Excel caches IMCSCH("13+5j") as 1.28233995888454E-06 and IMSECH("7+25j")'s imaginary part as
0.000241377759972715: below 1, a 15-digit decimal longer than 21 characters is written in
scientific notation. 9.99999999999999E-9&"" now gives 9.99999999999999E-09, not
0.00000000999999999999999. Affects every number-to-text conversion (&, CONCAT, TEXTJOIN, IM
functions, formula printing, labels).
Excel rejects a text criterion longer than 255 characters in COUNTIF(S),
SUMIF(S), AVERAGEIF(S), MAXIFS and MINIFS with #VALUE!; GridCraft matched
it anyway. Found by the calc oracle on an Excel-saved report where a
268-character concatenated criterion caches #VALUE!.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A formula under a criteria label that names no field is evaluated for each record: its relative references move from the first record to that record, and the database labels name the record's fields, as in Excel. The calc evaluator passes the criteria reference to the database functions through Ctx::computed_criterion.
The parser stripped _xludf. like _xlfn., but the two mean opposite things:
_xludf. marks a name Excel must not treat as a built-in. Keeping it in the
call name makes the call an unknown function (#NAME?, as in Excel) and the
writer emits it unchanged, so _xludf.EOMONTH() no longer saves as an
EOMONTH() call Excel repairs away.
Excel's EDATE and EOMONTH read serial 0 as 0 January 1900 and end
February 1900 on the 28th (serial 59), although the sheet itself keeps
the Lotus 29 February 1900 (serial 60). GridCraft clamped day 0 up to
day 1 and used the sheet's 29-day February, so EDATE(0,0) was 1,
EDATE(0,1) 32, EDATE(31,1) and EOMONTH(59,0) 60.
EDATE now keeps day 0 and both functions take the real calendar's month
lengths. Serial 60 still reads as 29 February, so EDATE(60,1) stays
29 March. DATE, DAY and the parser keep the leap day.
Tests: the Excel 365 table from the issue; the #124 cases that expected
60 at the end of February 1900 now expect 59.
Restores "Counts whole years, months or days between two dates." and
appends the units, in the multi-line form rustfmt gives a long entry.
Drops the 256-argument test case: the parser already refuses more than
255 arguments, so it never reached the entry check. (993c108's message
says entry "also refuses" them; it already did.)
Claude-Session: https://claude.ai/code/session_01BGBBGBC2gRpQ7GDuwQ6WPt
FnSpec::accepts holds the arity check (including Excel's 255-argument
limit) that functions::call and the entry check both use, so entry also
refuses 256 arguments to a variadic function. The evaluator's own check
in calc is unchanged.
Claude-Session: https://claude.ai/code/session_01BGBBGBC2gRpQ7GDuwQ6WPt