I wrote the question because, initially, I did genuinely wonder why (as I had existing code that used
softKeys). However, the reason was obvious on reflection and I decided to post it here, in case someone else also uses
softKeys and was wondering the same thing.
In short, the reason was that
softKeys never made any sense in the first place. Thus, its initial inclusion was in itself a mistake, one which the Guava developers are rectifying via deprecation.
In general, you use soft references if you want the object to stick around for a little after all the strong references are gone; in contrast, with weak references, the object usually gets collected soon once there are no strong or soft references left. This is useful for cached values that you want to hold on to temporarily, so that a lookup using the corresponding key will "revive" a strong reference for the value.
However, this behaviour doesn't make any sense for keys:
weakKeysmaps use an identity-based lookup, the only way to get at an entry of interest is to posess a strong reference to its key.† Thus, once there are no strong key references left, the entry is effectively dead (with no possibility of revival).
weakKeysis how long an entry remains in the map after all the strong references to its key are gone. Since such entries are dead anyway, using
weakKeysonly delays the entry's eviction, for no good reason.
Thus, most times when coming across code that uses
softKeys, a much more suitable replacement is
† I am not considering the case of fetching the entry via iteration, or anything other than key-based lookup, since maps are primarily about key-based operations.