wgpu's default DX12 compiler (Auto) loads the first dxcompiler.dll on the
DLL search path. GridCraft ships none, so another program's copy is used;
without a matching dxil.dll its 'DXIL signing library not found' warning
fails wgpu's internal compute pipelines, the device is lost and the app
exits before the first frame. Pin FXC (part of Windows) unless
WGPU_DX12_COMPILER asks for another compiler.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Cells with shrinkToFit drew at full size: text spilled or was clipped and
numbers showed ####. Too-wide content now gets a smaller font that fits the
cell or merged area, matching the PDF/print path.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
XLOOKUP returned only values, so it couldn't end a range or feed
functions that need a reference: SUM(XLOOKUP("b",A1:A4,B1:B4):
XLOOKUP("d",A1:A4,B1:B4)) and SUM(B1:XLOOKUP(…)) gave #VALUE!, and so
did CELL("address",XLOOKUP(…)). A match on an empty cell gave 0, so
ISBLANK(XLOOKUP(…)) was FALSE and "x"&XLOOKUP(…) gave "x0" (#374). In
Excel XLOOKUP returns a reference to the matching row (or column) of
its result range. The evaluator now does so, finding the position with
XMATCH; arrays of lookup values, array arguments and errors still go
through the library function.
Excel stores a merged range's outline on the cells along each edge; the grid
only used the anchor cell's borders, so a boxed merge showed only top and left.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Validation formulas were evaluated as written, at every cell of the
range, and against the cell's old value. A Custom rule =A1>10 on A1:A9
rejected 20 typed into A1 (A1 was still empty) and checked A1, not A5,
when typing into A5; a Whole number rule "<= B1" on C1:C9 compared C2
with B1. As in Excel, the formulas are now relative to the first cell
of the range and see the value being entered.
A name whose references have no $ was evaluated as if they were
absolute. In Excel such references are relative: defined from B2,
=Sheet1!A2 means "the cell to the left", and the file stores it
relative to A1 (Sheet1!XFD1, wrapping around the sheet's edge).
Name formulas with relative references are now moved by the using
cell's offset from A1, wrapping at the edges, and formulas.defineName
takes relative references as relative to the active cell, as Excel's
Name Manager does, and stores them relative to A1.
INDEX took one row and one column number, so INDEX(A1:C3,{1;3},{1,3})
gave a single value and SUM(INDEX(A1:C3,{1,2,3},2)) summed one cell.
In Excel, arrays of positions return an array of the values at each
(broadcast) position. A position that picks a whole row or column gives
#CALC! there, as arrays can't nest.
These were missing from the list of functions newer than the original
file format, so they were saved without the prefix, and Excel shows
#NAME? for such formulas.
A formula calling a built-in function with too few or too many
arguments (=DATEDIF(a,b), =SUM(), =DATE(2024,1)) was stored and showed
#VALUE! with no hint about what was wrong. Entry now refuses it with
the function's signature, e.g. "DATEDIF has too few arguments. Use
DATEDIF(start_date, end_date, unit)", and leaves the cell unchanged.
Only registered functions are checked: special forms the evaluator
handles itself (IF, LET, LAMBDA...), LAMBDA names and unknown functions
are skipped. Paste and Find & Replace keep turning input that fails
entry into text. DATEDIF's description now lists its units.
Fixes#363
Claude-Session: https://claude.ai/code/session_01BGBBGBC2gRpQ7GDuwQ6WPt
The default periods (a third of the history, up to ~33k) bypassed
MAX_PERIODS, and each forecast row's three ETS formulas refit over the whole
history, so 100k history rows x 10k periods froze the app. History is now at
most 10,000 rows, periods at most 10,000, and history x periods at most
500,000 (about 3 us per history point per fit: a few seconds of recalc at
the cap). The default is clamped to the cap; explicit periods or an end past
it are refused.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
U+25FB is a real glyph in DejaVu, so the test could pass while symbols still rendered as boxes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Content types (and extensions) of kept parts come from the input file and
were written into [Content_Types].xml unescaped, so a quote or angle bracket
injected XML; they are now attribute-escaped. Each sheet re-read and kept
its VML drawing (up to 16 MB) outside MAX_KEPT_TOTAL, so many sheets sharing
one highly compressed drawing amplified memory; the drawing is now read
within and charged to the remaining kept-bytes allowance.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Brandon Thomas