fix: keep empty crosstab rows and columns through the reshape

[martin1english]

Reported by PamGene (bionavigator) after upgrading to 1.1.1:

Failed to run operator, exit code = 1
Error in setnames(x, value) : Can't assign 5 names to a 4-column data.table
Calls: colnames<- ... names<-.data.table -> setnames -> stopf

Cause

dcast only emits a row/column for each .ri/.ci value that actually occurs in the data. A crosstab row or column with no observations at all therefore drops out of the wide table, while the headers are still built from the full rselect() / cselect(). The two then disagree:

df_wide  <- dcast(df_long, .ri ~ .ci, value.var = ".y")   # narrower than cselect()
raw_data <- df_wide[order(.ri)][, !".ri"]
...
colnames(df_out) <- c(row_block$names, new_col_names)     # boom

The error is the good outcome

When the number of surviving rows divides evenly into the expected count, cbind() recycles instead of erroring and the export silently contains the wrong values, with only a warning:

4 crosstab rows, rows 2 and 4 empty:
      Row     0     1     2
1:     r1     1     2     3
2:     r2     4     5     6
3:     r3     1     2     3   <- should be empty; carries r1's data
4:     r4     4     5     6   <- should be empty; carries r2's data

The default crosstab-view path (collapse_cols = FALSE) is affected the same way. So an unknown number of previously "successful" exports from tables with empty rows/columns may contain misaligned data.

Fix

Pin the .ri/.ci factor levels to the full index range and reshape with drop = FALSE, so the data block always lines up with the headers and genuinely empty cells come through as NA.

Verification

Ran the real main.R against a mock context, comparing every cell value against the expected grid.

Before (as released in 1.1.1):

== collapse_cols = TRUE ==
  no holes (regression)    -> PASS
  one empty column         -> ERROR: Can't assign 5 names to a 4-column data.table
  one empty row            -> ERROR: SILENT RECYCLING
  empty row + column       -> ERROR: SILENT RECYCLING
  2 of 4 rows empty        -> FAIL values misaligned/recycled
== collapse_cols = FALSE ==
  no holes (regression)    -> PASS
  one empty column         -> FAIL shape 3x3 want 3x4
  one empty row            -> ERROR: SILENT RECYCLING
  empty row + column       -> ERROR: SILENT RECYCLING
  2 of 4 rows empty        -> FAIL values misaligned/recycled

After: all 10 PASS.

Regression checks:

• shipped tests/test_1 fixture reproduces its expected output byte-identical
• CSV, TSV and XLSX all write valid output in both collapse_cols modes, and with collapse_rows = TRUE

🤖 Generated with Claude Code

https://claude.ai/code/session_01EeoFKnewHFjCdPK45kWjDG