Select asset templates by the final extension only

[youdie006]

compileTemplatesFromDir and compileTemplatesFromAsset implement the same rule -- load every file whose extension is in Options.Extensions -- but they compute ext differently.

render.go:291, with the comment stating the case it exists for:

// Fix same-extension-dirs bug: some dir might be named to: "users.tmpl", "local.html".
// These dirs should be excluded as they are not valid golang templates, but files under
// them should be treat as normal.
...
ext = filepath.Ext(rel)

render.go:363, the asset twin, never got that fix:

ext = "." + strings.Join(strings.Split(rel, ".")[1:], ".")

For rel = "users.tmpl/index.tmpl" that produces ".tmpl/index.tmpl". No extension anyone can configure contains a path separator, so the for _, extension := range r.opt.Extensions below it is vacuously false and the template is skipped. The same happens for any multi-dot name -- a.b.tmpl gives ".b.tmpl".

Nothing fails at startup. New returns fine and the template is simply missing, so it surfaces as a render error at request time.

files := map[string]string{
	"testdata/dotdir/users.tmpl/index.tmpl": "Hello asset",
	"testdata/dotdir/a.b.tmpl":              "Hello ab",
}
dir   := New(Options{Directory: "testdata/dotdir"})
asset := New(Options{Directory: "testdata/dotdir", Asset: ..., AssetNames: ...})

dir.TemplateLookup("users.tmpl/index")    // non-nil
asset.TemplateLookup("users.tmpl/index")  // nil

The fix

Use filepath.Ext in the asset loader too, so both paths select the same files and produce the same template names. One line.

Behaviour change

Asset/go-bindata mode only, and only in the direction of loading files that were previously never loaded: anything under a dot-named directory, and any multi-dot filename. The resulting template names match what directory mode already produces (users.tmpl/index, a.b). No existing test row changes -- the suite is green unmodified. Every current fixture is single-dot, which is why this never showed up.

Verification

make ci (go test -cover -race -count=1 ./...): ok github.com/unrolled/render 1.071s, coverage: 88.4%. go vet ./... clean. golangci-lint run ./...: 0 issues, the same count as on v1, so I have not added any.

One test added to render_html_test.go driving the same two files through both loaders, plus two small fixtures. On v1 it reports:

render_html_test.go:553: asset mode did not load "users.tmpl/index"
render_html_test.go:553: asset mode did not load "a.b"

I checked the boundary from three directions rather than only confirming it goes green:

mutation	result
revert to the Split(rel, ".")[1:] form	fails
take everything after the first dot of the basename instead	fails
keep the Split form but take the last element	passes

The third is deliberate: it is semantically filepath.Ext, and its passing shows the test pins "only the final extension" rather than the literal call I happened to write.

What I did not change

• compileTemplatesFromDir -- already correct, and touching it would move directory-mode output.
• The if strings.Contains(rel, ".") guard on both sites. It is redundant now, since filepath.Ext returns "" when there is no dot, but removing it is cosmetic and widens the diff.
• Both break statements at render.go:311 and render.go:383 -- they fire on a match, which is correct.
• name := rel[0:len(rel)-len(ext)] -- correct once ext is, and it now matches the directory path exactly.

---

Disclosure: AI-assisted. I found and prepared this with an AI assistant, and I ran and verified everything above myself.