Comment author: @gasche

I think that the arguments are applied in the reverse order: if the function argument has type ?l1:t1 -> ?l2:t2 -> ?l3:t3 -> t -> u, then your patch produces the application (foo ~l3:None ~l2:None ~l1:None) (because "args" are acumulated by a tail-rec function in reverse position). This is correct as label order is not significant, but I would still have a preference for having the label order, in this generated code, mirror the order in the type, for easier readability of the resulting code.

Finally, it's the second time I've had to read this piece of code and it still made my head hurt. In the hope of helping future reviews of the code, I have written a more verbose comment about what it's doing (patch attached). Jacques, would you agree to include it in the code? (Do you have a policy for comments in general? Adding a tad more comment is something I sometimes wish to do when reading the type-checker code.)

Comment author: turpin

I didn't pay attention to the order of arguments, I didn't change the order in which they were produced but since the labels were missing we had "args = rev args", which is no longer true. Is the tail-recursion really needed here ? Who needs so many optional arguments ?

Comment author: @garrigue

This was already fixed in trunk by Alain, but I cannot find it in Mantis.
Anyway, I merged Alain's fix (commits 12919 and 12920) in 4.00@13050.

Comment author: @garrigue

At 13050 in 4.00 and 12919-12920 in trunk.