Re: [PATCH 2/2] chainlint: reduce annotation noise-factor

From: Junio C Hamano <gitster@pobox.com>
Date: Aug 30, 2024, 23:51 UTC
Message-ID: <xmqqle0dpo1p.fsf@gitster.g>
In-Reply-To: <CAPig+cSZ8Sot9oq+rmzBTmQU-Fnay92roTO=Mk0uT+-JUzMcXw@mail.gmail.com>

Eric Sunshine <sunshine@sunshineco.com> writes:

Show 8 quoted lines

> It may be possible to do something like this instead (untested), but
> I'm not sure it's worth the complexity:
>
>     $checked .= substr($body, $start, $pos - $start);
>     $checked .= ' ' unless $checked =~ /\s$/;
>     $checked .= "$erropenERR $err$errclose";
>     $checked .= ' ' unless $pos + 1 >= length($body) ||
>         substr($body, $pos + 1, 1) =~ /\s/;

I think the complexity you mention is the updates to existing code to get to the above end state? Using some setup like ...

	($erropen, errclose) = 
		$colored_output ? ("?!", "?!") : ("<RED>", "<RESET>");

... and then using a code like the above would be quite straightforward and the end result cannot become simpler than that ;-)

Show 9 quoted lines

> As first implemented, there was no structured "problem description".
> chainlint originally just output a stream of raw parse tokens (not the
> original test text), and when a problem was discovered the "?!...?!"
> annotations were embedded directly in the output stream. This was
> still the case even when colored output was implemented[1]; in fact,
> the annotations were colored after-the-fact by searching for "?!...?!"
> in the output stream. It was only when chainlint was taught to output
> the original test text verbatim[2] that problem descriptions became
> structured data.

Exactly.

Thanks.