On Tue, Aug 09, 2016 at 07:49:38AM -0400, Jeff King wrote:
> On Tue, Aug 09, 2016 at 11:33:37AM +0000, Torsten Bögershausen wrote:
>
> > > The second one seems plausible, given the history of issues with
> > > changing CRLF settings for an existing checkout. I'm not sure if it
> > > would be feasible to reset the index completely before each tested
> > > command, but that would probably solve it.
> > The content of the file has been changed (we appended the letter 'Z' to it,
> > so even if mtime is the same, st.st_size should differ.
> > And it seems as if the commit is triggered, see below.
>
> I don't think I made myself clear. It's not a question of whether there
> is something to commit. It's that when git asks the index "what is the
> sha1 of the content at this path?", the index may be able to answer
> directly (the file is up-to-date, so we return the index value), or it
> may have to go to the filesystem and read the file content. It is this
> latter which triggers convert_to_git(), which is what generates the
> message in question.
>
> For a more stripped-down example, try:
>
> git add foo
> git commit -m msg
>
> versus:
>
> git add foo
> sleep 1
> git commit -m msg
>
> In the latter case, we should not generally need convert_to_git() in the
> "commit" step. It was already done by "git add", and we reuse the cached
> result.
>
> Whereas in the first one, we may run into the racy-index problem and
> have to re-read the file to be on the safe side.
>
> -Peff
Thanks for the explanation, so there are 2 chances for a race.
I assume that the suggested "touch" will fix race#2 in most cases.
In my understanding, the change of the file size will be more reliable:
diff --git a/t/t0027-auto-crlf.sh b/t/t0027-auto-crlf.sh
index 2860d2d..9933a9b 100755
--- a/t/t0027-auto-crlf.sh
+++ b/t/t0027-auto-crlf.sh
@@ -120,6 +120,7 @@ commit_chk_wrnNNO () {
cp $f $fname &&
printf Z >>"$fname" &&
git -c core.autocrlf=$crlf add $fname 2>/dev/null &&
+ printf Z >>"$fname" &&
git -c core.autocrlf=$crlf commit -m "commit_$fname" $fname >"${pfx}_$f.err" 2>&1
done
-------------------
Does anybody agree ?
And, by the way, the convert warning may be issued twice, once in
"git add" and once in "git commit".
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html