The water is [too salty to drink].
I'd interpret this as "The water is so salty that it should not be drunk".
I agree. Although I used "cannot" instead of "should not" in my paraphrase, I regard them as equivalent here, "cannot" being hyperbolic for "should not."
The infinitival clause "to drink" is licensed by "too", but it is not a complement of "too". It is a complement of the word "salty".
The reason is that there may be multiple occurrences of "too" followed by a single infinitival clause:
The water is too salty, too polluted, to drink.
Consequently, it is preferable to analyse the infinitival clause as a complement in clause structure rather than as a complement of "too" itself.
That's an interesting alternative analysis. My parsing of the infinitival clause as the complement of the degree word (Deg)
too was inspired by my memory of
Hankamer and Mikkelsen (2012), "CP Complements to D," an extremely advanced syntax paper (some of it over my head) that deals with this "too"-construction, and other constructions related to it, on pages 17-20. I especially recommend checking out tree (64) on page 18, where "too" originates with the infinitival CP in complement position and then "raises" to the higher position in the degP, leaving the string "too heavy for there to be . . ." in surface structure.
I believe that your example with
too salty, too polluted, to drink could be dealt with as a case of Right Node Raising, each instance of
too taking the same complement (
to drink) in underlying structure. Thus, the possibility of such sentences need not be interpreted as evidence against the complement analysis.