Re: [PATCH 1/2] grep Added --blame so that grep can show result tagged with blame entries

From: Junio C Hamano <gitster@pobox.com>
Date: Mar 10, 2009, 09:46 UTC
Message-ID: <7v7i2xaewh.fsf@gitster.siamese.dyndns.org>
In-Reply-To: <49B51791.4030801@gmail.com>

pi song <pi.songs@gmail.com> writes:

Show 14 quoted lines

> This part:-
> 1) Move reusable bits from builtin-blame.c into blame.c, blame.h
> 2) Replace static variables with context struct
>
> Signed-off-by: Pi Song <pi.songs@gmail.com>
> ---
> Makefile        |    2 +
> blame.c         | 1919 +++++++++++++++++++++++++++++++++++++++++++++++++++
> blame.h         |  222 ++++++
> builtin-blame.c | 2072
> +++----------------------------------------------------
> 4 files changed, 2245 insertions(+), 1970 deletions(-)
> create mode 100644 blame.c
> create mode 100644 blame.h

This is simply too big to ask anybody sane to review (besides as we can clearly see in the above the patch is severely whitespace damaged to be usable nor mechanically reviewable).

I suspect that most of the file-scope static variables can be moved to the scoreboard, and when the reusable parts are made public, some structure and function names may need to become a bit more blame specific to avoid namespace contamination.

Perhaps the first two patches in an equivalent series that is rerolled to be reviewable would look like:

 (1) move file-scope static variables to the scoreboard, and necessary
     code changes associated with it;

     renaming of some structures and functions (if needed---I didn't
     check);

 (2) create blame.c and blame.h, and move lines from builtin-blame.c *and*
     do NOTHING OTHER THAN

     - adding #include "blame.h" to builtin-blame.c,

     - adding necessary #include at the top of blame.c,

     - surrounding blame.h with necessary

         #ifndef BLAME_H
         #define BLAME_H
	 ...
         #endif

       and finally

     - updating Makefile to add blame.c and blame.h

     This step will make a HUGE patch, and it is crucial for reviewability
     for it not to do anything other than line movement.  Ideally, the
     patch shouldn't even reorder the structures and function definitions
     during this step.

     Then we can use "git blame" itself to make sure that no other changes
     were sneaked in by mistake.  "git blame -C blame.h" and "git blame -C
     blame.c" would show everything came from builtin-blame.c.

At this point, there shouldn't still be any behaviour change nor new feature. And the titles of these preparatory step will never say anything about "grep". They are only refactoring "blame".

Once this becomes solid, you can start adding features to blame.c to support your new caller, and we can be reasonably sure that such patches can be reviewed to decide if its change breaks the existing (and so far the only) caller builtin_blame() or not.