Allow any non-numeric character to delimit timestamps

srt.py

@@ -12,15 +12,17 @@
 log = logging.getLogger(__name__)
 
 SRT_REGEX = re.compile(
-    r'(\d+)\n(\d+:\d+:\d+[,.]\d+) --> (\d+:\d+:\d+[,.]\d+) ?([^\n]*)\n(.*?)'
+    # We use "(\S+)" as the timestamp regex to avoid duplicating work and code
+    # that we would do in srt_timestamp_to_timedelta later anyway.
+    r'(\d+)\n(\S+) --> (\S+) ?([^\n]*)\n(.*?)'
     # Many sub editors don't add a blank line to the end, and many editors
     # accept it. We allow it in input.
     r'(?:\n|\Z)(?:\n|\Z)'
     # Some SRT blocks, while this is technically invalid, have blank lines
     # inside the subtitle content. We look ahead a little to check that the
     # next lines look like an index and a timestamp as a best-effort
     # solution to work around these.
-    r'(?=(?:\d+\n\d+:|\Z))',
+    r'(?=(?:\d+\n[^\n]+ -->|\Z))',
     re.DOTALL,
 )
 
@@ -144,14 +146,21 @@ def srt_timestamp_to_timedelta(srt_timestamp):
     r'''
     Convert an SRT timestamp to a :py:class:`~datetime.timedelta`.
 
+    We're intentionally extremely liberal with the delimiters we will accept.
+    The only truly "acceptable" ones are ":" to delimit hours, minutes, and
+    seconds, and "," to delimit the milliseconds, but we will accept any
+    non-numeric character as a delimiter. We do this as many SRT editors use
+    non-standard delimiters, including "." and "|", but it's quite possible
+    (and likely) that there are even more that we can't predict.
+
     .. doctest::
 
         >>> srt_timestamp_to_timedelta('01:23:04,000')
         datetime.timedelta(0, 4984)
     '''
-    # "." is not technically a legal separator, but some subtitle editors use
-    # it to delimit msecs, and some players accept it.
-    hrs, mins, secs, msecs = (int(x) for x in re.split('[,:.]', srt_timestamp))
+    hrs, mins, secs, msecs = (
+        int(x) for x in re.split(r'[^\d]', srt_timestamp)
+    )
     return timedelta(hours=hrs, minutes=mins, seconds=secs, milliseconds=msecs)
 
 

tests/test_srt.py

@@ -279,7 +279,7 @@ def test_parser_noncontiguous(subs, fake_idx, fake_hours, garbage):
     # don't really delimit subtitles, it has to look at least a little like an
     # SRT block.
     composed = composed.replace(
-        '\n\n', '\n\n%d\n%d:%s' % (
+        '\n\n', '\n\n%d\n%d -->%s' % (
             fake_idx, fake_hours, garbage,
         )
     )
@@ -294,7 +294,7 @@ def test_parser_noncontiguous(subs, fake_idx, fake_hours, garbage):
 )
 def test_parser_didnt_match_to_end_raises(subs, fake_idx, fake_hours, garbage):
     srt_blocks = [sub.to_srt() for sub in subs]
-    garbage = '\n\n%d\n%d:%s' % (fake_idx, fake_hours, garbage)
+    garbage = '\n\n%d\n%d -->%s' % (fake_idx, fake_hours, garbage)
     srt_blocks.append(garbage)
     composed = ''.join(srt_blocks)