NEWS

polyglotr 1.7.5.9000

google_get_supported_languages() now returns the bundled google_supported_languages snapshot (columns Language and ISO-639 code) instead of scraping the Google Cloud documentation. The scrape took the first table on the page, which is no longer the language list, so it returned a different set of languages with different columns. Getter and google_is_valid_language_code() now agree by construction.

translate_file(overwrite = TRUE) no longer fails with "can only write character objects".
translate_file() no longer silently replaces lines it cannot translate with an empty line; the original line is kept and a warning reports how many lines were skipped.
translate_file() no longer doubles the indentation of indented lines, and no longer sends blank lines to the translation service. A roxygen line that fails to translate keeps its text instead of becoming a bare #', and whitespace after #' (e.g. indented @examples code) is preserved.
batch_translate() now does what its documentation says: it calls translate_file() once per target language and writes one file each. It is now exported. All language codes are validated before any file is written.
translate_file() no longer produces a trailing dot (README_de_translated.) for input files without an extension.

translate_file() and batch_translate() return the path(s) they wrote, invisibly, instead of NULL.
translate_file() now warns when lines could not be translated (code run with options(warn = 2) will stop there) and checks language codes before reading the file. batch_translate() stops if the input file does not exist.
Internal: chunking of long texts moved out of google_translate() into split_translation_chunks(); create_translation_table() and create_transliteration_table() share one implementation. No behaviour change.
Removed RCurl, rvest and rlang from Imports; they are no longer used. dplyr stays: purrr::map_dfr() in apertium_get_language_pairs() needs it.
Documented the source and staleness of the google_supported_languages dataset.
Offline tests (mocked with testthat::local_mocked_bindings()) for the functions above.

Fixed google_translate() after Google's mobile HTML endpoint began redirecting automated requests to an HTTP 429 bot-detection page. Translation now uses the JSON endpoint and dict-chrome-ex client identifier (issue #32).
Updated language_detect() to use the same working client identifier and to parse the detector's structured response instead of scraping the raw array.

Fixed 301-redirect URL for QCRI in README.md: replaced qcri.org with hbku.edu.qa/en/qcri.
Removed linguee_external_sources(), linguee_translation_examples(), and linguee_word_translation() — the upstream API (linguee-api.fly.dev) is no longer available.

All functions that use internet resources now fail gracefully with an informative message when the service is unavailable (CRAN policy compliance). Network-level errors (DNS failure, connection refused, timeout) are caught via tryCatch() and reported via message() rather than propagating as errors.
wikipedia_get_language_names() example wrapped in \donttest{}.