Repository navigation
Conversation
c-knof
force-pushed
the
feature/trunk/ck/po-files-migration
branch
from
October 28, 2026 08:01
b20a913 to
0ac48c3
Compare
… introduce Activities
The seven Activities now get `ILIAS\AccessControl\PublicInterface\RBAC` via `$use` instead of reaching into `$GLOBALS['DIC']->rbac()`, and call `system()` only in `isAllowedToPerform()`. The Setup agent uses `InstallLanguage::forSetup()`/`UpdateLanguage::forSetup()`, since Setup checks no permissions and needs no RBAC.
`InstalledLanguageDatabaseRepository`, `LanguageInstallationManager`, `AddLanguageEntry` and the `ilSetupLanguage` of the web Activities now get `ILIAS\Database\Connection` via `$pull` instead of resolving `$GLOBALS['ilDB']`. `ilSetupLanguage` takes an optional connection; Setup keeps handing in its database via `setDbHandler()`, also for the metrics, and the file-only checks need no database.
The Setup CLI asks `ilLanguagesInstalledAndUpdatedObjective` for its label before `achieve()`, so `getLabel()` only worked through a leftover `$GLOBALS['ilDB']`. `isApplicable()`, which the CLI calls first, now reads the languages from the Setup database and `getLabel()` lists them without database access.
The repository and the installation manager in the component graph now get the root from `ILIAS\Environment\Configuration\Installation\Directories` at resolve time instead of a path computed in `init()`. Callers outside the graph keep passing the path themselves.
… container `AllModernComponents` bound `$DIC[ILIAS\Language\Language::class]` to the `$use`d language, whose legacy adapter itself delegates to `$DIC->language()`. Nothing reads the key, so the constructor parameter and the binding are removed.
No component pulls `InstalledLanguageRepository` or `LanguageInstallationManager`, so they stay internal services of the component; `ilSetupLanguage` remains the entry point for callers outside it.
The Activities took the ref_id of the language folder as `int|\Closure` defaulting to 0. They now require `ILIAS\Language\LanguageFolder`, whose `LanguageFolderLegacyProxy` looks the ref_id up only when a permission is checked.
Keep only what the code does not say itself (the why, a constraint or a pitfall), mostly in one line, and drop the history of how it got there.
…vities A denied permission is now a `DomainException` with code 403 instead of a localized string, and `Failure::of()` masks every error outside the 4xx range as a generic 500, so callers other than the GUI no longer see internal messages. `InvalidInputException` carries 400, `AmbiguousLanguageTitleException` 409; `RendersActivityErrors` decides by code, which replaces the `SafeToDisplayActivityError` marker.
…the File Activities `LanguageActivity::isAllowedToPerform()` now rejects unknown and anonymous users before asking RBAC, as `FileUpload\Activities\AccessCheck` does for the File Activities. `maybePerformAs()` hands the trusted user id to `perform()` directly, which replaces the `additionalPerformParameters()` hook.
…ections like the File Activities The descriptions are built with `ILIAS\Data\Factory` instead of an own markdown helper, so the Activities no longer need the Refinery. The input of every Activity is a titled `section` instead of an untitled `group`, as in the File Activities.
…Provider` like the File Activities The language objects, language entries, the module cache, the user lookup and the settings are reached through one `RepositoryProvider` instead of optional closures with legacy defaults in every Activity. `UninstallLanguage`, `RemoveLocalLanguageChanges`, `AddLanguageEntry` and the two settings Activities take it as their only dependency besides language, RBAC and language folder; the tests mock it.
A `PDOException` carries its SQLSTATE as string code, and `'42S02'` compares as a 4xx code, so `Failure::of()` and `RendersActivityErrors` showed the database message as input error. Both now require an integer code. The test doubles no longer call the removed `markdown()` helper.
Keeps only what the code does not say itself, as in the File component: the reasons in `Language.php` shrink to the two real traps (no `final` Activities, the order of the contributions), and the tests lose their history notes.
`LanguageActivity` no longer takes `Language`: only `AddLanguageEntry` needs it, for the labels of its translation fields. The other six Activities, their `forSetup()` variants and the wiring drop the parameter.
… of the component translation classes
…ndent of their order
…tead of a path read during wiring
…y proxy instead of a global $DIC fallback
…st CSRF The table actions were read from the query by `executeCommand()` without a command, so ilCtrl never checked a token. `ilObjLanguageFolderGUI` now implements `ilCtrlSecurityInterface`, runs the table actions only through the unsafe command `executeTableAction` and dispatches an exact allowlist of commands, since ilCtrl compares commands case-sensitively but PHP calls methods case-insensitively. Takes effect after the ctrl security artifact is rebuilt (`composer dump-autoload` or `setup build`).
…ObjLanguageFolderGUI` Listing and downloading the deprecated entries failed with an error, because `ilObjLanguageFolderGUI` still included `ilLangDeprecated` from the former `Services/Language` directory.
…uageFolderGUI` `confirmRefreshSelectedObject()`, `confirmUninstallObject()`, `confirmUninstallChangesObject()` and the table action `edit` could no longer be reached since the table actions confirm in modals, and their confirmations failed for lack of language ids.
Before CLIENT_DATA_DIR is defined the client data directory was always composed with "[clients] default" of ilias.ini.php, also when CLIENT_ID was already defined (e.g. by the plugin Setup objectives or a CLI run for another client). The defined client id wins now, like it does in ilInitialisation.
…uild ilLanguage kept the client data directory resolved from ilias.ini.php for the whole process, also when none was found, and cached per module whether it is shipped. MigratedTranslations::reset() (e.g. after the build in the Setup) did not clear either, so objectives later in the same process (install creates the client data directory) kept serving without overlay. The client data directory is now resolved by MigratedTranslations, a found one is kept until reset(), a missing one is looked for again. The per-module cache in ilLanguage is replaced by an index of the language file directories, rebuilt when the directory manager in $DIC changes.
Without the component graph a module was recognised as shipped by a
glob over components/*/*/lang/*.pot, which also ran in every web
request for each plugin module, and missed a contribution whose template
is named or placed differently.
The build now records every contributed module in current.json
("shipped"); with a build that lists them only these count, the glob is
the fallback without one (or with a current.json written before). The
build warns about a contribution without a lang/<prefix>.pot, which
Setup and CLI would not recognise without a build.
Cover the editable state of a shipped module (shipped .po with or without overlay, unreadable files), that the administration shows, saves, deletes and adds entries of shipped modules without lng_data, the contributed modules recorded in current.json with the build warning for a missing template, the client data directory resolved by CLIENT_ID and forgotten on reset, and the web path with a filled module cache and the identifier path without a build. Rename the tests that still named the database fallback.
…languages
Every module is shipped in PO files now, so lang/ilias_<lang>.lang holds
only the header up to `<!-- language file start -->`. The check of all
languages in the administration took the empty content behind the
header for a missing header and reported every file as invalid ("wrong
header"). Only a missing start marker is a wrong header.
If the content of a shipped module could not be read, saving in the administration and "add new variable" still wrote the lng_data rows with local_change before giving up on lng_modules and the overlay; a later setup update took them over into the overlay unnoticed. And with an unreadable overlay the save was based on the shipped values alone, which dropped the overlay's other local changes. The content is now checked per module before anything is written: a module whose shipped .po or overlay cannot be read is skipped entirely and reported as not written. Only the display still falls back to the shipped values for an unreadable overlay.
An invalid `shipped` entry in current.json no longer invalidates the whole pointer: only the list is dropped, and shipped modules are recognised by their templates instead. A contribution with an invalid file name pattern is listed as well.
…uild is reset Resolving the client data directory reads ilias.ini.php; while the directory does not exist yet (setup install), do not read it again for every lookup but only after MigratedTranslations::reset(). Keep the memo of shipped .po files without a build in MigratedTranslations as well, so reset() clears it too.
Setup read the local changes of a shipped module from lng_data on every update and let them win over the overlay, so stale or partially written rows overrode the overlay. lng_data is only the rollback copy. The local changes in lng_data are now taken over into the overlay once, while the module has no overlay for the language (upgrade from ILIAS 10/11). Afterwards, recorded per language in <client data dir>/lang/taken-over.<lang>.json, and whenever an overlay exists, the overlay is the only source of local changes and remarks; an overlay that cannot be read is still rebuilt from lng_data, the only copy left. Uninstalling a language forgets the record. Plugin modules are unchanged, the dual-write stays. Tests that let a lng_data local change win over the overlay now expect the overlay's value; installing without local changes writes the record.
…nges once Cover that the administration and AddLanguageEntry write nothing of a module whose shipped .po or overlay cannot be read (and keep the local changes of other keys), that setup update takes over the local changes in lng_data only for a module without overlay and records it, that it leaves a module with an unreadable overlay unchanged and reports it, and the remaining cache and pointer details. Rename the tests that still named the database as source of local changes.
…ers in Setup An overlay or a `taken-over.<lang>.json` entry of a module that is no longer shipped (e.g. a reverted migration) would win over the changes made in `lng_data` once the module is migrated again, so `ilLanguagesInstalledAndUpdatedObjective` now reports such modules, as well as local changes whose take-over could not be recorded. Nothing is removed. The README explains the rollback steps.
…nguage file is read-only Merging the local changes into the shipped files ran only if `lang/ilias_<lang>.lang` was writable, although `ShippedPoMerger` checks each `.po` itself. `ilObjLanguageExtGUI` now always merges into the `.po` files, writes the `.lang` file only if it is writable and reports both results.
Close the array shape of `ShippedPoMerger::merge()`, declare the exceptions of `MigratedLanguageFileSync::overlayIfShipped()` and `markMergedIntoShipped()`, add missing value types, drop `array_values()` calls on lists and let `GrindingInputAccess::assertInternal()` assert the input type.
…rror Installing or updating a language with invalid language files threw an `InvalidInputException` (400) whose English text the administration showed untranslated. `InstallLanguage` and `UpdateLanguage` now throw `InvalidLanguageFilesException` (409), and `RendersActivityErrors` shows the new text `lng_invalid_language_files` with the language keys.
`scan_shipped_cross_module_duplicates.php` only read `lang/ilias_*.lang`, which no longer holds any module once all modules are maintained in PO files. The build reports collisions between modules itself.
Every `setup update` and `composer dump-autoload` repeated all warnings of the language build, about 340 lines. `ShippedLanguageFilesCompiledObjective` now prints them in detail only when the build is written, with collisions grouped by modules, and otherwise a single line counting them with the path of the manifest.
…tGUI` ilCtrl compares the unsafe commands case-sensitively, but PHP calls methods case-insensitively, so `cmd=maintainexecute` ran the maintenance without a token. `ilObjLanguageExtGUI` now only calls the commands it uses and shows the list for any other command.
Introduce the read-only `TranslationRepository` with `TranslationGettextRepository` as its implementation, which takes over the read path of `MigratedTranslations`. `ilLanguage` accepts a repository as optional second argument, so unit tests can replace the build and gettext with an in-memory repository.
…build `ShippedTranslationsBuild::catalogCompatibility()` left out `TranslationCatalog`, `TranslationEntry` and `PluralForms`, so a `.po` that does not compile could be served from the catalog of a build compiled by other code. The compatibility and the build id now hash the same list `CATALOG_CLASSES`.
…module Importing a language file could add an identifier that already existed in another module, whose value then replaced the one of that module. Like a new entry, `ilObjLanguageExt::importLanguageFile()` now refuses such entries: an uploaded file is rejected as a whole, the customizing file skips them, and both name the entries with the new text `lng_identifier_not_unique`.
…stomizing file `InstallLanguage` in the mode `install_local` reported invalid shipped language files only as an aborted action. It now checks the languages like a plain install and throws `InvalidLanguageFilesException`, so the administration names them.
…s in another module Installing or updating a language applied the lines of `lang/customizing/ilias_<lang>.lang.local` even if their identifier belonged to another module. `LanguageInstallationManager` now skips such lines like lines with disallowed markup, and Setup and the administration name them.
The major upgrade section of docs/configuration/install.md tells admins that the build compiles the shipped .po files, that the first setup update copies the local changes from the database into the client data directory once, and that the setup has to run as the web server user.
The build of the PO files needs ext-gettext. composer.json and composer.lock now equal the state of the dependency PRs for ext-gettext and gettext/gettext, so that this branch merges without a conflict in the content-hash once they are merged.
c-knof
force-pushed
the
feature/trunk/ck/po-files-migration
branch
from
October 28, 2026 09:34
e9842f6 to
3f65aa2
Compare
c-knof
added a commit
that referenced
this pull request
Oct 28, 2026
Requires ext-gettext and names it with the C.utf8 locale in the installation guide and the GitHub workflows. Needed by #12261.
c-knof
added a commit
that referenced
this pull request
Oct 28, 2026
Requires gettext/gettext ^5.7.3 to read, write and compile the PO files of the language modules. Needed by #12261.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Implements the accepted feature request "PO files for improving language handling"
(https://docu.ilias.de/go/wiki/wpage_8951_1357), including both requested changes (native gettext, identifier enums).
Depends on #12090 (Language on the Component Revision)
What it does
lang/<module>.potplus<module>_<lang>.poinstead of lines inlang/ilias_<lang>.lang. Setup (andcomposer dump-autoload) builds them into.mocatalogs belowartifacts/language/,ilLanguageserves them through native gettext.lng_data/lng_modulesstay a rollback copy. Local changes of earlier versions are taken over once on the first update.txt($id, $n). GeneratedLanguageIdentifierenums per component:txt(PollLanguageIdentifier::POLL_ADD).tos,poll,cmps,lng. All other modules follow in the wave PRs below.Breaking change
Language::txt(string|LanguageIdentifier $a_topic, ?int $n = null). The fallback module parameter andntxt()are removed, all 13 core callers are adapted (one commit per component: Exercise, Survey, WebServices, OrgUnit, Forum, Repository, WebDAV, UI/Refinery test doubles).Changes outside the Language component
TermsOfService, Poll, Component (module files and contribution), the callers listed above,
composer.json/composer.lock(gettext/gettext),docs/development/language.md,lang/ilias_*.lang(pilot modules removed, newlng_*texts).Requirements and operation
ext-gettext(CLI and web server) and theC.utf8locale.composer dump-autoloadorsetup buildafter every change of a shipped.po; Setup runs as the web server user.artifacts/language/<build>/manifest.json.composer dump-autoload,setup update(see README, "Rollback").Behaviour changes
-identifier-instead of the value of anotherloaded module.
docs/development/language.md)..pofiles of the components.