The spreadsheet engine parsed typed input with the invariant culture while
displaying values with the current culture, breaking the edit round-trip on
non-en-US hosts and making comma-decimal entry (10,50) impossible.
Workbook gains a Culture property (defaults to CurrentCulture) which the
component stamps from its inherited Culture parameter. It drives:
- Cell input parsing (CellData type inference), including day-month date
handling for cultures where the group and date separators collide (de-DE)
- Edit text and display rendering (GetValue, GetValueAsString, GetDisplayText)
- Number format rendering - format codes stay canonical invariant tokens while
separators, month names, and AM/PM designators follow the culture
- Formula entry and display via the new FormulaLocalizer (Excel FormulaLocal
semantics: ';' argument separators and ',' decimals in comma-decimal
cultures, lenient comma acceptance where unambiguous, canonical invariant
storage)
- Dialog input (data validation, conditional format, filter) via shared
conversion helpers on SpreadsheetDialogBase
XLSX and CSV files read and write canonical invariant values regardless of
the workbook culture, including autofit column widths. Malformed formulas now
surface as error trees instead of an unhandled lexer exception. Number format
parsing is cached (hard-bounded) since CellView reparsed per render.
Includes localization demos for the Spreadsheet and Document Processing
sections and culture test suites.
Note: headless code on non-en-US hosts now parses string values with the
host culture; set Workbook.Culture explicitly (or to InvariantCulture) for
host-independent processing.
Remove narration comments, stale/incorrect comments, and XML doc
comments on private and internal API across the Spreadsheet model,
formula functions, editor, and tools. Keep comments documenting
non-obvious rationale and XML docs on public API.
Six call sites that build a string once and either return or assign it
now go through the thread-local cache instead of allocating a fresh
StringBuilder on every call:
- NumberFormat.FormatDate
- NumberFormat.FormatNumber (outer builder; the inner thousand-grouping
builder stays a direct allocation because it is nested inside the
outer cached one)
- NumberFormat.FormatScientific
- CsvWriter.BuildContent
- Worksheet.InvalidateFormulasReferencing (formula rebuild)
- ColumnRef.ToString — additionally drops the stackalloc<char>;
appends digits least-significant first then reverses in place via
the StringBuilder indexer
Sites left as direct 'new StringBuilder()': CsvReader.ParseCsv,
NumberFormatParser.SplitSections / ParseSection, FormulaLexer
ScanTrivia / ScanStringLiteral / ScanNumericLiteral. Each either holds
two parallel builders, reuses one across many fields with Clear, or
has multiple ToString() / return paths that would need try/finally
release plumbing.
Four implementations of 'coerce a cell value to a double' had drifted
across the spreadsheet codebase, with subtle differences:
- AdvancedFilterCriteria.TryCoerceToDouble: handled DateTime via
ToOADate
- Worksheet.Filter.TryCoerce: did NOT handle DateTime at all (bug —
date filters silently failed on DateTime-typed cells)
- ConditionalFormatRules.GreaterThanRule.TryGetNumber: did NOT handle
DateTime explicitly
- NumberFormat.TryGetNumber: handled DateTime via ToNumber (the
Lotus 1-2-3-compatible variant)
The codebase canonical is ToNumber() — it correctly applies the
phantom-Feb-29-1900 quirk for dates before March 1, 1900. ToOADate
diverged for that pre-1900 range.
Consolidate into NumericCoercion.TryCoerceToDouble. Filter and
ConditionalFormat now correctly handle DateTime. NumberFormat keeps
its CellDataType gate (intentional) but delegates non-string paths
to the shared helper.
Also fixes DataValidationRule's lingering ToOADate calls so date
validation agrees with date filtering and conditional formatting
for pre-1900 dates, AND collapses 16 empty visit overrides in
FilterCriterionVisitorBase plus removes 50+ lines of trivial Accept
wrappers + dead defensive null guards in StringFilterCriterion.
Two sweeps over Radzen.Blazor/Documents/Spreadsheet and Radzen.Blazor/Spreadsheet:
- Delete inline `//` comments that merely restate the adjacent code
(e.g. "// Decrease column count" above `ColumnCount--;`, the "// Parse X"
labels above the matching ParseX(...) calls). Comments explaining *why*,
Excel/OOXML quirks, ordering requirements and trade-offs are kept.
- Delete XML doc comments (`///`) on non-public declarations — private/internal
members and the members of internal types (XlsxWriter/Reader, CsvWriter/Reader,
the formula lexer/parser, the Functions classes, AutofillCommand, etc.). Docs
on the public API surface are untouched (the build enforces this via CS1591).
Comment-only changes; no behavior change. Build clean, 1477 spreadsheet tests pass.
Replaced all == null with is null and != null with is not null
across 109 spreadsheet files. The is null pattern uses a direct
reference check that cannot be overridden by operator overloads.
Three sequential foreach loops (color extraction, date detection,
number format parsing) merged into one. The isDate flag is checked
inline so number-format fields are skipped for date sections.
137 files with namespace Radzen.Documents.Spreadsheet now live under
Documents/Spreadsheet/ directory, matching their namespace. 76 function
files in Documents/Spreadsheet/Functions/. 27 UI files remain in
Spreadsheet/.
