Re: [RFC 0/4] ref-filter: remove printing from formatting logic

From: Martin Ågren <martin.agren@gmail.com>
Date: Mar 13, 2018, 19:26 UTC
Message-ID: <CAN0heSp-Vsvr95zc3J0iokRMJt74pxFH7OsoiR_3gnkMxpAAKw@mail.gmail.com>
In-Reply-To: <CAL21BmkmXKzdwYHu1pNxuHhaxqei4ekVbutbuv2jmv6=GgcG_A@mail.gmail.com>

Hi Olga

On 13 March 2018 at 11:25, Оля Тележная <olyatelezhnaya@gmail.com> wrote:

Show 17 quoted lines

> The main idea of the patch is, if you want to format the output by
> ref-filter, you should have an ability to work with errors by yourself
> if you want to.
> So I decided not to touch signature of show_ref_array_item(), but to
> move all printing (I mean errors) to it. So that we could invoke
> format_ref_array_item() and be sure that we cold handle errors by
> ourselves.
>
> The patch is not finished, but I decided to show it to you. There are
> still many places where we could die in the middle of formatting
> process. But, if you like the general idea, I will finish and re-send
> it.
>
> Another question is about verify_ref_format(). Do we need to allow its
> users also to manage errors by themselves? I left the old scenario,
> printing everything in verify_ref_format() and die. If you have better
> ideas, please share them.

I think it is a good idea to stop die-ing in "libgit". This seems like a good way of achieving that, or isolating the issue. Do you have any particular use-case for this, i.e., are you setting up the stage for a patch "5" where you add a new user of one of these?

I do wonder whether a helper function to call strbuf_addstr() and return -1 would be a good idea though. I mentioned it in patch 2, then with patches 3 and 4, it started to seem like a reasonably good idea. It would be a shame if this sort of "boilerplate" for handling errors could have an impact on code clarity / obviousness.

Another issue is whether passing NULL for an error-strbuf should be a way of saying "I don't care; die() so I do not have to". Well, right now I guess passing NULL would indeed terminate the program. ;-) Such a construct might be another reason for providing error_strbuf_addstr()... Of course, it also means we keep die-ing in libgit..

I feel I'm just talking out loud. Maybe you find my input useful.

Martin