BlinkReceipt/blinkreceipt-android 2.4.1 on GitHub

Recognizer SDK (2.4.1)

Fixed: an out-of-memory crash at the end of a scan, while looking up the merchant from the receipt's raw text. The parser materializes line-number gaps as literal newlines when it assembles the combined raw text, so a corrupt line index padded that text with megabytes of consecutive newlines — ahead of, between, or after the real lines — and tokenizing it for the raw-text retailer request then failed on a single 44 MB–1 GB allocation. The combined raw text is now sanitized at its one entry point into the Java pipeline (OcrResultsTransformer): newline runs beyond 100 collapse to 100, and the result is hard-capped at 262,144 characters, which also bounds the session-long ScanResults retention of that text and its Parcel write. The raw-text retailer request mapper additionally tokenizes lines in place, without materializing a per-line array, and bounds its output at 50,000 tokens, as defense in depth for any input shape. Receipt-sized text is untouched and the wire payload is pinned byte-identical to the previous tokenization, so no integration changes are required.
Stability fixes and improvements