latex/tabularx multcolumn and column width

Question

I was reading the tabularx documentation and stumbled over a few things unclear to me.

1. In 4.3. Point 1 suggests that the the sum (of the fractions before \hsize) should always be the number of columns used. Will it do any harm if this does not fit (except that if you calculated the numbers in advance, the columns width won't exactly fit the calculation (I'm asking just, because sometimes you do the maths but then you think that one column should be a bit bigger, so one just want to increase that one number instead of recalculating everything))

2. In 4.3. Point 2: What is meant by crossing? Should one have no multicolumn [lXl] (going over the left and right bound of a X column), [XX] (spanning over multiple X columns) or even [lX] (contain a X column)? - And why is that a problem, when the size does not fit, the package will just recalculate the sizes (from my understanding)

3. Why should \hsize be used to "weight" the column width and not another length like \testwidht or \linewidth (from what I know, they are somewhat similar (and on most occasions the same)?

Answer 1

1. tabularx really won't do anything sensible if you don't make the sizes add up. It does something and in some cases you may accept the result but any output in that case is accidental untested code paths

2. it means if you have an X column in column 2 you shouldn't have a \multicolumn{..}{c}{...} which spans over column 2, but has no X. You can break this rule sometimes and sometimes you can't. Details are a bit fuzzy 30 years after I wrote the code:-)

3. because you need to set \hsize as that is the tex primitive used for linebreaking widths. It's a good idea to set \linewidth to match (as shown in that item) so then nested latex lists will use the right length.

The reason (2) is a problem is that it essentially makes you violate (1). If the width or a column declared as X is actually forced by an entry with a fixed natural length \multicolumn{1}{c}{long text here} then tabularx essentially ignores the X in that column and works as if there is one less X column in the table, and re-calulates. that's fine if it is X that was being spanned but if it was >{\hsize=.5\hsize}X then when this column is ignored and the number of active X columns is reduced the remaining modified widths will not meet the constraint set by (1).