git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: git diff-index with relative git-dir does not work

From
Yasushi SHOJI <yashi@atmark-techno.com>
Date
Feb 9, 2010, 13:10 UTC
Message-ID
<87wrymwo5a.wl@dns1.atmark-techno.com>
In-Reply-To
<fcaeb9bf1002090358v3d7f69d5ra80c186d30a1304d@mail.gmail.com>

At Tue, 9 Feb 2010 18:58:51 +0700, Nguyen Thai Ngoc Duy wrote:

Show 16 quoted lines
> 
> On 2/9/10, Yasushi SHOJI <yashi@atmark-techno.com> wrote:
> >  ...
> >  This is because static variable 'base' in sha1_file_name is already
> >  assigned _before_ setup_work_tree() from cmd_diff_index() is
> >  called. setup_work_tree() eventually chdir to the given work tree dir,
> >  but we use the old base to generate object file path. And that cause
> >  open(2) to fail because the object file path and the current dir is
> >  not in sync any more.
> >
> >  So, is it correct to assume that we must call setup_work_tree()
> >  _before_ any function which call getter/setter in environment.c?  This
> >  including open_sha1_file, in this case.
> 
> We must if gitdir is relative to cwd (and will be moved by
> setup_work_tree). Or just make gitdir absolute path.
ok. thanks.
Show 7 quoted lines
> >  Also, would it be a good idea to make all builtin command to
> >  _explicitly_ call setup_* functions, so that we can find calling order
> >  bug?
> 
> If you agree that writing "RUN_SETUP" in git.c is explicit, then all
> builtin commands do explictly call setup_*. It's about relative
> directories and cwd being moved around.
oops.  I had omitted too much words.

In the diff-index case, it, indeed, has RUN_SETUP explicitly set. however, it does not have NEED_WORK_TREE set. And, this is correct in the current semantics because diff-index is a tool to compare the index and the object store. it does not need a work tree.

However, diff-index is used in describe which need a work tree if --dirty is given. That means that diff-index might be called with --work-tree.

> >  In that case, we must change the setup functions signature to
> >  allow marking "not interested" or something.
> 
> I'm not sure I get your idea.

Given that in the current form of git, many built-in command is called by many other built-in commands. It is hard to predict what is needed and what's not. Plus, --git-dir and --work-tree are options to git itself not built-in's. So, I thought it might be a good idea to call, say, setup_work_tree_with_abs_path(), regardless of NEED_WORK_TREE, to explicitly setup run time environment before any other part of the code call, say, open_sha1_file.

If calling those functions are not acceptable due to speed or other issues, making them debug / poison code might be enough.

-- 
         yashi
Previous: Nguyen Thai Ngoc DuyNext: Nguyen Thai Ngoc Duy
Message 3 of 4 in “git diff-index with relative git-dir does not work”
  1. Yasushi SHOJIFeb 9, 2010
  2. Nguyen Thai Ngoc DuyFeb 9, 2010
  3. Yasushi SHOJIFeb 9, 2010
  4. Nguyen Thai Ngoc DuyFeb 11, 2010

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.