Fixed
Details
Details
Assignee
Markus Scherer
Markus SchererReporter
Markus Scherer
Markus SchererComponents
Labels
Priority
Time Needed
Days
Fix versions
Created June 28, 2018 at 5:20 PM
Updated October 3, 2018 at 10:57 PM
Resolved July 1, 2018 at 8:44 PM
The LDML spec (CLDR 23) says: "Since resets always work on the existing state, the rule entries must be in the proper order."
Richard Wordingham reports that &a=x/b &v=b makes x sort like av and greater than ac, and I just verified that with the online demo.
I think Richard interprets the spec as it is intended – at the time that x is assigned CEs, it should get CE(b) appended which at that time should not have been tailored yet.
The ICU User Guide lists "Known Limitations" including "Recursive Expansion" with examples. It looks to me like ICU does not quite implement either recursive expansion or using the current state. For example, with "&a = b / c &d = c / e" I would expect either b=ac (using the current state when b is assigned) or b=ade (recursive expansion) but actually we get b=ad, see below.
We should clarify what the intended behavior is, with an eye towards difficulty of implementation, and check ICU and possibly other existing implementations.