Re: [PATCH 6/6] fsmonitor: Use fsmonitor data in `git diff`

[Ben Peart]

On 1/5/2018 5:22 PM, Junio C Hamano wrote:

Johannes Schindelin  writes:


diff --git a/diff-lib.c b/diff-lib.c
index 8104603a3..13ff00d81 100644
--- a/diff-lib.c
+++ b/diff-lib.c
@@ -95,6 +95,9 @@ int run_diff_files(struct rev_info *revs, unsigned int option)
  
  	diff_set_mnemonic_prefix(>diffopt, "i/", "w/");
  
+	if (!(option & DIFF_SKIP_FSMONITOR))

+   refresh_fsmonitor(_index);
+
if (diff_unmerged_stage < 0)
diff_unmerged_stage = 2;


I read over this hunk five times, and only now am I able to wrap my head
around this: if we do *not* want to skip the fsmonitor data, we refresh
the fsmonitor data in the index.

That feels a bit like an unneeded double negation. Speaking for myself, I
would prefore `DIFF_IGNORE_FSMONITOR` instead, it would feel less like a
double negation then. But I am not a native speaker, so I might be wrong.


I do find the logic a bit convoluted with double negative.



It's great to see more use of the fsmonitor data.  Thanks for doing this!

I agree with the sentiment that the logic as written is confusing.  I'll 
also point out that DIFF_IGNORE_FSMONITOR would be more consistent with 
the similar CE_MATCH_IGNORE_FSMONITOR flag and logic.


I'm also confused why we would not want to use the fsmonitor data in the 
'add' case.  When would you ever need to add a file that had not been 
modified?

Re: [PATCH 6/6] fsmonitor: Use fsmonitor data in `git diff`

[Junio C Hamano]

Johannes Schindelin  writes:

>> diff --git a/diff-lib.c b/diff-lib.c
>> index 8104603a3..13ff00d81 100644
>> --- a/diff-lib.c
>> +++ b/diff-lib.c
>> @@ -95,6 +95,9 @@ int run_diff_files(struct rev_info *revs, unsigned int 
>> option)
>>  
>>  diff_set_mnemonic_prefix(>diffopt, "i/", "w/");
>>  
>> +if (!(option & DIFF_SKIP_FSMONITOR))
>> +refresh_fsmonitor(_index);
>> +
>>  if (diff_unmerged_stage < 0)
>>  diff_unmerged_stage = 2;
>
> I read over this hunk five times, and only now am I able to wrap my head
> around this: if we do *not* want to skip the fsmonitor data, we refresh
> the fsmonitor data in the index.
>
> That feels a bit like an unneeded double negation. Speaking for myself, I
> would prefore `DIFF_IGNORE_FSMONITOR` instead, it would feel less like a
> double negation then. But I am not a native speaker, so I might be wrong.

I do find the logic a bit convoluted with double negative.

Re: [PATCH 6/6] fsmonitor: Use fsmonitor data in `git diff`

[Johannes Schindelin]

Hi Alex,

On Tue, 2 Jan 2018, Alex Vandiver wrote:

> diff --git a/diff-lib.c b/diff-lib.c
> index 8104603a3..13ff00d81 100644
> --- a/diff-lib.c
> +++ b/diff-lib.c
> @@ -95,6 +95,9 @@ int run_diff_files(struct rev_info *revs, unsigned int 
> option)
>  
>   diff_set_mnemonic_prefix(>diffopt, "i/", "w/");
>  
> + if (!(option & DIFF_SKIP_FSMONITOR))
> + refresh_fsmonitor(_index);
> +
>   if (diff_unmerged_stage < 0)
>   diff_unmerged_stage = 2;

I read over this hunk five times, and only now am I able to wrap my head
around this: if we do *not* want to skip the fsmonitor data, we refresh
the fsmonitor data in the index.

That feels a bit like an unneeded double negation. Speaking for myself, I
would prefore `DIFF_IGNORE_FSMONITOR` instead, it would feel less like a
double negation then. But I am not a native speaker, so I might be wrong.

> +   if (ce->ce_flags & CE_FSMONITOR_VALID && !(option & 
> DIFF_SKIP_FSMONITOR))
> +   continue;

Since we do expect this to be called without the DIFF_SKIP_FSMONITOR flag,
I guess it makes sense to order it this way.

I still have troubles to understand why we ignore the fsmonitor data with
`git add`, though... we want to add only modified files, right? I thought
that the fsmonitor data could help performance exactly there (I am
thinking of a certain insanely large code base where a developer might
want to change only one or maybe 3 files out of an entire machine workshop
of files, and with fsmonitor it should be a really fast operation because
it should ignore all but those few files, right?)... Could you maybe try
to help me understand that better?

Thanks,
Johannes

[PATCH 6/6] fsmonitor: Use fsmonitor data in `git diff`

[Alex Vandiver]

With fsmonitor enabled, the first call to match_stat_with_submodule
calls refresh_fsmonitor, incurring the overhead of reading the list of
updated files -- but run_diff_files does not respect the
CE_FSMONITOR_VALID flag.

Make use of the fsmonitor extension to skip lstat() calls on files
that fsmonitor judged as unmodified.  Skip use of the fsmonitor
extension when called by "add", as the format_callback in such cases
expects to be called even when the file is believed to be "up to date"
with the index.

Notably, this change improves performance of the git shell prompt when
GIT_PS1_SHOWDIRTYSTATE is set.

Signed-off-by: Alex Vandiver 
---
 builtin/add.c | 2 +-
 diff-lib.c| 6 ++
 diff.h| 2 ++
 3 files changed, 9 insertions(+), 1 deletion(-)

diff --git a/builtin/add.c b/builtin/add.c
index bf01d89e2..bba20b46e 100644
--- a/builtin/add.c
+++ b/builtin/add.c
@@ -119,7 +119,7 @@ int add_files_to_cache(const char *prefix,
rev.diffopt.format_callback_data = 
rev.diffopt.flags.override_submodule_config = 1;
rev.max_count = 0; /* do not compare unmerged paths with stage #2 */
-   run_diff_files(, DIFF_RACY_IS_MODIFIED);
+   run_diff_files(, DIFF_RACY_IS_MODIFIED | DIFF_SKIP_FSMONITOR);
clear_pathspec(_data);
return !!data.add_errors;
 }
diff --git a/diff-lib.c b/diff-lib.c
index 8104603a3..13ff00d81 100644
--- a/diff-lib.c
+++ b/diff-lib.c
@@ -95,6 +95,9 @@ int run_diff_files(struct rev_info *revs, unsigned int option)
 
diff_set_mnemonic_prefix(>diffopt, "i/", "w/");
 
+   if (!(option & DIFF_SKIP_FSMONITOR))
+   refresh_fsmonitor(_index);
+
if (diff_unmerged_stage < 0)
diff_unmerged_stage = 2;
entries = active_nr;
@@ -197,6 +200,9 @@ int run_diff_files(struct rev_info *revs, unsigned int 
option)
if (ce_uptodate(ce) || ce_skip_worktree(ce))
continue;
 
+   if (ce->ce_flags & CE_FSMONITOR_VALID && !(option & 
DIFF_SKIP_FSMONITOR))
+   continue;
+
/* If CE_VALID is set, don't look at workdir for file removal */
if (ce->ce_flags & CE_VALID) {
changed = 0;
diff --git a/diff.h b/diff.h
index 36a09624f..1060bc495 100644
--- a/diff.h
+++ b/diff.h
@@ -395,6 +395,8 @@ extern const char *diff_aligned_abbrev(const struct 
object_id *sha1, int);
 #define DIFF_SILENT_ON_REMOVED 01
 /* report racily-clean paths as modified */
 #define DIFF_RACY_IS_MODIFIED 02
+/* skip loading the fsmonitor data */
+#define DIFF_SKIP_FSMONITOR 04
 extern int run_diff_files(struct rev_info *revs, unsigned int option);
 extern int run_diff_index(struct rev_info *revs, int cached);
 
-- 
2.15.1.626.gc4617b774