On 06/12, Jeff King wrote: > On Mon, Jun 12, 2017 at 02:53:52PM -0700, Brandon Williams wrote: > > > > These all seem reasonable to me. Patch 3 made me shrug a little, because > > > it seems like the majority of C files end up including it anyway. I > > > suspect you could break config.h down into two files: the few things > > > that everybody needs (git_config() and the few parsing functions needed > > > in callbacks) and the ones for commands that actually manipulate the > > > config. > > > > > > That would reduce the surface area of the module that most callers look > > > at, but I don't think there's a huge benefit to doing so (mostly it just > > > makes re-compiling faster by decreasing the chance that a dependent > > > header has changed for each file). > > > > Yes, ultimately I think it would be a good thing to break config.c down > > into at least 2 more files (the file parsing logic and the config_set > > logic) but that can be done at a later point. I started looking at > > doing that now but that logic is a little more entangled than I thought > > it was. > > To be clear, I don't mind that sort of module refactoring and like the > results. But it almost certainly isn't the biggest bang-for-buck in > terms of the time it takes versus the benefit it brings. So take my > comment as "we could also do..." but not "we should not take your patch > because it does not go far enough". Yes, that's how I understood your comment. I realize that there's a lot of improvements that can be done across our code base and one rabbit hole that is easy to fall into is "fix all the things right now!" hence why I decided to defer doing that to a later date. And I agree that doing that extra work doesn't buy us 'all too much'. Having said that I do think it was worthwhile to remove the config declarations from cache.h. I think in the long term it would be nice to limit cache.h's scope as it is very unapproachable in its current form. -- Brandon Williams