IDS Model Check's IDS editor lets you write a buildingSMART Information Delivery Specification by picking a class, a property set and a property. Until today a between row was offered only on numbers. Requirements like "the warranty starts on or after 1 January 2024" or "the work day is at most PT8H" could not be written.
IDS 1.0 allows them. Its DataTypes table gives IFCDATE the base xs:date, IFCDATETIME xs:dateTime, IFCTIME xs:time and IFCDURATION xs:duration, and allows minInclusive, maxInclusive, minExclusive and maxExclusive on all four. So the editor now writes them:
<property dataType="IFCDATE">
<propertySet><simpleValue>Pset_Warranty</simpleValue></propertySet>
<baseName><simpleValue>WarrantyStartDate</simpleValue></baseName>
<value><xs:restriction base="xs:date"><xs:minInclusive value="2024-01-01"/></xs:restriction></value>
</property>
Then we checked it against IfcTester
Each editor feature gets cross-checked against IfcOpenShell's IfcTester, the checker behind Bonsai (BlenderBIM) and many CI pipelines. We built a test model of 61 doors, each with a Pset_Warranty start and end date spread over 2022 to 2027, and wrote 16 ranges through the editor.
IfcTester 0.9.0 failed all 61 doors on every range. On "starts on or after 2024-01-01" this checker fails 30 doors and IfcTester fails 61.
The reason is one line in ifctester/facet.py: a range bound is compared as float(other) < float(value). float("2024-01-01") raises a ValueError, the restriction catches it, and the object fails. Durations go the same way. The file is valid and the standard says what it means, but one widely used reader can't evaluate it.
A pattern both read the same way
A pattern is the workaround that works in both checkers. Dates in YYYY-MM-DD form have a fixed width, so they sort as strings in date order. A date range is therefore a range of strings, and a range of fixed-width digit strings can be written as a regular expression one digit at a time. Take the lower bound's digit and recurse, take any digit between the bounds and allow anything after it, or take the upper bound's digit and recurse. For the calendar year 2026:
2026-(0(1-(0[1-9]|[1-9]\d)|[2-9]-\d\d)|1([0-1]-\d\d|2-([0-2]\d|3[0-1])))(Z|[+\-]\d{2}:\d{2})?
Exclusive bounds move one day inward. The optional suffix accepts a time zone. The pattern checks the shape, not the calendar: it would accept 2026-02-31, which no model holds.
The editor now warns on every date or duration range that IfcTester fails every object on it. On date rows a Write as a pattern button turns the range into this pattern. On the 61-door model the ten date ranges, written as patterns, got the same verdict and the same failing doors in both checkers, 10 of 10. The unit test compares 157 ranges against every third day from 1985 to 2045, with and without a time zone, and finds no difference. Durations have no such rewrite, because P1Y and P365D are not the same length to every reader. There the row suggests listing exact values instead.
Smaller things in the same change
- An opened IDS that asks only that a standard property be present, written with no
dataType(Rotterdam's permit checks and bim2sim's window U-value check do this), now opens as an editable row instead of read-only XML. Over 64 public IDS files, 86.4% of facets are now editable (was 85.2%), and 17 files are fully editable (was 14). All 64 still write back asking the same thing. - With a model open, a standard property the model already carried showed "The model has it on 61 of 61. None of the model's 61 IFCDOOR object(s) has it yet." The second sentence is gone when it isn't true.
- Exclusive numeric bounds now say that IDS ranges take no rounding tolerance: "less than 2100" is compared exactly, unlike "exactly 2100".
The measurements and the scripts that reproduce them are in eval/date-ranges-2026-11-17.md. The checker and the editor run in your browser, and your files stay on your machine. IDS Model Check is MIT licensed.