fix(caldav): expand RRULE for recurring events

internal/icalfix/icalfix.go

@@ -27,8 +27,20 @@ import "regexp"
 var tzidParamRe = regexp.MustCompile(`TZID=("?)([^";:\r\n]+)("?)`)
 
 // tzidLineRe matches a standalone "TZID:<name>" property line, as found
-// inside a VTIMEZONE component.
-var tzidLineRe = regexp.MustCompile(`(?m)^TZID:([^\r\n]+)$`)
+// inside a VTIMEZONE component. The trailing "(\r?)" capture group is
+// required, not cosmetic: Go's RE2 engine only matches "$" in multi-line
+// mode immediately before a literal "\n", not before "\r\n" — so without
+// consuming (and re-emitting) an optional trailing "\r" explicitly, this
+// regex silently fails to match on any CRLF-terminated feed (which is
+// every Outlook/Exchange-published ICS feed in practice), leaving the
+// VTIMEZONE's own TZID line un-normalized even though the "TZID=..."
+// parameter form (tzidParamRe, used on DTSTART/DTEND/etc.) still matches
+// fine. The practical effect was a VTIMEZONE whose declared TZID
+// ("W. Europe Standard Time") never matched the now-normalized
+// "TZID=Europe/Berlin" parameters referencing it elsewhere in the same
+// object, so any code trying to find "the VTIMEZONE for Europe/Berlin"
+// (see internal/icssub.Series.referencedTimezones) would never find one.
+var tzidLineRe = regexp.MustCompile(`(?m)^TZID:([^\r\n]+)(\r?)$`)
 
 // NormalizeTimeZones rewrites any recognized Windows timezone identifier
 // in data to its IANA equivalent, leaving everything else (including
@@ -51,7 +63,12 @@ func NormalizeTimeZones(data []byte) []byte {
 		if !ok {
 			return m
 		}
-		return []byte("TZID:" + iana)
+		// sub[2] is the optional trailing "\r" the regex consumed as part
+		// of the match (see the doc comment above) — it must be re-emitted
+		// here, or a CRLF-terminated line loses its "\r" and every
+		// subsequent line in the file ends up misaligned relative to the
+		// original byte offsets a caller might have recorded.
+		return []byte("TZID:" + iana + string(sub[2]))
 	})
 	return data
 }

internal/icalfix/icalfix_test.go

@@ -69,3 +69,36 @@ func TestNormalizeTimeZonesLeavesUnknownAndIANAZonesAlone(t *testing.T) {
 		t.Fatalf("expected data to be unchanged, got:\n%s", got)
 	}
 }
+
+// TestNormalizeTimeZonesFixesVTIMEZONECRLFLine reproduces a real bug: the
+// standalone "TZID:<name>" property line inside a VTIMEZONE component
+// (as opposed to a "TZID=<name>" parameter on DTSTART/DTEND/etc.) was
+// never being normalized on real-world feeds, because every
+// Outlook/Exchange-published ICS uses CRLF line endings and Go's RE2 "$"
+// anchor in multi-line mode only matches immediately before a literal
+// "\n" — not before "\r\n". So on a line like
+// "TZID:W. Europe Standard Time\r\n", the old
+// `^TZID:([^\r\n]+)$` pattern never matched at all, leaving the
+// VTIMEZONE's own declared TZID un-normalized while every "TZID=..."
+// parameter elsewhere in the same object WAS normalized (that regex has
+// no such issue) — so nothing referencing the VTIMEZONE by name could
+// ever find it again. See internal/icssub.Series.referencedTimezones,
+// which depends on exactly this match to embed the right VTIMEZONE in a
+// synthetic per-series VCALENDAR.
+func TestNormalizeTimeZonesFixesVTIMEZONECRLFLine(t *testing.T) {
+	const raw = "BEGIN:VCALENDAR\r\n" +
+		"BEGIN:VTIMEZONE\r\n" +
+		"TZID:W. Europe Standard Time\r\n" +
+		"END:VTIMEZONE\r\n" +
+		"END:VCALENDAR\r\n"
+	const want = "BEGIN:VCALENDAR\r\n" +
+		"BEGIN:VTIMEZONE\r\n" +
+		"TZID:Europe/Berlin\r\n" +
+		"END:VTIMEZONE\r\n" +
+		"END:VCALENDAR\r\n"
+
+	got := string(NormalizeTimeZones([]byte(raw)))
+	if got != want {
+		t.Fatalf("CRLF-terminated VTIMEZONE TZID line was not normalized:\ngot:  %q\nwant: %q", got, want)
+	}
+}