On Mon, Dec 11, 2017 at 10:32:49AM -0800, Jonathan Tan wrote: > > Other than that philosophical point, the documentation you added looks > > pretty good to me. Two possible improvements to the API we could do on > > top: > > > > 1. Should there be a DECORATION_INIT macro (possibly taking the "name" > > as an argument)? (Actually, the whole name thing seems like a > > confusing and bad API design in the first place). > > Agreed about the "name" thing. I'll add a DECORATION_INIT when I make > the next reroll, but I think that having it with no argument is best > (and instantiating "name" with NULL).
That will leave callers like the one in log-tree unable to use the macro, since it uses a static initializer. I didn't dig, though. That may be the only one. Most of the rest seem to just get explicitly zero-initialized (some via xcalloc of a larger struct, so maybe we should just promise that zero-initialization is always OK). > > 2. This is really just an oidmap to a void pointer. I wonder if we > > ought to be wrapping that code (I think we still want some > > interface so that the caller doesn't have to declare their own > > structs). > > It is slightly different from oidmap in that this uses "struct object *" > as a key whereas oidmap uses "struct object_id", meaning that a user of > "decorate" must already have objects allocated or be willing to allocate > them, whereas a user of "oidmap" doesn't. Ah, right. I was thinking the difference was only on the "value" half being a pointer versus a struct. It's nice in the current code that decorations do not incur the extra cost of storing the oid twice (once in the "struct object", and then again in the map key). OTOH, that might well be premature optimization. -Peff
