We were allocating a vector sized for the total number of table columns, taking account of colspan. This goes wrong when there's a ridiculous colspan (an example was ~1<<53). Instead count the number of column indices where a cell starts, and renumber everything to collapse duplicate spans.
We were allocating a vector sized for the total number of table columns, taking account of colspan. This goes wrong when there's a ridiculous colspan (an example was ~1<<53). Instead count the number of column indices where a cell starts, and renumber everything to collapse duplicate spans.