git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [PATCH 3/3] diff: convert flags to be stored in bitfields

From
Junio C Hamano <gitster@pobox.com>
Date
Oct 29, 2017, 01:55 UTC
Message-ID
<xmqqk1zeafaq.fsf@gitster.mtv.corp.google.com>
In-Reply-To
<20171027222853.180981-4-bmwill@google.com>
Brandon Williams <bmwill@google.com> writes:
> We have have reached the limit of the number of flags that can be stored
s/have have/have/; but bit #9 is unused.

"We cannot add many more flags even if we wanted to" would be more flexible and does not take this change hostage to whatever topic that tries to claim that bit, I would think.

Show 14 quoted lines
> in a single unsigned int.  In order to allow for more flags to be added
> to the diff machinery in the future this patch converts the flags to be
> stored in bitfields in 'struct diff_flags'.
>
> Signed-off-by: Brandon Williams <bmwill@google.com>
> ---
>  builtin/commit.c |  7 +++--
>  builtin/log.c    |  2 +-
>  diff-lib.c       |  6 ++--
>  diff.c           |  3 +-
>  diff.h           | 96 +++++++++++++++++++++++++++++++++-----------------------
>  sequencer.c      |  5 +--
>  6 files changed, 70 insertions(+), 49 deletions(-)
>
Show 22 quoted lines
> diff --git a/diff.h b/diff.h
> index aca150ba2..d58f06106 100644
> --- a/diff.h
> +++ b/diff.h
> @@ -60,42 +60,59 @@ typedef struct strbuf *(*diff_prefix_fn_t)(struct diff_options *opt, void *data)
>  
>  #define DIFF_FORMAT_CALLBACK	0x1000
>  
> -#define DIFF_OPT_RECURSIVE           (1 <<  0)
> -#define DIFF_OPT_TREE_IN_RECURSIVE   (1 <<  1)
> -#define DIFF_OPT_BINARY              (1 <<  2)
> -...
> -#define DIFF_OPT_FUNCCONTEXT         (1 << 29)
> -#define DIFF_OPT_PICKAXE_IGNORE_CASE (1 << 30)
> -#define DIFF_OPT_DEFAULT_FOLLOW_RENAMES (1U << 31)
> -
> -#define DIFF_OPT_TST(opts, flag)    ((opts)->flags & DIFF_OPT_##flag)
> -#define DIFF_OPT_TOUCHED(opts, flag)    ((opts)->touched_flags & DIFF_OPT_##flag)
> -#define DIFF_OPT_SET(opts, flag)    (((opts)->flags |= DIFF_OPT_##flag),((opts)->touched_flags |= DIFF_OPT_##flag))
> -#define DIFF_OPT_CLR(opts, flag)    (((opts)->flags &= ~DIFF_OPT_##flag),((opts)->touched_flags |= DIFF_OPT_##flag))
> +#define DIFF_FLAGS_INIT { 0 }
> +extern const struct diff_flags diff_flags_cleared;

This thing is curious. Seeing the scary diff_flags_or(), I would have expected we'd have diff_flags_clear(&flags).

Show 22 quoted lines
> +struct diff_flags {
> +	unsigned RECURSIVE:1;
> +	unsigned TREE_IN_RECURSIVE:1;
> +	unsigned BINARY:1;
> +...
> +	unsigned FUNCCONTEXT:1;
> +	unsigned PICKAXE_IGNORE_CASE:1;
> +	unsigned DEFAULT_FOLLOW_RENAMES:1;
> +};
> +
> +static inline struct diff_flags diff_flags_or(struct diff_flags a,
> +					      struct diff_flags b)
> +{
> +	char *tmp_a = (char *)&a;
> +	char *tmp_b = (char *)&b;
> +	int i;
> +
> +	for (i = 0; i < sizeof(struct diff_flags); i++)
> +		tmp_a[i] |= tmp_b[i];
> +
> +	return a;
> +}

This is doubly scary, but let's see why we need it by looking at the callers.

> +#define DIFF_OPT_TST(opts, flag)	((opts)->flags.flag)
> +#define DIFF_OPT_TOUCHED(opts, flag)	((opts)->touched_flags.flag)
> +#define DIFF_OPT_SET(opts, flag)	(((opts)->flags.flag = 1),((opts)->touched_flags.flag = 1))
> +#define DIFF_OPT_CLR(opts, flag)	(((opts)->flags.flag = 0),((opts)->touched_flags.flag = 1))
These are trivial and straight-forward.
Show 22 quoted lines
> +
>  #define DIFF_XDL_TST(opts, flag)    ((opts)->xdl_opts & XDF_##flag)
>  #define DIFF_XDL_SET(opts, flag)    ((opts)->xdl_opts |= XDF_##flag)
>  #define DIFF_XDL_CLR(opts, flag)    ((opts)->xdl_opts &= ~XDF_##flag)
> @@ -122,8 +139,8 @@ struct diff_options {
>  	const char *a_prefix, *b_prefix;
>  	const char *line_prefix;
>  	size_t line_prefix_length;
> -	unsigned flags;
> -	unsigned touched_flags;
> +	struct diff_flags flags;
> +	struct diff_flags touched_flags;
>  
>  	/* diff-filter bits */
>  	unsigned int filter;
> @@ -388,7 +405,8 @@ extern int diff_result_code(struct diff_options *, int);
>  
>  extern void diff_no_index(struct rev_info *, int, const char **);
>  
> -extern int index_differs_from(const char *def, int diff_flags, int ita_invisible_in_index);
> +extern int index_differs_from(const char *def, struct diff_flags flags,
> +			      int ita_invisible_in_index);

OK. I tend to think twice before passing any struct by value (even something that starts its life as a small/single-word struct), but let's see how much simpler this allows callers to become.

Show 14 quoted lines
> diff --git a/builtin/commit.c b/builtin/commit.c
> index d75b3805e..de08c2594 100644
> --- a/builtin/commit.c
> +++ b/builtin/commit.c
> @@ -912,11 +912,12 @@ static int prepare_to_commit(const char *index_file, const char *prefix,
>  			 * submodules which were manually staged, which would
>  			 * be really confusing.
>  			 */
> -			int diff_flags = DIFF_OPT_OVERRIDE_SUBMODULE_CONFIG;
> +			struct diff_flags flags = DIFF_FLAGS_INIT;
> +			flags.OVERRIDE_SUBMODULE_CONFIG = 1;
>  			if (ignore_submodule_arg &&
>  			    !strcmp(ignore_submodule_arg, "all"))
> -				diff_flags |= DIFF_OPT_IGNORE_SUBMODULES;

This couldn't have been done in patches 1+2/3 because DIFF_OPT_SET() does not take a bare 'flags' but diffopt itself, which was a bit unfortunate, but the end result after this step becomes a lot more sensible.

> -			commitable = index_differs_from(parent, diff_flags, 1);
> +				flags.IGNORE_SUBMODULES = 1;
> +			commitable = index_differs_from(parent, flags, 1);
OK.
Show 10 quoted lines
> diff --git a/builtin/log.c b/builtin/log.c
> index d81a09051..780975ed4 100644
> --- a/builtin/log.c
> +++ b/builtin/log.c
> @@ -134,7 +134,7 @@ static void cmd_log_init_defaults(struct rev_info *rev)
>  
>  	if (default_date_mode)
>  		parse_date_format(default_date_mode, &rev->date_mode);
> -	rev->diffopt.touched_flags = 0;
> +	rev->diffopt.touched_flags = diff_flags_cleared;

So this structure assignment is a more kosher way to clear everything than memset(&touched_flags, '\0', sizeof(...)); I'd still prefer

	diff_flags_clear(&rev->diffopt.touched_flags);

tough, as it is easy to forget diff_flags_cleared is a singleton constant specifically created to be assigned for clearing another flags struct.

Show 6 quoted lines
> @@ -546,7 +546,7 @@ int index_differs_from(const char *def, int diff_flags,
>  	setup_revisions(0, NULL, &rev, &opt);
>  	DIFF_OPT_SET(&rev.diffopt, QUICK);
>  	DIFF_OPT_SET(&rev.diffopt, EXIT_WITH_STATUS);
> -	rev.diffopt.flags |= diff_flags;
> +	rev.diffopt.flags = diff_flags_or(rev.diffopt.flags, flags);

In a more general case, we cannot know what flags setup_revisions() gave to rev.diffopt, and because of that we cannot do

	rev.diffopt.flags = diff_flags;
	DIFF_OPT_SET(&rev.diffopt, QUICK);
	DIFF_OPT_SET(&rev.diffopt, EXIT_WITH_STATUS);

without using the flags_or() thing. In this codepath, that currently is not a problem (because we give ac/av = 0/NULL), but it probably is a good idea to avoid depending on that.

I still haven't brought myself to like the structure being passed by value and the singleton diff_flags_cleared thing, but I suspect that we may get used to them once we start using these. I dunno.

Thanks.
Previous: Brandon WilliamsNext: Junio C Hamano
Message 7 of 47 in “convert diff flags to be stored in a struct”
  1. 0/3 convert diff flags to be stored in a structBrandon Williams, Oct 27, 2017
  2. 1/3 add: use DIFF_OPT_SET macro to set a diff flagBrandon Williams, Oct 27, 2017
  3. 2/3 reset: use DIFF_OPT_SET macro to set a diff flagBrandon Williams, Oct 27, 2017
  4. Junio C HamanoOct 29, 2017
  5. Brandon WilliamsOct 30, 2017
  6. 3/3 diff: convert flags to be stored in bitfieldsBrandon Williams, Oct 27, 2017
  7. Junio C HamanoOct 29, 2017
  8. Junio C HamanoOct 30, 2017
  9. Brandon WilliamsOct 30, 2017
  10. Junio C HamanoOct 31, 2017
  11. Brandon WilliamsOct 30, 2017
  12. Junio C HamanoOct 29, 2017
  13. Stefan BellerOct 29, 2017
  14. 0/4 convert diff flags to be stored in a structBrandon Williams, Oct 30, 2017
  15. 1/4 add, reset: use DIFF_OPT_SET macro to set a diff flagBrandon Williams, Oct 30, 2017
  16. 2/4 diff: convert flags to be stored in bitfieldsBrandon Williams, Oct 30, 2017
  17. Junio C HamanoOct 31, 2017
  18. 5/4 diff: avoid returning a struct by value from diff_flags_or()Junio C Hamano, Oct 31, 2017
  19. Brandon WilliamsOct 31, 2017
  20. 4/4 diff: remove touched flagsBrandon Williams, Oct 30, 2017
  21. 3/4 diff: add flag to indicate textconv was set via cmdlineBrandon Williams, Oct 30, 2017
  22. Stefan BellerOct 30, 2017
  23. Brandon WilliamsOct 30, 2017
  24. Brandon WilliamsOct 30, 2017
  25. Junio C HamanoOct 31, 2017
  26. 5/4 diff: set TEXTCONV_VIA_CMDLINE only when it is set to trueJunio C Hamano, Oct 31, 2017
  27. Brandon WilliamsOct 31, 2017
  28. 5/4 diff: remove DIFF_OPT_TST macroBrandon Williams, Oct 30, 2017
  29. 7/4 diff: remove DIFF_OPT_CLR macroBrandon Williams, Oct 30, 2017
  30. 6/4 diff: remove DIFF_OPT_SET macroBrandon Williams, Oct 30, 2017
  31. 8/4 diff: make struct diff_flags members lowercaseBrandon Williams, Oct 30, 2017
  32. 0/8 convert diff flags to be stored in a structBrandon Williams, Oct 31, 2017
  33. 1/8 add, reset: use DIFF_OPT_SET macro to set a diff flagBrandon Williams, Oct 31, 2017
  34. 2/8 diff: convert flags to be stored in bitfieldsBrandon Williams, Oct 31, 2017
  35. Stefan BellerOct 31, 2017
  36. Junio C HamanoNov 1, 2017
  37. Stefan BellerNov 1, 2017
  38. 4/8 diff: remove touched flagsBrandon Williams, Oct 31, 2017
  39. 5/8 diff: remove DIFF_OPT_TST macroBrandon Williams, Oct 31, 2017
  40. 3/8 diff: add flag to indicate textconv was set via cmdlineBrandon Williams, Oct 31, 2017
  41. 6/8 diff: remove DIFF_OPT_SET macroBrandon Williams, Oct 31, 2017
  42. 7/8 diff: remove DIFF_OPT_CLR macroBrandon Williams, Oct 31, 2017
  43. Stefan BellerOct 31, 2017
  44. Junio C HamanoNov 1, 2017
  45. 8/8 diff: make struct diff_flags members lowercaseBrandon Williams, Oct 31, 2017
  46. Stefan BellerOct 31, 2017
  47. Junio C HamanoNov 1, 2017

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.