Re: [PATCH v1 2/8] format-patch: confirmation whenever patches exist

From: Junio C Hamano <gitster@pobox.com>
Date: May 6, 2021, 23:48 UTC
Message-ID: <xmqqk0obwhi2.fsf@gitster.g>
In-Reply-To: <20210506165102.123739-3-firminmartin24@gmail.com>

Firmin Martin <firminmartin24@gmail.com> writes:

Show 5 quoted lines

> Currently, git-format-patch, along with the option --cover-letter,
> unconditionaly overwrites a cover letter with the same name (if
> present). Although this is a desired behaviour for patches which are
> auto-generated from Git commits log, it might not be the case for a
> cover letter whose the content is meticulously written manually.

True. But if we require confirmation before overwriting patches, that would be overall worsening the end-user experience, I am afraid. In a 5-patch series with a cover-letter that was formatted, proofread, corrected with "rebase -i" and then re-formatted, unless you rephrased the titles of the patches, you'd get prompted once for the cover letter (which *IS* valuable) and five-times for patches (which is annoying).

It also is unfortunate that this change does not address another annoyance after retitling a patch---the stale output from the previous run is not overwritten with the updated one but is left in the same directory as the output files from the latest run.

So, while I very much do welcome the motivation, I am not onboard with this particular design.

Show 19 quoted lines

> diff --git a/builtin/log.c b/builtin/log.c
> index 6102893fcc..bada3db9eb 100644
> --- a/builtin/log.c
> +++ b/builtin/log.c
> @@ -35,6 +35,7 @@
>  #include "repository.h"
>  #include "commit-reach.h"
>  #include "range-diff.h"
> +#include "prompt.h"
>  
>  #define MAIL_DEFAULT_WRAP 72
>  #define COVER_FROM_AUTO_MAX_SUBJECT_LEN 100
> @@ -959,6 +960,10 @@ static int open_next_file(struct commit *commit, const char *subject,
>  			 struct rev_info *rev, int quiet)
>  {
>  	struct strbuf filename = STRBUF_INIT;
> +	struct strbuf file_exists_prompt = STRBUF_INIT;
> +	const char *yesno;
> +	static int not_prompted = 1;

When the API passes a structure that keeps track of state (like "struct rev_info *rev", used to hold rev->diffopt.file which is clearly a state), add a member to it, instead of inventing a function local static that will hurt reuse of the API you are touching. This static variable will make it impossible from a single process to format two patch series, but if it is made a part of rev_info, you do not have to introduce such a limitation.