Testfx Localization Validation

TestFx localization validation

Run from the repository root with Python 3.12 or newer; no third-party packages or .NET build are needed for the guard:

python .github\scripts\test_check_localization.py
python .github\scripts\check_localization.py

On Linux/macOS use / in command paths. The standalone check-localization.yml workflow runs the same commands on localization PRs. The guard is read-only. Exit codes are 0 for clean, 1 for catalog findings, and 2 for invocation/discovery errors (including no catalogs).

Runtime choice

Python is a measured exception to the preference for PowerShell when there is no performance, feature or security impact. On Windows with Python 3.12.10 and PowerShell 7.6.6, nine interleaved warm-cache runs (including process startup) gave these medians:

The dependency-free probe used .NET XML readers, cached neutral resources and ordinal IDs. It implemented only a subset of validation, not feature parity; it already exceeded the complete guard's runtime. Python is retained rather than accepting that regression or weakening the checks. These measurements are not a claim about every possible PowerShell implementation or other operating systems. A future port must demonstrate no performance regression and pass the committed regression suite, including case-sensitive IDs/tokens and literal/escaped placeholder semantics. Explicit .NET XML-reader security settings and DTD handling were independently exercised in the feasibility probe, not by the Python regression suite. A port must separately verify those settings and compare accepted XML-input profiles; passing the suite alone does not prove XML security parity.

Rules and limits

• TestFx uses XLIFF 1.2 with the OASIS namespace and a relative file original pointing to the neutral RESX. Resolve it relative to the catalog, not the project or repository. Do not confuse locale cs (Czech) with a C# file.
• The XLIFF 1.2 transitional schema requires one direct source and permits zero or one direct target per trans-unit. Do not require a target to exist. Targets inside alt-trans are alternatives, not duplicates. Groups are legal. Unit IDs are unique within a file across trans-unit and bin-unit, not globally across files.
• Generated units must correspond to neutral string resource IDs, and their source text must match the neutral value exactly. When the neutral comment is nonempty, a note must match it exactly. Match XliffTasks' RESX exclusions: typed/binary entries, designer names (>>..., *.LayoutSettings), whole-value {Locked} comments, and empty/whitespace values do not require units. Additional notes are legal. An absent target or empty unfinished target is allowed; unfinished translations are not evidence of a structural defect.
• Numeric composite-format argument indices must be preserved; reordering, repetition, alignment and format specifier changes are not automatically translation defects. Escaped braces are supported. Named templates such as {tfm} and literal JSON are not assumed to be .NET composite formats. When source text is not unambiguously a composite format, manually check any embedded numeric placeholders. This is not a complete String.Format parser.
• Every {Locked="..."} token must occur verbatim in the neutral value and in a nonempty/completed target. Matching is case-sensitive and substring-based. The guard checks presence, not grammar or a word's translatability. Manually check for collisions with other words in the same value: const also matches constant. Prefer the longest unambiguous invariant token. Lock punctuation only when invariant or necessary to disambiguate.
• This is a focused guard, not a full XSD validator or translation-quality review. Inline XLIFF codes require producer-specific rendering and are reported for manual handling instead of silently flattened. Do not invent new bans on otherwise legal XLIFF solely to make the checker pass.