threads / patch / 54188

patchsubmodule: suppress checking for file name ambiguity for object ids

Subject: [PATCH] submodule: suppress checking for file name ambiguity for object ids

## tl;dr

7 messages between Sep 4, 2020 and Sep 6, 2020. Diffs are folded; open one to read it.

replies: 6people: 3as markdown or json

Orgad Shaneh via GitGitGadget· Sep 4, 2020, 14:53 UTC · lore
From: Orgad Shaneh <orgads@gmail.com>

The argv argument of collect_changed_submodules() contains obly object ids (submodule references).

Notify setup_revisions() that the input is not filenames by passing assume_dashdash, so it can avoid redundant stat for each ref.

A better improvement would be to pass oid_array instead of stringified argv, but that will require a larger change, which can be done later.

Signed-off-by: Orgad Shaneh <orgads@gmail.com>
---
    submodule: suppress checking for file name ambiguity for object ids
    
    The argv argument of collect_changed_submodules() contains obly object
    ids (submodule references).
    
    Notify setup_revisions() that the input is not filenames by passing
    assume_dashdash, so it can avoid redundant stat for each ref.
    
    A better improvement would be to pass oid_array instead of stringified
    argv, but that will require a larger change, which can be done later.
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-725%2Forgads%2Fsubmodule-not-filename-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-725/orgads/submodule-not-filename-v1
Pull-Request: https://github.com/gitgitgadget/git/pull/725
 submodule.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
Show changes to submodule.c +4 −1
diff --git a/submodule.c b/submodule.c
index 3cbcf40dfc..9b5bfb12a3 100644
--- a/submodule.c
+++ b/submodule.c
@@ -840,9 +840,12 @@ static void collect_changed_submodules(struct repository *r,
 {
 	struct rev_info rev;
 	const struct commit *commit;
+	struct setup_revision_opt s_r_opt = {
+		.assume_dashdash = 1,
+	};
 
 	repo_init_revisions(r, &rev, NULL);
-	setup_revisions(argv->nr, argv->v, &rev, NULL);
+	setup_revisions(argv->nr, argv->v, &rev, &s_r_opt);
 	if (prepare_revision_walk(&rev))
 		die(_("revision walk setup failed"));
 

base-commit: 3a238e539bcdfe3f9eb5010fd218640c1b499f7a

-- 
gitgitgadget
Orgad Shaneh via GitGitGadget· Sep 6, 2020, 05:34 UTC · re: Orgad Shaneh via GitGitGadget · lore

[PATCH v2] submodule: suppress checking for file name ambiguity for object ids

From: Orgad Shaneh <orgads@gmail.com>

The argv argument of collect_changed_submodules() contains obly object ids (the objects references of all the refs).

Notify setup_revisions() that the input is not filenames by passing assume_dashdash, so it can avoid redundant stat for each ref.

A better improvement would be to pass oid_array instead of stringified argv, but that will require a larger change, which can be done later.

Signed-off-by: Orgad Shaneh <orgads@gmail.com>
---
    submodule: suppress checking for file name ambiguity for object ids
    
    The argv argument of collect_changed_submodules() contains obly object
    ids (submodule references).
    
    Notify setup_revisions() that the input is not filenames by passing
    assume_dashdash, so it can avoid redundant stat for each ref.
    
    A better improvement would be to pass oid_array instead of stringified
    argv, but that will require a larger change, which can be done later.
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-725%2Forgads%2Fsubmodule-not-filename-v2
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-725/orgads/submodule-not-filename-v2
Pull-Request: https://github.com/gitgitgadget/git/pull/725
Range-diff vs v1:
 1:  f12112cc88 ! 1:  501ce90e9a submodule: suppress checking for file name ambiguity for object ids
     @@ Commit message
          submodule: suppress checking for file name ambiguity for object ids
      
          The argv argument of collect_changed_submodules() contains obly object ids
     -    (submodule references).
     +    (the objects references of all the refs).
      
          Notify setup_revisions() that the input is not filenames by passing
          assume_dashdash, so it can avoid redundant stat for each ref.
 submodule.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
Show changes to submodule.c +4 −1
diff --git a/submodule.c b/submodule.c
index 3cbcf40dfc..9b5bfb12a3 100644
--- a/submodule.c
+++ b/submodule.c
@@ -840,9 +840,12 @@ static void collect_changed_submodules(struct repository *r,
 {
 	struct rev_info rev;
 	const struct commit *commit;
+	struct setup_revision_opt s_r_opt = {
+		.assume_dashdash = 1,
+	};
 
 	repo_init_revisions(r, &rev, NULL);
-	setup_revisions(argv->nr, argv->v, &rev, NULL);
+	setup_revisions(argv->nr, argv->v, &rev, &s_r_opt);
 	if (prepare_revision_walk(&rev))
 		die(_("revision walk setup failed"));
 

base-commit: 3a238e539bcdfe3f9eb5010fd218640c1b499f7a

-- 
gitgitgadget
Orgad Shaneh· Sep 6, 2020, 19:25 UTC · re: Orgad Shaneh via GitGitGadget · lore

Re: [PATCH v2] submodule: suppress checking for file name ambiguity for object ids

On Sun, Sep 6, 2020 at 8:34 AM Orgad Shaneh via GitGitGadget <gitgitgadget@gmail.com> wrote:

Show 11 quoted lines
>
> From: Orgad Shaneh <orgads@gmail.com>
>
> The argv argument of collect_changed_submodules() contains obly object ids
> (the objects references of all the refs).
>
> Notify setup_revisions() that the input is not filenames by passing
> assume_dashdash, so it can avoid redundant stat for each ref.
>
> A better improvement would be to pass oid_array instead of stringified argv,
> but that will require a larger change, which can be done later.

I'm wondering if it would be possible to track all the commits that were received via the transport, instead of resolving them by ref changes, because resolving from refs requires excluding all the previously-known refs, which can be a lot. Our repository has ~35K tags, and I believe there are larger repos out there...

What do you say?
- Orgad
Orgad Shaneh via GitGitGadget· Sep 6, 2020, 20:47 UTC · re: Orgad Shaneh via GitGitGadget · lore

[PATCH v3] submodule: suppress checking for file name and ref ambiguity for object ids

From: Orgad Shaneh <orgads@gmail.com>

The argv argument of collect_changed_submodules() contains obly object ids (the objects references of all the refs).

Notify setup_revisions() that the input is not filenames by passing assume_dashdash, so it can avoid redundant stat for each ref.

Also suppress refname_ambiguity flag to avoid filesystem lookups for each object. Similar logic can be found in cat-file, pack-objects and more.

Signed-off-by: Orgad Shaneh <orgads@gmail.com>
---
    submodule: suppress checking for file name and ref ambiguity for object
    ids
    
    The argv argument of collect_changed_submodules() contains obly object
    ids (submodule references).
    
    Notify setup_revisions() that the input is not filenames by passing
    assume_dashdash, so it can avoid redundant stat for each ref.
    
    A better improvement would be to pass oid_array instead of stringified
    argv, but that will require a larger change, which can be done later.
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-725%2Forgads%2Fsubmodule-not-filename-v3
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-725/orgads/submodule-not-filename-v3
Pull-Request: https://github.com/gitgitgadget/git/pull/725
Range-diff vs v2:
 1:  501ce90e9a ! 1:  128a7244f9 submodule: suppress checking for file name ambiguity for object ids
     @@ Metadata
      Author: Orgad Shaneh <orgads@gmail.com>
      
       ## Commit message ##
     -    submodule: suppress checking for file name ambiguity for object ids
     +    submodule: suppress checking for file name and ref ambiguity for object ids
      
          The argv argument of collect_changed_submodules() contains obly object ids
          (the objects references of all the refs).
     @@ Commit message
          Notify setup_revisions() that the input is not filenames by passing
          assume_dashdash, so it can avoid redundant stat for each ref.
      
     -    A better improvement would be to pass oid_array instead of stringified argv,
     -    but that will require a larger change, which can be done later.
     +    Also suppress refname_ambiguity flag to avoid filesystem lookups for
     +    each object. Similar logic can be found in cat-file, pack-objects and more.
      
          Signed-off-by: Orgad Shaneh <orgads@gmail.com>
      
     @@ submodule.c: static void collect_changed_submodules(struct repository *r,
       {
       	struct rev_info rev;
       	const struct commit *commit;
     ++	int save_warning;
      +	struct setup_revision_opt s_r_opt = {
      +		.assume_dashdash = 1,
      +	};
       
     ++	save_warning = warn_on_object_refname_ambiguity;
     ++	warn_on_object_refname_ambiguity = 0;
       	repo_init_revisions(r, &rev, NULL);
      -	setup_revisions(argv->nr, argv->v, &rev, NULL);
      +	setup_revisions(argv->nr, argv->v, &rev, &s_r_opt);
     ++	warn_on_object_refname_ambiguity = save_warning;
       	if (prepare_revision_walk(&rev))
       		die(_("revision walk setup failed"));
       
 submodule.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
Show changes to submodule.c +8 −1
diff --git a/submodule.c b/submodule.c
index 3cbcf40dfc..e48710e423 100644
--- a/submodule.c
+++ b/submodule.c
@@ -840,9 +840,16 @@ static void collect_changed_submodules(struct repository *r,
 {
 	struct rev_info rev;
 	const struct commit *commit;
+	int save_warning;
+	struct setup_revision_opt s_r_opt = {
+		.assume_dashdash = 1,
+	};
 
+	save_warning = warn_on_object_refname_ambiguity;
+	warn_on_object_refname_ambiguity = 0;
 	repo_init_revisions(r, &rev, NULL);
-	setup_revisions(argv->nr, argv->v, &rev, NULL);
+	setup_revisions(argv->nr, argv->v, &rev, &s_r_opt);
+	warn_on_object_refname_ambiguity = save_warning;
 	if (prepare_revision_walk(&rev))
 		die(_("revision walk setup failed"));
 

base-commit: 3a238e539bcdfe3f9eb5010fd218640c1b499f7a

-- 
gitgitgadget
Orgad Shaneh via GitGitGadget· Sep 6, 2020, 20:53 UTC · re: Orgad Shaneh via GitGitGadget · lore

[PATCH v4] submodule: suppress checking for file name and ref ambiguity for object ids

From: Orgad Shaneh <orgads@gmail.com>

The argv argument of collect_changed_submodules() contains obly object ids (the objects references of all the refs).

Notify setup_revisions() that the input is not filenames by passing assume_dashdash, so it can avoid redundant stat for each ref.

Also suppress refname_ambiguity flag to avoid filesystem lookups for each object. Similar logic can be found in cat-file, pack-objects and more.

This change reduces the time for git fetch in my repo from 25s to 6s.
Signed-off-by: Orgad Shaneh <orgads@gmail.com>
---
    submodule: suppress checking for file name and ref ambiguity for object
    ids
    
    The argv argument of collect_changed_submodules() contains obly object
    ids (submodule references).
    
    Notify setup_revisions() that the input is not filenames by passing
    assume_dashdash, so it can avoid redundant stat for each ref.
    
    A better improvement would be to pass oid_array instead of stringified
    argv, but that will require a larger change, which can be done later.
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-725%2Forgads%2Fsubmodule-not-filename-v4
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-725/orgads/submodule-not-filename-v4
Pull-Request: https://github.com/gitgitgadget/git/pull/725
Range-diff vs v3:
 1:  128a7244f9 ! 1:  7134a87921 submodule: suppress checking for file name and ref ambiguity for object ids
     @@ Commit message
          Also suppress refname_ambiguity flag to avoid filesystem lookups for
          each object. Similar logic can be found in cat-file, pack-objects and more.
      
     +    This change reduces the time for git fetch in my repo from 25s to 6s.
     +
          Signed-off-by: Orgad Shaneh <orgads@gmail.com>
      
       ## submodule.c ##
 submodule.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
Show changes to submodule.c +8 −1
diff --git a/submodule.c b/submodule.c
index 3cbcf40dfc..e48710e423 100644
--- a/submodule.c
+++ b/submodule.c
@@ -840,9 +840,16 @@ static void collect_changed_submodules(struct repository *r,
 {
 	struct rev_info rev;
 	const struct commit *commit;
+	int save_warning;
+	struct setup_revision_opt s_r_opt = {
+		.assume_dashdash = 1,
+	};
 
+	save_warning = warn_on_object_refname_ambiguity;
+	warn_on_object_refname_ambiguity = 0;
 	repo_init_revisions(r, &rev, NULL);
-	setup_revisions(argv->nr, argv->v, &rev, NULL);
+	setup_revisions(argv->nr, argv->v, &rev, &s_r_opt);
+	warn_on_object_refname_ambiguity = save_warning;
 	if (prepare_revision_walk(&rev))
 		die(_("revision walk setup failed"));
 

base-commit: 3a238e539bcdfe3f9eb5010fd218640c1b499f7a

-- 
gitgitgadget
Ramsay Jones· Sep 6, 2020, 20:59 UTC · re: Orgad Shaneh via GitGitGadget · lore

Re: [PATCH v4] submodule: suppress checking for file name and ref ambiguity for object ids

On 06/09/2020 21:53, Orgad Shaneh via GitGitGadget wrote:
> From: Orgad Shaneh <orgads@gmail.com>
> 
> The argv argument of collect_changed_submodules() contains obly object ids
s/obly/only/

ATB, Ramsay Jones

Junio C Hamano· Sep 6, 2020, 21:59 UTC · re: Orgad Shaneh via GitGitGadget · lore

Re: [PATCH v2] submodule: suppress checking for file name ambiguity for object ids

"Orgad Shaneh via GitGitGadget" <gitgitgadget@gmail.com> writes:
> From: Orgad Shaneh <orgads@gmail.com>
>
> The argv argument of collect_changed_submodules() contains obly object ids
obly??? s/b/n/.
Show 7 quoted lines
> (the objects references of all the refs).
>
> Notify setup_revisions() that the input is not filenames by passing
> assume_dashdash, so it can avoid redundant stat for each ref.
>
> A better improvement would be to pass oid_array instead of stringified argv,
> but that will require a larger change, which can be done later.

A naïve way is to append "--" to the argv strvec, but since 6d5b93f2 (cherry-pick: do not expect file arguments, 2012-04-14) we made it unnecessary by introducing the flag. This is exactly how the flag was designed to be used.

Good job.
Thanks.
Show 55 quoted lines
>
> Signed-off-by: Orgad Shaneh <orgads@gmail.com>
> ---
>     submodule: suppress checking for file name ambiguity for object ids
>     
>     The argv argument of collect_changed_submodules() contains obly object
>     ids (submodule references).
>     
>     Notify setup_revisions() that the input is not filenames by passing
>     assume_dashdash, so it can avoid redundant stat for each ref.
>     
>     A better improvement would be to pass oid_array instead of stringified
>     argv, but that will require a larger change, which can be done later.
>
> Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-725%2Forgads%2Fsubmodule-not-filename-v2
> Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-725/orgads/submodule-not-filename-v2
> Pull-Request: https://github.com/gitgitgadget/git/pull/725
>
> Range-diff vs v1:
>
>  1:  f12112cc88 ! 1:  501ce90e9a submodule: suppress checking for file name ambiguity for object ids
>      @@ Commit message
>           submodule: suppress checking for file name ambiguity for object ids
>       
>           The argv argument of collect_changed_submodules() contains obly object ids
>      -    (submodule references).
>      +    (the objects references of all the refs).
>       
>           Notify setup_revisions() that the input is not filenames by passing
>           assume_dashdash, so it can avoid redundant stat for each ref.
>
>
>  submodule.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/submodule.c b/submodule.c
> index 3cbcf40dfc..9b5bfb12a3 100644
> --- a/submodule.c
> +++ b/submodule.c
> @@ -840,9 +840,12 @@ static void collect_changed_submodules(struct repository *r,
>  {
>  	struct rev_info rev;
>  	const struct commit *commit;
> +	struct setup_revision_opt s_r_opt = {
> +		.assume_dashdash = 1,
> +	};
>  
>  	repo_init_revisions(r, &rev, NULL);
> -	setup_revisions(argv->nr, argv->v, &rev, NULL);
> +	setup_revisions(argv->nr, argv->v, &rev, &s_r_opt);
>  	if (prepare_revision_walk(&rev))
>  		die(_("revision walk setup failed"));
>  
>
> base-commit: 3a238e539bcdfe3f9eb5010fd218640c1b499f7a

← back to recent threads