From: Phillip Wood Date: Wed, 21 Feb 2018 11:42:30 GMT Subject: Re: [PATCH v2 4/9] t3701: don't hard code sha1 hash values Message-ID: <8f39e6b1-0031-5d97-f636-ae67cfdb3f1c@talktalk.net> In-Reply-To: On 20/02/18 17:39, Junio C Hamano wrote: > Phillip Wood writes: > >> From: Phillip Wood >> >> Purge the index lines from diffs so we're not hard coding sha1 hash >> values in the expected output. > > The motivation of this patch is clear, but all-zero object name for > missing side of deletion or creation patch should not change when we > transition to any hash function. Neither the permission bits shown > in the output (and whether the index line has the bits are shown on > it in the first place, i.e. the index line of a creation patch does > not, whilethe one in a modification patch does). > > So I am a bit ambivalent about this change. > > Perhaps have a filter that redacts, instead of removes, selected > pieces of information that are likely to change while hash > transition, and use that consistently to filter both the expected > output and the actual output before comparing? > Keeping the permission bits makes sense (I'd not thought of them when I created the patch) as we want to check that the file has the correct permissions. As for the all-zero object name, is it really worth leaving it in - if a file has been created or deleted then we'll still have /dev/null as the file name for one side or the other and the diff lines will show it as well. As these tests are just to check the state of the index then I'm not sure the hash values add anything. How do you feel about a filter like sed "/^ index/s/ [0-9a-f][0-9a-f]*\.\.[0-9a-f][0-9a-f]*/x/" Best Wishes Phillip