Re: [Libusb-devel] >=autoconf-2.63 no grokmultiline AC_CONFIG_FILES?

[Pete B. <pb...@gm...>]

On 2010.12.05 22:19, Michael Plante wrote:
> This is really
> strange, as though git has some memory across branch switches other than
> what's in the config -- like it's not cleaning up the working copy if it
> only changes in line endings when you switch branches...this may be a git
> bug.

Mmm, that could explain the weird results I had on supertest.

> As an example, if I apply the patch on top of your master now in Linux
> (with autocrlf/safecrlf unspecified, so false), it does work:
>
> $ git clone http://git.libusb.org/libusb-pbatard.git .

OK, if you clone my master, the files should be CRLF in your local git repo

> $ emacs .gitattributes
> [...changed to eol=crlf...]
>
> $ git status
>    #       modified:   msvc/libusb.dsw
>    no changes added to commit (use "git add" and/or "git commit -a")
>
> $ hexdump -C msvc/libusb.dsw |less
> 00000030  74 20 56 65 72 73 69 6f  6e 20 36 2e 30 30 0d 0a  |t Version
> 6.00..|
>                                                       ^^ ^^
> Weird?  git marks it modified, yet the working copy is clearly still CRLF.
> (Note that I did not do any add/commit ops here.)

Maybe git expects that if you use eol=crlf, the repo should have LF 
(because CRLF conversion is applied on checkout)
But since repo is CRLF we have the following when diffing:

local CRLF -> stripped to LF (because eol=crlf) -> compared to CRLF
  => flagged as modified.

Could that explain this behaviour?

Regards,

/Pete