Re: change the filetype from binary to text after the file is commited to a git repo

From: Jeff King <peff@peff.net>
Date: Jul 24, 2017, 18:18 UTC
Message-ID: <20170724181835.zo3lcv7pxmkif2jr@sigill.intra.peff.net>
In-Reply-To: <CAEor4NHDNL4CQP_FpCuJ_3F86-A3P8i=yvp1AZr-NKt0i1=cXw@mail.gmail.com>

On Mon, Jul 24, 2017 at 07:11:06AM +0200, tonka tonka wrote:

Show 8 quoted lines

> I have a problem with an already committed file into my repo. This git
> repo was converted from svn to git some years ago. Last week I have
> change some lines in a file and I saw in the diff that it is marked as
> binary (it's a simple .cpp file). I think on the first commit it was
> detected as an utf-16 file (on windows). But no matter what I do I
> can't get it back to a "normal text" text file (git does not detect
> that), but I is now only utf-8. I also replace the whole content of
> the file with just 'a' and git say it's binary.

Git doesn't store a flag for "binary-ness" on each file (though see below). As the diffs are generated on the fly when you ask to compare two versions, so too is the determination of "is this binary".

The default heuristic looks at file size (by default, if the file is over 500MB it's considered binary) and whether it has any zero-byte characters in the first few kilobytes. But note that if _either_ side of a diff is considered binary, then Git won't show a text diff.

If you want a particular diff to show all content, even if it doesn't look like text, add "-a" to your git invocation (e.g., "git show -a").

That said, you can also use .gitattributes (see "git help attributes") to mark a file as binary or not-binary, skipping the heuristic check. I'm guessing since you converted from svn that you don't have a .gitattributes file, but it's possible that somebody later added one that marks the file as binary (and so the solution would be to drop that entry).

-Peff