Le 08/12/2014 19:02, Michael Brand a écrit : > > Good. My opinion is about to replace it with what. > > https://github.com/tbanel/orgaggregate#empty-and-malformed-input-cells > says: > > An input cell may be empty. In this case, it is silently replaced > by zero. In an output cell, if the computed result is zero, it not > output, leaving a blank cell. This allows for empty input cells to > result in empty output cells. > > I understand the intention very well (the proof is in the references > at the bottom ;-) ). Nevertheless I find the compromise goes too far > when in the following example the sum and mean for a0 and b0 are > empty. I would prefer 0 there even when for the time being it is at > the cost of that c results in a sum and mean of 0 too.
You instantly found the weakness of the current design! > It would mean to remove the above "In an output cell, if the computed > result is zero, it not output, leaving a blank cell. [...]". I'm inclined to agree with you. Dropping the /zero output becomes blank/ feature would be the best short-term compromise. > Or - when > you want to bother with the implementation - to change it into "If all > input cells of a computation are empty then the result cell is left > empty.". Yes, this is the correct specification. Testing that *all* inputs are blank. Unfortunately, implementing this is a lot of work, because we need to create new data structures to remember whether input fields are blank. > #+TBLNAME: original > | Item | Value | > |------+-------| > | a2 | 1 | > | a2 | 1 | > | a0 | -1 | > | a0 | 1 | > | b2 | 2 | > | b2 | | > | b0 | 0 | > | b0 | | > | c | | > | c | | > > #+BEGIN: aggregate :table original :cols "Item sum(Value) mean(Value)" > | Item | sum(Value) | mean(Value) | > |------+------------+-------------| > | a2 | 2 | 1 | > | a0 | | | > | b2 | 2 | 1 | > | b0 | | | > | c | | | > #+END > > Could you please add this example or something in the same sense to > the unittests.org before any other change? Good idea. I'll do that. > https://github.com/tbanel/orgaggregate#empty-and-malformed-input-cells > continues: > > The empty cell handling may be changed in the futur. For instance, > we may want to compute an average aggregation ignoring empty cells > (right now, empty cells contribute to the average by pulling it > toward zero). > > As I understand orgaggregate already uses Calc vectors. Yes. > Maybe then it > could use and benefit from org-table-make-reference which has the > necessary arguments and asks for a Lisp list? See > testing/lisp/test-org-table.el: > - The application of the mode string variations for TBLFM are in > test-org-table/references/mode-string-*. > - The same variations for org-table-make-reference are in > test-org-table/org-table-make-reference/mode-string-* > Definitely interesting. Someone else has already bumped into the empty cells thing. > Michael Thierry
