Request: a way to ignore .gitattributes

Request: a way to ignore .gitattributes

[Sean MacLennan]

Hi,

We have a git repo that is a mirror of an svn repo. The tools create a
huge .gitattribute file that is about 83,000 entries. Almost all are of
the form: <path> -text

This file kills git grep. A grep that takes <1s without the file takes
almost 2 minutes with the file. So git grep it unusable.

My current solution is to rename the file:

-#define GITATTRIBUTES_FILE ".gitattributes"
+#define GITATTRIBUTES_FILE ".gitattributes-no"

But I would like a cleaner solution so I don't have to maintain my own
git.

My request is more for what would be the best gitish way to solve this;
I am willing to do the actual patch.

This needs to be something in the local config and not global. My
current thoughts are:

1) A way to override the default name (much like I do now):
   gitattributesfile = .gitattributes-no

2) A flag variable:
   gitattributes = ignore
        or
   ignoreattributes = true

I personally like 1) because it would also work for the case where you
want to override a bad .gitattributes file (that you have no control
over) with a good .gitattributes file. I just think it would be more
generally useful.

But I would like to pick a solution that is most likely to get
accepted.

Cheers,
   Sean MacLennan

RE: Request: a way to ignore .gitattributes

[rsbecker]

On March 12, 2022 1:07 PM, Sean MacLennan wrote:
>We have a git repo that is a mirror of an svn repo. The tools create a huge
>.gitattribute file that is about 83,000 entries. Almost all are of the
form: <path> -
>text
>
>This file kills git grep. A grep that takes <1s without the file takes
almost 2 minutes
>with the file. So git grep it unusable.
>
>My current solution is to rename the file:
>
>-#define GITATTRIBUTES_FILE ".gitattributes"
>+#define GITATTRIBUTES_FILE ".gitattributes-no"
>
>But I would like a cleaner solution so I don't have to maintain my own git.
>
>My request is more for what would be the best gitish way to solve this; I
am willing
>to do the actual patch.
>
>This needs to be something in the local config and not global. My current
thoughts
>are:
>
>1) A way to override the default name (much like I do now):
>   gitattributesfile = .gitattributes-no
>
>2) A flag variable:
>   gitattributes = ignore
>        or
>   ignoreattributes = true
>
>I personally like 1) because it would also work for the case where you want
to
>override a bad .gitattributes file (that you have no control
>over) with a good .gitattributes file. I just think it would be more
generally useful.
>
>But I would like to pick a solution that is most likely to get accepted.

These may seem a little off the wall, but:
1. Could you use a clean/smudge approach to mess with your bad
.gitattributes file before it gets put down on disk? I realize that
registering the filter might not be possible given that you need to muck
with .gitattributes to do it, but if you have any control at all and can get
the filter in, perhaps that might be a way to clean up the bad
.gitattributes file.
2. What about a post-checkout hook that fixes .gitattributes and then does
an update-index --assume-unchanged on it. This is under the assumption that
you will never change .gitattributes in your clone.

My $0.02,
--Randall

--
Brief whoami: NonStop&UNIX developer since approximately
UNIX(421664400)
NonStop(211288444200000000)
-- In real life, I talk too much.

Re: Request: a way to ignore .gitattributes

[Junio C Hamano]

<rsbecker@nexbridge.com> writes:

> These may seem a little off the wall, but:
> 1. Could you use a clean/smudge approach to mess with your bad
> .gitattributes file before it gets put down on disk? I realize that
> registering the filter might not be possible given that you need to muck
> with .gitattributes to do it, but if you have any control at all and can get
> the filter in, perhaps that might be a way to clean up the bad
> .gitattributes file.
> 2. What about a post-checkout hook that fixes .gitattributes and then does
> an update-index --assume-unchanged on it. This is under the assumption that
> you will never change .gitattributes in your clone.

Ahh, that's cleverer than what I had initially as a knee-jerk
reaction, which was "fix the tool that emits '$path -text' for each
and every path in the repository and instead use '* -text' perhaps?".

Re: Request: a way to ignore .gitattributes

[Sean MacLennan]

On Sat, 12 Mar 2022 16:13:26 -0500
<rsbecker@nexbridge.com> wrote:

> These may seem a little off the wall, but:
> 1. Could you use a clean/smudge approach to mess with your bad
> .gitattributes file before it gets put down on disk? I realize that
> registering the filter might not be possible given that you need to
> muck with .gitattributes to do it, but if you have any control at all
> and can get the filter in, perhaps that might be a way to clean up
> the bad .gitattributes file.

I have no access to the servers.

> 2. What about a post-checkout hook that fixes .gitattributes and then
> does an update-index --assume-unchanged on it. This is under the
> assumption that you will never change .gitattributes in your clone.

That seems to work. I did not know about update-index.

> My $0.02,
> --Randall

Cheers,
   Sean