Re: [PATCH v3 02/12] rerere: convert flags into an enum
- From
Patrick Steinhardt <ps@pks.im>
- Date
- Nov 11, 2026, 11:02 UTC
- Message-ID
- <avRLM38NKe81u9fg@pks.im>
- In-Reply-To
- <xmqqik25zuku.fsf@gitster.g>
On Mon, Nov 09, 2026 at 11:13:53AM -0800, Junio C Hamano wrote:
Show 28 quoted lines
> Junio C Hamano <gitster@pobox.com> writes: > > > Patrick Steinhardt <ps@pks.im> writes: > > > >>> What BOTH versions lack is an explanation why RERERE_NOAUTOUPDATE is > >>> needed in the first place. In other words, "How is it different > >>> from the lack of RERERE_AUTOUPDATE bit?" is not answered here. And > >>> I think that is a lot more important thing to clarify. > >> > >> Heh, in this case _I_ thought that it's implicitly clear from the > >> comment of `RERERE_AUTOUPDATE`: it overrides "rerere.autoUpdate". I > >> figured that updating these comments would cause some discussions, and > >> that's exactly why I didn't dare to update them at all in v1 :) > > > > ;-) > > The command line can give us not two (which could be represented > with a single bit) but three choices, (0) the command line has no > preference on --[no-]autoupdate and follow the configured default, > (1) --autoupdate, or (2) --no-autoupdate, and that is exactly why we > have these two independent-looking bits. > > But two bits can express another combination (i.e., both bits on), > which probably is invalid. The underlying OPT_UYN() calls > parse_opt_tertiary() which only chooses between 1 and 2 so we should > be safe on the setting side, but I offhand do not remember if we > detect when both bits are on and BUG() out where we consume the > value.
We have this bit in `rerere_cache_setup()`:
if (flags & (RERERE_AUTOUPDATE|RERERE_NOAUTOUPDATE)) cache->autoupdate = !!(flags & RERERE_AUTOUPDATE);
So `RERERE_AUTOUPDATE` takes precedence over `RERERE_NOAUTOUPDATE`, but all callers know to only pass one of these flags. I agree though that we should detect this misuse and complain about it, it's certainly not obvious at all.
Patrick