Keep street address matches on one line

RELEASE_NOTES.md

@@ -4,6 +4,7 @@ Notable changes to the `phileas-redact` package, most recent first.
 
 ## Unreleased
 
+* The `streetAddress` filter no longer matches across a line break. The parts of an address were joined by `\s`, which includes line breaks, so a number at the end of one line and a word and a street type on the next were read as an address: in `Cedar Falls, IA 50613`, a blank line, and `Dear Dr. Rivera,`, the text from `50613` to `Dr` was detected, with `Dr` read as Drive. The parts are now joined by spaces and tabs, as in Phileas (Java), so an address on one line and the street line of a multi-line address block are found as before. Parts separated by other whitespace, such as a no-break space, no longer form an address. This is the parity port of the Phileas (Java) fix for philterd/phileas#474.
 * The `phoneNumber` filter now finds a number written with a bare `1` country code and an area code beginning with 1, such as `1-112-997-5633`, `1 112 997 5633`, or `1-109-299-5128 x512`, when a configured `region` uses country code 1 (such as `US`, the default). libphonenumber read the `1` as the country code and the area code's `1` as a national prefix, leaving too few digits, so these were missed while `+1 112-997-5633` was found. A number found before keeps its span and confidence, and a number with no separators, such as `11129975633`, is still not found. Same as the Java library (philterd/phileas#458).
 * The `RELATIVE` strategy now expresses a detected date relative to the current date (`3 months ago`, `2 years 1 months ago`), matching Phileas (Java). Previously it returned the date unchanged, leaving it in the document. A future date is written as `in 3 months` when the strategy sets `futureDates` to `true` and is redacted otherwise.
 * A strategy no longer leaves a detected value in place when it cannot be applied. An unrecognized date under `SHIFT`, `TRUNCATE_TO_YEAR`, or `RELATIVE`, a value of 4 characters or fewer under `LAST_4` or `TRUNCATE`, and a filter type with no random generator under `RANDOM_REPLACE` are now redacted with the strategy's `redactionFormat`. A date-shaped value that is not a real date, such as `02/31/2020`, is treated as unrecognized instead of raising an error.

phileas/filters/street_address_filter.py

@@ -21,9 +21,12 @@
 from .base import BaseFilter, FilterType
 
 
+# The parts are joined by spaces and tabs only, as in the Java filter, so a match does not
+# run from one line into the next (a ZIP code at the end of a line and "Dear Dr." on the
+# next are not an address).
 _PATTERNS = [
     re.compile(
-        r"\b\d{1,5}\s+(?:[A-Za-z0-9#.,\-]+\s+){1,4}"
+        r"\b\d{1,5}[ \t]+(?:[A-Za-z0-9#.,\-]+[ \t]+){1,4}"
         r"(?:Street|St|Avenue|Ave|Boulevard|Blvd|Road|Rd|Lane|Ln|Drive|Dr|Court|Ct|Place|Pl|Way|Circle|Cir|Terrace|Ter|Trail|Trl)\b",
         re.IGNORECASE,
     ),

tests/test_street_address_detection.py

@@ -116,6 +116,26 @@ def test_long_street_name_within_word_limit(street_filter):
     assert spans[0].text == addr
 
 
+def test_no_match_across_line_breaks(street_filter):
+    # A ZIP code at the end of a line and "Dear Dr." on the next line are not an address.
+    text = "Address: 214 Juniper Court, Cedar Falls, IA 50613\n\nDear Dr. Rivera,"
+    spans = street_filter.detect(text)
+    assert [s.text for s in spans] == ["214 Juniper Court"]
+
+
+@pytest.mark.parametrize(
+    "text,expected",
+    [
+        ("214 Juniper Court\nCedar Falls, IA 50613", "214 Juniper Court"),
+        ("123 Main St Apt 4B\r\nSpringfield, IL 62704", "123 Main St"),
+        ("Jane Doe\n1 Martin Luther King Jr Blvd\nAtlanta, GA 30303", "1 Martin Luther King Jr Blvd"),
+    ],
+)
+def test_street_line_of_address_block(street_filter, text, expected):
+    spans = street_filter.detect(text)
+    assert [s.text for s in spans] == [expected]
+
+
 def test_default_context_returns_results(street_filter):
     spans = street_filter.detect("123 Main Street", context="default")
     assert len(spans) == 1
@@ -146,6 +166,9 @@ def test_custom_context_still_detects(street_filter):
         "123 Main Streets",  # trailing 's' breaks the suffix word boundary
         "123 Main Streetss",
         "123456 Main Street",  # number has six digits (exceeds \\d{1,5})
+        "Cedar Falls, IA 50613\nDear Dr. Rivera,",  # parts on different lines
+        "Cedar Falls, IA 50613\r\nDear Dr. Rivera,",
+        "Unit 12\nMain Street",
     ],
 )
 def test_no_detection(street_filter, text):