is not always legal to do so and it's easier to avoid the issue than to face it.
+
+// SortedMap is a slice of KeyValue pairs that simplifies sorting
+// and iterating over map entries.
+//
+// Each KeyValue pair contains a map key and its corresponding value.
+type SortedMap []KeyValue
+
+// KeyValue holds a single key and value pair found in a map.
+type KeyValue struct {
+	Key, Value reflect.Value
+}
+
+// Sort accepts a map and returns a SortedMap that has the same keys and
+// values but in a stable sorted order according to the keys, modulo issues
+// raised by unorderable key values such as NaNs.
+//
+// The ordering rules are more general than with Go's < operator:
+//
+//   - when applicable, nil compares low
+//   - ints, floats, and strings order by <
+//   - NaN compares less than non-NaN floats
+//   - bool compares false before true
+//   - complex compares real, then imag
+//   - pointers compare by machine address
+//   - channel values compare by machine address
+//   - structs compare each field in turn
+//   - arrays compare each element in turn.
+//     Otherwise identical arrays compare by length.
+//   - interface values compare first by reflect.Type describing the concrete type
+//     and then by concrete value as described in the previous rules.
+func Sort(mapValue reflect.Value) SortedMap {
+	if mapValue.Type().Kind() != reflect.Map {
+		return nil
+	}
+	// Note: this code is arranged to not panic even in the presence
+	// of a concurrent map update. The runtime is responsible for
+	// yelling loudly if that happens. See issue 33275.
+	n := mapValue.Len()
+	sorted := make(SortedMap, 0, n)
+	iter := mapValue.MapRange()
+	for iter.Next() {
+		sorted = append(sorted, KeyValue{iter.Key(), iter.Value()})
+	}
+	slices.SortStableFunc(sorted, func(a, b KeyValue) int {
+		return compare(a.Key, b.Key)
+	})
+	return sorted
+}
+
+// compare compares two values of the same type. It returns -1, 0, 1
+// according to whether a > b (1), a == b (0), or a < b (-1).
+// If the types differ, it returns -1.
+// See the comment on Sort for the comparison rules.
+func compare(aVal, bVal reflect.Value) int {
+	aType, bType := aVal.Type(), bVal.Type()
+	if aType != bType {
+		return -1 // No good answer possible, but don't return 0: they're not equal.
+	}
+	switch aVal.Kind() {
+	case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
+		return cmp.Compare(aVal.Int(), bVal.Int())
+	case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
+		return cmp.Compare(aVal.Uint(), bVal.Uint())
+	case reflect.String:
+		return cmp.Compare(aVal.String(), bVal.String())
+	case reflect.Float32, reflect.Float64:
+		return cmp.Compare(aVal.Float(), bVal.Float())
+	case reflect.Complex64, reflect.Complex128:
+		a, b := aVal.Complex(), bVal.Complex()
+		if c := cmp.Compare(real(a), real(b)); c != 0 {
+			return c
+		}
+		return cmp.Compare(imag(a), imag(b))
+	case reflect.Bool:
+		a, b := aVal.Bool(), bVal.Bool()
+		switch {
+		case a == b:
+			return 0
+		case a:
+			return 1
+		default:
+			return -1
+		}
+	case reflect.Pointer, reflect.UnsafePointer:
+		return cmp.Compare(aVal.Pointer(), bVal.Pointer())
+	case reflect.Chan:
+		if c, ok := nilCompare(aVal, bVal); ok {
+			return c
+		}
+		return cmp.Compare(aVal.Pointer(), bVal.Pointer())
+	case reflect.Struct:
+		for i := 0; i < aVal.NumField(); i++ {
+			if c := compare(aVal.Field(i), bVal.Field(i)); c != 0 {
+				return c
+			}
+		}
+		return 0
+	case reflect.Array:
+		for i := 0; i < aVal.Len(); i++ {
+			if c := compare(aVal.Index(i), bVal.Index(i)); c != 0 {
+				return c
+			}
+		}
+		return 0
+	case reflect.Interface:
+		if c, ok := nilCompare(aVal, bVal); ok {
+			return c
+		}
+		c := compare(reflect.ValueOf(aVal.Elem().Type()), reflect.ValueOf(bVal.Elem().Type()))
+		if c != 0 {
+			return c
+		}
+		return compare(aVal.Elem(), bVal.Elem())
+	default:
+		// Certain types cannot appear as keys (maps, funcs, slices), but be explicit.
+		panic("bad type in compare: " + aType.String())
+	}
+}
+
+// nilCompare checks whether either value is nil. If not, the boolean is false.
+// If either value is nil, the boolean is true and the integer is the comparison
+// value. The comparison is defined to be 0 if both are nil, otherwise the one
+// nil value compares low. Both arguments must represent a chan, func,
+// interface, map, pointer, or slice.
+func nilCompare(aVal, bVal reflect.Value) (int, bool) {
+	if aVal.IsNil() {
+		if bVal.IsNil() {
+			return 0, true
+		}
+		return -1, true
+	}
+	if bVal.IsNil() {
+		return 1, true
+	}
+	return 0, false
+}

internal/methodlesstemplate/option.go

@@ -0,0 +1,72 @@
+// Copyright 2015 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// This file contains the code to handle template options.
+
+package template
+
+import "strings"
+
+// missingKeyAction defines how to respond to indexing a map with a key that is not present.
+type missingKeyAction int
+
+const (
+	mapInvalid   missingKeyAction = iota // Return an invalid reflect.Value.
+	mapZeroValue                         // Return the zero value for the map element.
+	mapError                             // Error out
+)
+
+type option struct {
+	missingKey missingKeyAction
+}
+
+// Option sets options for the template. Options are described by
+// strings, either a simple string or "key=value". There can be at
+// most one equals sign in an option string. If the option string
+// is unrecognized or otherwise invalid, Option panics.
+//
+// Known options:
+//
+// missingkey: Control the behavior during execution if a map is
+// indexed with a key that is not present in the map.
+//
+//	"missingkey=default" or "missingkey=invalid"
+//		The default behavior: Do nothing and continue execution.
+//		If printed, the result of the index operation is the string
+//		"<no value>".
+//	"missingkey=zero"
+//		The operation returns the zero value for the map type's element.
+//	"missingkey=error"
+//		Execution stops immediately with an error.
+func (t *Template) Option(opt ...string) *Template {
+	t.init()
+	for _, s := range opt {
+		t.setOption(s)
+	}
+	return t
+}
+
+func (t *Template) setOption(opt string) {
+	if opt == "" {
+		panic("empty option string")
+	}
+	// key=value
+	if key, value, ok := strings.Cut(opt, "="); ok {
+		switch key {
+		case "missingkey":
+			switch value {
+			case "invalid", "default":
+				t.option.missingKey = mapInvalid
+				return
+			case "zero":
+				t.option.missingKey = mapZeroValue
+				return
+			case "error":
+				t.option.missingKey = mapError
+				return
+			}
+		}
+	}
+	panic("unrecognized option: " + opt)
+}

internal/methodlesstemplate/template.go

@@ -0,0 +1,236 @@
+// Copyright 2011 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package template
+
+import (
+	"maps"
+	"reflect"
+	"sync"
+	"text/template/parse"
+)
+
+// common holds the information shared by related templates.
+type common struct {
+	tmpl   map[string]*Template // Map from name to defined templates.
+	muTmpl sync.RWMutex         // protects tmpl
+	option option
+	// We use two maps, one for parsing and one for execution.
+	// This separation makes the API cleaner since it doesn't
+	// expose reflection to the client.
+	muFuncs    sync.RWMutex // protects parseFuncs and execFuncs
+	parseFuncs FuncMap
+	execFuncs  map[string]reflect.Value
+}
+
+// Template is the representation of a parsed template. The *parse.Tree
+// field is exported only for use by [html/template] and should be treated
+// as unexported by all other clients.
+type Template struct {
+	name string
+	*parse.Tree
+	*common
+	leftDelim  string
+	rightDelim string
+}
+
+// New allocates a new, undefined template with the given name.
+func New(name string) *Template {
+	t := &Template{
+		name: name,
+	}
+	t.init()
+	return t
+}
+
+// Name returns the name of the template.
+func (t *Template) Name() string {
+	return t.name
+}
+
+// New allocates a new, undefined template associated with the given one and with the same
+// delimiters. The association, which is transitive, allows one template to
+// invoke another with a {{template}} action.
+//
+// Because associated templates share underlying data, template construction
+// cannot be done safely in parallel. Once the templates are constructed, they
+// can be executed in parallel.
+func (t *Template) New(name string) *Template {
+	t.init()
+	nt := &Template{
+		name:       name,
+		common:     t.common,
+		leftDelim:  t.leftDelim,
+		rightDelim: t.rightDelim,
+	}
+	return nt
+}
+
+// init guarantees that t has a valid common structure.
+func (t *Template) init() {
+	if t.common == nil {
+		c := new(common)
+		c.tmpl = make(map[string]*Template)
+		c.parseFuncs = make(FuncMap)
+		c.execFuncs = make(map[string]reflect.Value)
+		t.common = c
+	}
+}
+
+// Clone returns a duplicate of the template, including all associated
+// templates. The actual representation is not copied, but the name space of
+// associated templates is, so further calls to [Template.Parse] in the copy will add
+// templates to the copy but not to the original. Clone can be used to prepare
+// common templates and use them with variant definitions for other templates
+// by adding the variants after the clone is made.
+func (t *Template) Clone() (*Template, error) {
+	nt := t.copy(nil)
+	nt.init()
+	if t.common == nil {
+		return nt, nil
+	}
+	nt.option = t.option
+	t.muTmpl.RLock()
+	defer t.muTmpl.RUnlock()
+	for k, v := range t.tmpl {
+		if k == t.name {
+			nt.tmpl[t.name] = nt
+			continue
+		}
+		// The associated templates share nt's common structure.
+		tmpl := v.copy(nt.common)
+		nt.tmpl[k] = tmpl
+	}
+	t.muFuncs.RLock()
+	defer t.muFuncs.RUnlock()
+	maps.Copy(nt.parseFuncs, t.parseFuncs)
+	maps.Copy(nt.execFuncs, t.execFuncs)
+	return nt, nil
+}
+
+// copy returns a shallow copy of t, with common set to the argument.
+func (t *Template) copy(c *common) *Template {
+	return &Template{
+		name:       t.name,
+		Tree:       t.Tree,
+		common:     c,
+		leftDelim:  t.leftDelim,
+		rightDelim: t.rightDelim,
+	}
+}
+
+// AddParseTree associates the argument parse tree with the template t, giving
+// it the specified name. If the template has not been defined, this tree becomes
+// its definition. If it has been defined and already has that name, the existing
+// definition is replaced; otherwise a new template is created, defined, and returned.
+func (t *Template) AddParseTree(name string, tree *parse.Tree) (*Template, error) {
+	t.init()
+	t.muTmpl.Lock()
+	defer t.muTmpl.Unlock()
+	nt := t
+	if name != t.name {
+		nt = t.New(name)
+	}
+	// Even if nt == t, we need to install it in the common.tmpl map.
+	if t.associate(nt, tree) || nt.Tree == nil {
+		nt.Tree = tree
+	}
+	return nt, nil
+}
+
+// Templates returns a slice of defined templates associated with t.
+func (t *Template) Templates() []*Template {
+	if t.common == nil {
+		return nil
+	}
+	// Return a slice so we don't expose the map.
+	t.muTmpl.RLock()
+	defer t.muTmpl.RUnlock()
+	m := make([]*Template, 0, len(t.tmpl))
+	for _, v := range t.tmpl {
+		m = append(m, v)
+	}
+	return m
+}
+
+// Delims sets the action delimiters to the specified strings, to be used in
+// subsequent calls to [Template.Parse], [Template.ParseFiles], or [Template.ParseGlob]. Nested template
+// definitions will inherit the settings. An empty delimiter stands for the
+// corresponding default: {{ or }}.
+// The return value is the template, so calls can be chained.
+func (t *Template) Delims(left, right string) *Template {
+	t.init()
+	t.leftDelim = left
+	t.rightDelim = right
+	return t
+}
+
+// Funcs adds the elements of the argument map to the template's function map.
+// It must be called before the template is parsed.
+// It panics if a value in the map is not a function with appropriate return
+// type or if the name cannot be used syntactically as a function in a template.
+// It is legal to overwrite elements of the map. The return value is the template,
+// so calls can be chained.
+func (t *Template) Funcs(funcMap FuncMap) *Template {
+	t.init()
+	t.muFuncs.Lock()
+	defer t.muFuncs.Unlock()
+	addValueFuncs(t.execFuncs, funcMap)
+	addFuncs(t.parseFuncs, funcMap)
+	return t
+}
+
+// Lookup returns the template with the given name that is associated with t.
+// It returns nil if there is no such template or the template has no definition.
+func (t *Template) Lookup(name string) *Template {
+	if t.common == nil {
+		return nil
+	}
+	t.muTmpl.RLock()
+	defer t.muTmpl.RUnlock()
+	return t.tmpl[name]
+}
+
+// Parse parses text as a template body for t.
+// Named template definitions ({{define ...}} or {{block ...}} statements) in text
+// define additional templates associated with t and are removed from the
+// definition of t itself.
+//
+// Templates can be redefined in successive calls to Parse.
+// A template definition with a body containing only white space and comments
+// is considered empty and will not replace an existing template's body.
+// This allows using Parse to add new named template definitions without
+// overwriting the main template body.
+func (t *Template) Parse(text string) (*Template, error) {
+	t.init()
+	t.muFuncs.RLock()
+	trees, err := parse.Parse(t.name, text, t.leftDelim, t.rightDelim, t.parseFuncs, builtins())
+	t.muFuncs.RUnlock()
+	if err != nil {
+		return nil, err
+	}
+	// Add the newly parsed trees, including the one for t, into our common structure.
+	for name, tree := range trees {
+		if _, err := t.AddParseTree(name, tree); err != nil {
+			return nil, err
+		}
+	}
+	return t, nil
+}
+
+// associate installs the new template into the group of templates associated
+// with t. The two are already known to share the common structure.
+// The boolean return value reports whether to store this tree as t.Tree.
+func (t *Template) associate(new *Template, tree *parse.Tree) bool {
+	if new.common != t.common {
+		panic("internal error: associate not common")
+	}
+	if old := t.tmpl[new.name]; old != nil && parse.IsEmptyTree(tree.Root) && old.Tree != nil {
+		// If a template by that name exists,
+		// don't replace it with an empty template.
+		return false
+	}
+	t.tmpl[new.name] = new
+	return true
+}

v1/topdown/template.go

@@ -3,7 +3,14 @@ package topdown
 import (
 	"bytes"
 	"strings"
-	"text/template"
+
+	// A method-less copy of text/template (see internal/methodlesstemplate). Rego values
+	// decode to map[string]any/[]any/scalars, which have no methods, so eliding
+	// method calls is a no-op here; it keeps text/template's evalField
+	// MethodByName off the reachable graph, which otherwise disables the Go
+	// linker's method-level dead-code elimination binary-wide (golang/go#72895,
+	// #7903).
+	template "github.com/open-policy-agent/opa/internal/methodlesstemplate"
 
 	"github.com/open-policy-agent/opa/v1/ast"
 	"github.com/open-policy-agent/opa/v1/topdown/builtins"

v1/topdown/template_dce_test.go

@@ -0,0 +1,45 @@
+package topdown
+
+import (
+	"go/parser"
+	"go/token"
+	"os"
+	"strings"
+	"testing"
+)
+
+// TestNoStdlibTextTemplateImport guards the linker dead-code-elimination fix
+// (#7903): render_template must use the method-less template package, never
+// stdlib text/template or html/template. Their evalField reaches
+// reflect.Value.MethodByName, whose reachability disables method-level DCE for
+// the whole binary of every OPA embedder. Reintroducing the import silently
+// regresses that. Scans every non-test file in the package so the guard holds
+// even if the import moves to another file or template.go is renamed.
+func TestNoStdlibTextTemplateImport(t *testing.T) {
+	entries, err := os.ReadDir(".")
+	if err != nil {
+		t.Fatal(err)
+	}
+	fset := token.NewFileSet()
+	scanned := 0
+	for _, e := range entries {
+		name := e.Name()
+		if e.IsDir() || !strings.HasSuffix(name, ".go") || strings.HasSuffix(name, "_test.go") {
+			continue
+		}
+		f, err := parser.ParseFile(fset, name, nil, parser.ImportsOnly)
+		if err != nil {
+			t.Fatalf("parse %s: %v", name, err)
+		}
+		scanned++
+		for _, imp := range f.Imports {
+			switch strings.Trim(imp.Path.Value, `"`) {
+			case "text/template", "html/template":
+				t.Errorf("%s imports %s — reintroduces the linker DCE-defeat trigger", name, imp.Path.Value)
+			}
+		}
+	}
+	if scanned == 0 {
+		t.Fatal("scanned no package files; test ran from the wrong directory")
+	}
+}