`XXXX-XX-XX` passes any date filter

[victorlin]

Scope

augur filter and augur subsample

Current Behavior

XXXX-XX-XX passes any --min-date or --max-date filter because it evaluates to [0001-01-01, present].

Expected behavior

It should behave as if the value was an empty string, i.e. dropped.

How to reproduce

Cram test:

Create metadata TSV file with two date values that should be functionally equivalent.

  $ cat >metadata.tsv <<~~
  > strain	date
  > SEQ_1	
  > SEQ_2	XXXX-XX-XX
  > ~~

BUG: SEQ_2 passes for --min-date and --max-date.

  $ ${AUGUR} filter \
  >  --metadata metadata.tsv \
  >  --min-date 2025 \
  >  --output-strains filtered_strains.txt 2>/dev/null
  $ cat filtered_strains.txt
  SEQ_2

  $ ${AUGUR} filter \
  >  --metadata metadata.tsv \
  >  --max-date 2025 \
  >  --output-strains filtered_strains.txt 2>/dev/null
  $ cat filtered_strains.txt
  SEQ_2

Possible solution

Write a specific regular expression for this special value and handle it before handling the ambiguous date pattern.

--- a/augur/dates/__init__.py
+++ b/augur/dates/__init__.py
@@ -193,10 +193,19 @@ Matches an Augur-style ambiguous date with 'XX' used to mask unknown parts of th
 Note that this can support any date format, not just YYYY-MM-DD.
 """
 
+RE_AUGUR_MISSING_DATE = re.compile(r'^XXXX-XX-XX$')
+"""
+Matches an Augur-style ambiguous date with all parts masked.
+This only supports YYYY-MM-DD format.
+"""
+
 @cache
 def get_numerical_date_from_value(value, fmt, min_max_year=None) -> Union[float, Tuple[float, float], None]:
     value = str(value)
 
+    if RE_AUGUR_AMBIGUOUS_DATE.match(value):
+        return None
+
     # 1. Check if value is an exact date in the specified format (fmt).
 
     try:

Additional context

Noticed in nextstrain/ebola@b27d1ba