Lists of dates in IDS: xs:date or xs:string, and why P24M is not P2Y

The IDS editor now writes date and duration lists on xs:date and xs:duration. Both checkers compare the values as text either way, so list every form.

An IDS requirement like "the warranty starts on one of these dates" is an XSD enumeration inside an xs:restriction. IDS 1.0's DataTypes table says which base that restriction should have: xs:date for an IFCDATE property, xs:dateTime for IFCDATETIME, xs:time for IFCTIME and xs:duration for IFCDURATION. The IDS editor already wrote date and duration ranges that way, but its "one of" lists still went out on xs:string. It also couldn't edit lists that other tools had already written on xs:date, so buildingSMART's own IDS-Audit-tool test files opened with those facets as read-only XML.

As of today, a "one of" row on a date, date-time, time or duration writes the DataTypes base. Files the editor wrote before today keep xs:string when you reopen them, so they are written back byte for byte. If you want to move one of those rows to the new base, the row has a Write on xs:date button.

Does the base change any verdicts?

No, and that matters more than the base itself. IfcOpenShell's IfcTester (0.9.0) compares an enumeration value with the model value's own type and never reads the base, so for a date it compares strings. This checker also compares the text. We wrote eight lists on a test model of 61 doors, each one on both bases. The two bases gave the same failing doors every time, and IfcTester agreed with this checker on all 16.

Two of those eight show the catch. A list holding P24M does not match a door whose warranty period is P2Y. A list holding 2024-05-01Z and 2024-05-01+00:00 does not match a door dated 2024-05-01. In XSD's value space, P24M and P2Y are the same duration, and the two zoned dates are the same day as each other. Neither checker reads them that way. A list has to name each form the model actually writes, and the row now tells you so. If you want "two years or more", use a range, which this checker compares by value. (IfcTester 0.9.0 still fails every object on a date range, as we found earlier; for dates, the editor's Write as a pattern gets around that.)

Opened files

Of the 29 date and duration lists in the IDS-Audit-tool's valid test files, 13 now open as editable rows. That raises the editable share of those files from 26.5% to 28.1%, and all 12 files still write back asking the same. The other 16 lists sit in specifications that select by an attribute with no class, a shape the editor can't edit yet. None of the 64 public IDS files we track lists dates.

Measurements, scripts and the IfcTester output: eval/date-oneof-2026-11-17.md.

All writing · More in Tools and projects · All tools and research