# Re: [BUG] repack --drop-filtered aborts when a filtered blob is referenced by the index

1 messages from 2026-11-12 to 2026-11-12. Participants: kongfankaikai.
Thread: https://gitlist.dev/t/66831

## kongfankaikai, 2026-11-12 07:49

Subject: Re: [BUG] repack --drop-filtered aborts when a filtered blob is referenced by the index
Message-ID: <tencent_EB6E67B22EC8C9D00D0A7D062B62BB95ED05@qq.com>
URL: https://gitlist.dev/e/tencent_EB6E67B22EC8C9D00D0A7D062B62BB95ED05%40qq.com

```
Hi Siddharth,

Thanks for the explanation and the update!

It's great to know that a fix is already being worked on. I'll keep an eye on the discussion.

Thanks



原始邮件
发件人：Siddharth Shrimali <r.siddharth.shrimali@gmail.com>
发件时间：2026年11月12日 15:29
收件人：kongfankaikai <kongfankaikai@foxmail.com>
抄送：git <git@vger.kernel.org>, Samuel Bronson <naesten@gmail.com>, Siddharth Shrimali <r.siddharth.shrimali@gmail.com>
主题：Re: [BUG] repack --drop-filtered aborts when a filtered blob is referenced by the index


Hi,

On Thu, Nov 12, 2026 at 12:51:07AM +0800, kongfankaikai wrote:
> When a blob selected by the filter for dropping is referenced by the
> current index, Git aborts the entire repack with:
>
> `fatal: cannot drop 'path/to/large-file.bin' (object-id): it is
> referenced by the current index`
>
> This stops other eligible blobs from being dropped, even though those
> blobs are not referenced by the index.

Thanks for the report. This is a known limitation and a fix is already
planned.

Samuel Bronson ran into exactly the same thing on a real llvm-project
clone and sent an RFC patch for it in September:

  https://lore.kernel.org/git/s0vqzjavw8p.fsf@gmail.com/

and I agreed with the direction there:

  https://lore.kernel.org/git/CAGWgyh9B=re06aofii9VFB1xOwEeTtxYE=7T14m9WFAx1ORpMg@mail.gmail.com/

> Is there a specific reason this condition is treated as fatal rather
> than skipping the affected objects?
>
> This behavior would be especially helpful for partial clones with
> large working trees.

No correctness reason. The guard only avoids dropping a blob that the
next command would lazy-fetch right back, so such a blob is never at
risk. Aborting was simply a conservative version of the guard, and as
you and Samuel point out it is a poor trade-off.

The fix is to replace the die() with a warning and drop the remaining
eligible blobs. I am working on this and will post it to the list.

Thanks,
Siddharth 




```
