Lets consider the following case: ふぇ
Assume that we have a* width of 2 characters*, and let's use the "|" symbol to mark the breaking opportunity finally selected when rendering the text.
The ICU breaker resolves the following boundaries (fBoundaries array) [0, 1, 3, ... ] which ends up producing this result: "ふ|ぇ　"
However, there should be a breaking opportunity after "ぇ" (at least it is what I get if the line was "ふぇぇ"), which generates [0, 1, 2, ...] boundaries array, resulting in "ふぇ|ぇ" when rendering the text with the same width constraints.
Assuming normal line-breaking rules, the unicode spec states that CJ characters should be handled as ID class:
" Characters of this class may be treated as either NS or ID."
Handling CJ graphemes as ID class should imply that there are breaking opportunities before and after any of this characters. The presence of an Ideographic space (U+3000), which is a BA class grapheme, should not forbid or prevent the breaking opportunity after ぇ to be used
Hence, I consider that the boundaries array in the original case defined in this issue should be [0, 1, 2, ...] .