On Tue, Sep 22, 2020 at 01:43:47PM -0400, Brian Foster wrote: > This does slightly change behavior in that this function currently > unconditionally results in logging the associated dquot in the > transaction. I'm not sure anything really depends on that with a delta > == 0, but it might be worth documenting in the commit log. > > Also, it does seem a little odd to bail out after we've potentially > allocated ->t_dqinfo as well as assigned the current dquot a slot in the > transaction. I think that means the effect of this change is lost if > another dquot happens to be modified (with delta != 0) in the same > transaction (which might also be an odd thing to do). Yes, it seems like we should probably bail out at the very beginning for delta == 0, and document what kind of changes this theoretically causes, and why they don't matter. Btw, the function could really use a reindent, the formatting is very strange.
