Re: [RFC PATCH 2/6] tree-walk: Add three new gentle helpers

From: Paul-Sebastian Ungureanu <ungureanupaulsebastian@gmail.com>
Date: Jul 18, 2018, 23:11 UTC
Message-ID: <ba38bdd0-c1c3-cded-0e18-f21e0d7fd9a8@gmail.com>
In-Reply-To: <xmqqin5dofor.fsf@gitster-ct.c.googlers.com>

Hello,

On 17.07.2018 21:55, Junio C Hamano wrote:

Show 13 quoted lines

> I wonder if the GENTLY option should apply to update_tree_entry()
> the same way as it would to the other codepaths that currently die
> to express "we were handed this string by the caller and told to
> give back object ID the string represents, and we found no good
> answer".  In this one (and the "bad ref" one), the existing failures
> in these two codepaths are not "we got a string and that does not
> resolve to an object name", but "we didn't have the data to work on
> to begin with (either a corrupt tree object or a corrupt ref").
> 
> In other words, it's not like "We were given HEAD:no-such-path and
> there is no such path in that tree"; it is "We tried to read HEAD:
> tree for no-such-path in it, but the tree was corrupt and we couldn't
> even tell if such a path is or is not in it", no?

I can definitely say there is a clear difference between these two cases, but I am not entirely sure how `GENTLY` should apply to `update_tree_entry()`.

On one side, even before this patch, there was the gentle version of this function, `update_tree_entry_gently()`, which could still die. And it makes sense. It should be ok to call `die()` when there was detected a "bigger" issue.

On the other side, in some cases like `read_ref_at()` I think that it could be useful if the caller could handle any error (which is what patches 3/6 and 4/6 try to accomplish).

I really do not know which way would be the best in this particular case.

Best, Paul