(btw, I'm obviously still reading this thread :) Thanks for the replies) On Mon, Apr 3, 2017 at 11:25 PM, Axel Wagner <axel.wagner...@googlemail.com> wrote:
> Hi, > > On Mon, Apr 3, 2017 at 11:02 PM, Kevin Powick <kpow...@gmail.com> wrote: > >> Check the error value. If an error is present, all bets are off with >> regard to any of the return values. It's simple, explicit and consistent. >> > > I've said it at least 5 times in this thread alone but: No one is actually > arguing this point (no matter how much some people seem to insist we do). > > The idea that "sometimes" an API call will return values that "may" be >> helpful, or partially useful, is a recipe of disaster. >> > > This is even further from what anyone is suggesting, than the normal level > of misunderstanding on this topic. If anything, the idea is to > *relatively* (!!!)ยน consistently return the most unhelpful value possible. > > >> Yes, some will ignore or forget to check the error value and their >> programs may continue to work (until they blow up) >> > > The idea is to prevent this by putting at least some effort into returning > obviously and intentionally invalid values that blow up immediately. It's > exactly the opposite of what you are arguing against. > > >> If anyone writing an API want's to help the users of their APIs, then >> provide meaningful, detailed information in any returned error. >> > > Again, not debated here, even in the slightest. > > > Again, the example from above should give some indication of what the > question is: In the case, that a function has a value partially initiated > when an error occurs, should it a) return a fully invalid value (be it nil, > the zero value of a primitive type or the zero value of a struct) even if > that means adding an extra branch or so and maybe an instruction or two, or > b) just return the partially initiated value. > > The argument is, that a) will help a developer, that forgets to check the > error, to discover and fix the bug more quickly and that it does very > little harm. > > > [1] if anyone suggests again that I intend to make this part of an API > contract, I'll scream. > -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
