Move the translation marker onto Translation

[jakejackson1]

Fixes #42.

GravityPdf\Upload\__() was a function in src/Upload/i18n.php, loaded by Composer's files autoload. That entry is Composer's own — an autoloader that indexes classes and nothing else, such as a classmap built over php-scoper output, never runs it. Every error path then fatals with Call to undefined function GravityPdf\Upload\__(), on a site where uploads had only ever succeeded.

It is now Translation::__(), a static method, so it travels with the class. composer.json declares PSR-4 and nothing else.

Verification

The issue's own repro, loading every class and no files entry:

before:  FATAL: Error: Call to undefined function GravityPdf\Upload\__()
after:   rejected cleanly: File contents do not match the "ttf" extension. Must be one of: font/sfnt

Extraction is unchanged and needs no configuration, upstream or downstream. xgettext matches a keyword against the trailing identifier and ignores the class prefix, so -k__:1 reads Translation::__(), \GravityPdf\Upload\Translation::__() and an aliased T::__() alike. composer i18n:pot regenerates i18n/upload.pot byte-identical across the move — that is the proof, and the i18n workflow is what keeps it true.

One correction to the issue

composer dump-autoload --classmap-authoritative does not drop autoload.files — tested against a fixture; autoload_files.php is still emitted and autoload_real.php still requires it. The trigger is specifically an autoloader that isn't Composer's, which is what a scoped WordPress plugin tree normally has. The changelog says it that way so nobody on plain --classmap-authoritative goes looking for a bug they don't have.

The namespace fallback goes with it

An unqualified function call resolves against the global namespace when the current one has no match, so a validator forgetting use function GravityPdf\Upload\__; reached WordPress's __() and translated at the throw. A class name has no such fallback: the same mistake now fatals at the call.

That is still only reached on an error path, so I18nTest::testEveryCallerCanReachTheMarker() is retargeted rather than deleted — it reads every file under src/ for the import instead of waiting for a rejected upload. testNoGlobalMarkerCanMaskAMissingImport() is gone; there is no fallback left for a global stub to mask. A new testTheMarkerTravelsWithTheClass() refuses both a bare function and a files entry coming back.

tools/translator-readme/marker-import.php existed to prove the "note the leading backslash" advice in the Laravel and WordPress recipes, which was load-bearing only while a consumer could import our __(). The fixture and the caveat both go; global-underscore.php stays, since the documented adapters still call a global __().

Naming

Kept __ rather than mark(). PHP documents the __ prefix as reserved by convention; __ alone has never been a defined magic method, and CI proves it on 7.3 through 8.5. mark() would cost an extractor flag in our script and in every consumer extracting their own marked strings.

No deprecation shim: 4.0.0 is unreleased (latest tag is 3.1.0), so the function was never published.

Checks

phpunit (698 tests, 1,392 assertions, 1 skipped) · phpstan level 9 · lint · check-syntax · i18n:pot (no diff) · base64-docs · psr7-readme · translator-readme — all pass locally.

🤖 Generated with Claude Code