Re: [PATCH v2 5/5] refs/debug: trim trailing LF from reflog message

From: Junio C Hamano <gitster@pobox.com>
Date: Nov 29, 2021, 19:19 UTC
Message-ID: <xmqq1r2y7ork.fsf@gitster.g>
In-Reply-To: <CAFQ2z_N20ESyzkPLdGbS9q8HEHGB7_gmaX8FUBR=jGqXLGcL1Q@mail.gmail.com>

Han-Wen Nienhuys <hanwen@google.com> writes:

Show 8 quoted lines

> On Fri, Nov 26, 2021 at 9:16 AM Junio C Hamano <gitster@pobox.com> wrote:
>>
>> The API promises to have only LF, not CRLF, at the end, so
>> strbuf_trim_trailing_newline() is a bit overkill (and if payload
>> happened to end with CR, we would lose it).
>
> it would be best if there was a way to escape characters (ie. "\n" =>
> "\\n"). Do we have a function for that?

Mere escaping would not work in a backward compatible way, without a trick. It was envisioned that we probably could encode *and* signal the fact that the message is encoded by appending a trailing SP at the end of the message. See the log message of 523fa69c (reflog: cleanse messages in the refs.c layer, 2020-07-10) for details.

Having said that, that is about introducing a whole new reflog message format (whose use is signalled by the trailing SP), and I would prefer it to happen

 (1) after we integrate with reftable, and

 (2) implemented as an option in the normalize_reflog_message()
     function, so that no ref backends has to worry about it.

outside this topic.

> I beg to differ - despite this being fewer lines of code, I think
> pointer arithmetic is best avoided if possible.

I think repeated allocation and deallocation is best avoided, and that is why I recommended it. I do not see anything to fear in poiter arithmetic, as long as it is done clearly (e.g. in a narrow scope) and correctly.

If trace_printf() does not allow counted bytes "%.*s", the whole discussion is moot; I didn't go back to check.