diff mbox series

[1/1] clone: document partial clone section

Message ID 20210506063046.23353-1-dyroneteng@gmail.com (mailing list archive)
State New
Headers show
Series [1/1] clone: document partial clone section | expand

Commit Message

Teng Long May 6, 2021, 6:30 a.m. UTC
Partial clones are created using 'git clone', but there is no related
help information in the git-clone documentation during a period. Add
a relevant section to help users understand what partial clones are
and how they differ from normal clones.

The section briefly introduces the applicable scenarios and some
precautions of partial clone. If users want to know more about its
technical design and other details, users can view the link of
git-partial-clone(7) according to the guidelines in the section.

Signed-off-by: Teng Long <dyroneteng@gmail.com>
---
 Documentation/git-clone.txt | 69 +++++++++++++++++++++++++++++++++++++
 1 file changed, 69 insertions(+)

Comments

Bagas Sanjaya May 7, 2021, 4 a.m. UTC | #1
On 06/05/21 13.30, Teng Long wrote:
> Partial clones are created using 'git clone', but there is no related
> help information in the git-clone documentation during a period. Add
> a relevant section to help users understand what partial clones are
> and how they differ from normal clones.
> 
> The section briefly introduces the applicable scenarios and some
> precautions of partial clone. If users want to know more about its
> technical design and other details, users can view the link of
> git-partial-clone(7) according to the guidelines in the section.
> 
> Signed-off-by: Teng Long <dyroneteng@gmail.com>
> ---
>   Documentation/git-clone.txt | 69 +++++++++++++++++++++++++++++++++++++
>   1 file changed, 69 insertions(+)
> 
> diff --git a/Documentation/git-clone.txt b/Documentation/git-clone.txt
> index c898310099..15495675a8 100644
> --- a/Documentation/git-clone.txt
> +++ b/Documentation/git-clone.txt
> @@ -308,6 +308,75 @@ or `--mirror` is given)
>   	for `host.xz:foo/.git`).  Cloning into an existing directory
>   	is only allowed if the directory is empty.
>   
> +Partial Clone
> +-------------
> +
> +By default, `git clone` will download every reachable object, including
> +every version of every file in the history of the repository. The **partial clone**
> +feature allows Git to transfer fewer objects and request them from the
> +remote only when they are needed, so some reachable objects can be
> +omitted from the initial `git clone` and subsequent `git fetch`
> +operations. In this way, a partial clone can reduce the network traffic
> +costs and disk space usage when git is working under a large repository.
> +
> +To use the partial clone feature, you can run `git clone` with the
> +`--filter=<filter-spec>` option. If the repository has a deep history
> +and you don't want to download any blobs, the form `filter=blob:none`
> +will omit all the blobs. If the repository has some large blobs and you
> +want to prevent some large blobs being downloaded by an appropriate
> +threshold, the form `--filter=blob:limit=<n>[kmg]` omits blobs larger
> +than n bytes or units (see linkgit:git-rev-list[1]).
> +
Why not the following?:

```
If the repository has some large blobs and you want to omit blobs larger
than desired size limit, use `--filter=blob:limit=<n>[kmg]`.
```

> +When using a partial clone, Git will request missing objects from the
> +remote(s) when necessary. Several commands that do not involve a request
> +over a network may now trigger these requests.
> +
> +For example, The <repository> contains two branches which names 'master'
> +and 'topic. Then, we clone the repository by
> +
> +    $ git clone --filter=blob:none --no-checkout <repository>
> +
> +With the `--filter=blob:none` option Git will omit all the blobs and
> +the `--no-checkout` option Git will not perform a checkout of HEAD
> +after the clone is complete. Then, we check out the remote tracking
> +'topic' branch by
> +
> +    $ git checkout -b topic origin/topic
> +
> +The output looks like
> +
> +------------
> +    remote: Enumerating objects: 1, done.
> +    remote: Counting objects: 100% (1/1), done.
> +    remote: Total 1 (delta 0), reused 0 (delta 0), pack-reused 0
> +    Receiving objects: 100% (1/1), 43 bytes | 43.00 KiB/s, done.
> +    Branch 'topic' set up to track remote branch 'topic' from 'origin'.
> +    Switched to a new branch 'topic'
> +------------
> +
> +The output is a bit surprising but it shows how partial clone works.
> +When we check out the branch 'topic' Git will request the missing blobs
> +because they are needed. Then, We can switch back to branch 'master' by
> +
> +    $ git checkout master
> +
> +This time the output looks like
> +
> +------------
> +    Switched to branch 'master'
> +    Your branch is up to date with 'origin/master'.
> +------------
> +
> +It shows that when we switch back to the previous location, the checkout
> +is done without a download because the repository has all the blobs that
> +were downloaded previously.
> +
> +`git log` may also make a surprise with partial clones. `git log
> +--<path>` will not cause downloads with the blob filters, because it's
> +only reading commits. `git log -p -- <path>` will download blobs to
> +generate the patch output and git log --raw will download all blobs
> +that changed at recent commits in order to compute renames.
> +
>   :git-clone: 1
>   include::urls.txt[]
>   
> 

This describes client-side usage. However, you missed the point that not
all server supports partial clone. If you have your own Git server, and
you want to enable partial clone, you need to invoke as user running the
server daemon (typically `git`):

`git config --global uploadpack.allowfilter true`

And you also missed the case when someone wants to remove partial clone
filters (in order to turn into full clone), for example when needed to
push to another repository as backup. See Gitlab docs for the instructions
[1].

However, brian m. carlson [CC'ed] suggested that "fetch missing objects"
step is instead be done using xargs to avoid "Argument list too long" error,
see [2].

[1]: https://docs.gitlab.com/ee/topics/git/partial_clone.html#remove-partial-clone-filtering
[2]: https://lore.kernel.org/git/YD7bczBsIR5rkqfc@camp.crustytoothpaste.net/
diff mbox series

Patch

diff --git a/Documentation/git-clone.txt b/Documentation/git-clone.txt
index c898310099..15495675a8 100644
--- a/Documentation/git-clone.txt
+++ b/Documentation/git-clone.txt
@@ -308,6 +308,75 @@  or `--mirror` is given)
 	for `host.xz:foo/.git`).  Cloning into an existing directory
 	is only allowed if the directory is empty.
 
+Partial Clone
+-------------
+
+By default, `git clone` will download every reachable object, including
+every version of every file in the history of the repository. The **partial clone**
+feature allows Git to transfer fewer objects and request them from the
+remote only when they are needed, so some reachable objects can be
+omitted from the initial `git clone` and subsequent `git fetch`
+operations. In this way, a partial clone can reduce the network traffic
+costs and disk space usage when git is working under a large repository.
+
+To use the partial clone feature, you can run `git clone` with the 
+`--filter=<filter-spec>` option. If the repository has a deep history
+and you don't want to download any blobs, the form `filter=blob:none`
+will omit all the blobs. If the repository has some large blobs and you
+want to prevent some large blobs being downloaded by an appropriate
+threshold, the form `--filter=blob:limit=<n>[kmg]` omits blobs larger
+than n bytes or units (see linkgit:git-rev-list[1]).
+
+When using a partial clone, Git will request missing objects from the
+remote(s) when necessary. Several commands that do not involve a request
+over a network may now trigger these requests.
+
+For example, The <repository> contains two branches which names 'master'
+and 'topic. Then, we clone the repository by
+
+    $ git clone --filter=blob:none --no-checkout <repository>
+
+With the `--filter=blob:none` option Git will omit all the blobs and
+the `--no-checkout` option Git will not perform a checkout of HEAD
+after the clone is complete. Then, we check out the remote tracking
+'topic' branch by
+
+    $ git checkout -b topic origin/topic 
+
+The output looks like
+
+------------
+    remote: Enumerating objects: 1, done.
+    remote: Counting objects: 100% (1/1), done.
+    remote: Total 1 (delta 0), reused 0 (delta 0), pack-reused 0
+    Receiving objects: 100% (1/1), 43 bytes | 43.00 KiB/s, done.
+    Branch 'topic' set up to track remote branch 'topic' from 'origin'.
+    Switched to a new branch 'topic'
+------------
+
+The output is a bit surprising but it shows how partial clone works.
+When we check out the branch 'topic' Git will request the missing blobs
+because they are needed. Then, We can switch back to branch 'master' by
+
+    $ git checkout master
+
+This time the output looks like
+
+------------
+    Switched to branch 'master'
+    Your branch is up to date with 'origin/master'.
+------------
+
+It shows that when we switch back to the previous location, the checkout
+is done without a download because the repository has all the blobs that
+were downloaded previously.
+
+`git log` may also make a surprise with partial clones. `git log
+--<path>` will not cause downloads with the blob filters, because it's
+only reading commits. `git log -p -- <path>` will download blobs to
+generate the patch output and git log --raw will download all blobs
+that changed at recent commits in order to compute renames.
+
 :git-clone: 1
 include::urls.txt[]