Legacy source projections

legacy_pairs.ts_pairs(root) and legacy_pairs.po_pairs(po, fuzzy=False) provide the explicit selection rules used by existing TMX extraction commands. They return ordered (source, target, context, plural) tuples. They leave duplicate rows in place so the caller can apply its existing deduplication policy.

Qt input is a parsed ElementTree TS root. The projection visits messages within contexts, strips the context name, and excludes missing/empty sources and targets marked unfinished, obsolete or vanished. Plurals use the first and last nonempty numerusform; a single form produces one singular row. Both use the same source.

Gettext input is a parsed polib.POFile. Obsolete, empty-source and untranslated entries are excluded. Fuzzy entries require fuzzy=True. Indexed plural forms are sorted; the first emits the singular source and the last emits msgid_plural. Empty end forms are omitted independently. Context and text retain their whitespace.

These are compatibility projections, not complete catalog conversions. Middle plural forms and other native metadata do not appear in the resulting tuples; Qt byte escapes and length variants are not interpreted by this historical rule. Use formats.ts, formats.po and conversion.convert() for complete native catalog handling and explicit loss reporting. Locale normalization, resource discovery, provenance and destination paths remain caller policy. Feed selected records to memory.tmx_write.write_records() for atomic XML serialization.

The input contracts were checked against the official Qt TS schema and polib API. Attribution for the adapted compatibility rules is retained in NOTICE.

Apple resource values

Install vexy-localizzy[sources] for apple_resources.parse_strings_text(bytes). It returns ordered key/value pairs, retaining duplicate keys, whitespace and comment delimiters inside quoted values. UTF-8 (with optional BOM) and BOM-marked UTF-16 are supported. Invalid syntax/encoding raises instead of returning partial or replacement-character text. Comments outside strings are omitted.

resource_items(bytes) additionally reads binary/XML property lists, retaining nested native values and dictionary order. flatten_value(key, value, device="mac") projects nonempty strings, format strings and all plural categories into ordered pairs with key|variable|category identifiers. Device dictionaries prefer the requested device and then other. This is an explicit extraction projection; keep original bytes for native editing and full device coverage.

These APIs do not discover installed applications, normalize language folders, choose source-language fallbacks or assign provenance. The caller supplies those policies and retains the original resource path. The text parser is the published OpenStep parser; XML/binary decoding uses plistlib.

Fluent resources

With the sources extra, fluent_resources.parse_ftl(text) yields ordered (key, text) pairs for messages, terms (prefixed -), and .attribute values. Select-expression branches append [key], except default branches, which have no suffix. Nested suffixes and multiple selectors preserve historical expansion order. Whitespace is collapsed, while variable/function/reference expressions remain serialized Fluent syntax. Repeated IDs remain available to the caller's duplicate policy; comments are omitted. Keep the original resource for editing.

Malformed entries raise ValueError before the first resource pair is emitted. flatten_pattern(pattern, max_variants=10000) and parse_ftl accept a positive integer per-pattern limit. Expansion exceeding it raises before that Cartesian product is allocated; no variants from the failing pattern are yielded. Earlier valid patterns may already have been yielded, so callers must finish iteration before publishing output. Directory discovery, source/target joining and locale fallback remain caller policy.

Properties resources

With the sources extra, properties_resources.parse_properties(text) returns all ordered key/value pairs, including duplicate keys and empty values. The caller decodes bytes and chooses a duplicate policy. The published Java-style parser handles escaped separators, whitespace, Unicode surrogate pairs and odd/even backslash continuation rules. It preserves trailing value whitespace and rejects malformed Unicode escapes before returning any pairs. This API implements Java properties syntax; Mozilla-specific properties/DTD extraction uses the API below.

Mozilla properties and DTD

mozilla_resources.parse_mozilla(text, kind) accepts properties or dtd and returns ordered key/value pairs from published moz.l10n. It keeps duplicate and empty values for caller policy. Properties escapes and continuations are decoded; valid UTF-16 surrogate pairs become Unicode characters, while isolated surrogates raise. Unknown escapes follow Mozilla syntax, which differs from strict Java properties (for example, \uNOPE becomes uNOPE).

DTD reference spelling stays literal (&, numeric references and references to other entries). Commented-out declarations are omitted. DTD include declarations are not translation entries and are never fetched; retain the original source files to preserve their metadata. Unsupported/unparsed content raises before any pairs return. This is a localization DTD projection, not a general DTD processor: moz.l10n 0.14.2 also rejects comment delimiters inside quoted entity values. Resource discovery, locale aliases and source pairing remain caller policy.