[PATCH v2] mergetools/p4merge: Honor $TMPDIR for the /dev/null placeholder

[PATCH v2] mergetools/p4merge: Honor $TMPDIR for the /dev/null placeholder

[David Aguilar]

Use $TMPDIR when creating the /dev/null placeholder for p4merge.
This keeps it out of the current directory.

Reported-by: Jeremy Morton <admin@game-point.net>
Signed-off-by: David Aguilar <davvid@gmail.com>
---
No mktemp usage in this round.

 mergetools/p4merge | 27 +++++++++++++--------------
 1 file changed, 13 insertions(+), 14 deletions(-)

diff --git a/mergetools/p4merge b/mergetools/p4merge
index 295361a..52f7c8f 100644
--- a/mergetools/p4merge
+++ b/mergetools/p4merge
@@ -1,29 +1,21 @@
 diff_cmd () {
+	empty_file=
+
 	# p4merge does not like /dev/null
-	rm_local=
-	rm_remote=
 	if test "/dev/null" = "$LOCAL"
 	then
-		LOCAL="./p4merge-dev-null.LOCAL.$$"
-		>"$LOCAL"
-		rm_local=true
+		LOCAL="$(create_empty_file)"
 	fi
 	if test "/dev/null" = "$REMOTE"
 	then
-		REMOTE="./p4merge-dev-null.REMOTE.$$"
-		>"$REMOTE"
-		rm_remote=true
+		REMOTE="$(create_empty_file)"
 	fi
 
 	"$merge_tool_path" "$LOCAL" "$REMOTE"
 
-	if test -n "$rm_local"
-	then
-		rm -f "$LOCAL"
-	fi
-	if test -n "$rm_remote"
+	if test -n "$empty_file"
 	then
-		rm -f "$REMOTE"
+		rm -f "$empty_file"
 	fi
 }
 
@@ -33,3 +25,10 @@ merge_cmd () {
 	"$merge_tool_path" "$BASE" "$LOCAL" "$REMOTE" "$MERGED"
 	check_unchanged
 }
+
+create_empty_file () {
+	empty_file="${TMPDIR:-/tmp}/git-difftool-p4merge-empty-file.$$"
+	>"$empty_file"
+
+	printf "$empty_file"
+}
-- 
1.8.1.rc2.6.g18499ba

Re: [PATCH v2] mergetools/p4merge: Honor $TMPDIR for the /dev/null placeholder

[Junio C Hamano]

David Aguilar <davvid@gmail.com> writes:

> Use $TMPDIR when creating the /dev/null placeholder for p4merge.
> This keeps it out of the current directory.

The usual $REMOTE "this is theirs" and $LOCAL "this is ours" are
still created in the current directory, no?  It is unclear why this
"this side does not exist" case wants to be outside of the current
directory in the first place.

In other words, "This keeps it out of the current directory" only
explains what this patch changes, without explaining why it is a
good thing to do in the first place.

> +create_empty_file () {
> +	empty_file="${TMPDIR:-/tmp}/git-difftool-p4merge-empty-file.$$"
> +	>"$empty_file"
> +
> +	printf "$empty_file"
> +}

Assuming that it makes sense to create only the "this side doe not
exist, and here is a placeholder empty file" in $TMPDIR, I think
this is probably sufficient.

By the way, who is going to remove this temporary file once the
command is done?

Re: [PATCH v2] mergetools/p4merge: Honor $TMPDIR for the /dev/null placeholder

[Junio C Hamano]

Junio C Hamano <gitster@pobox.com> writes:

> By the way, who is going to remove this temporary file once the
> command is done?

Nevermind; I can see that once the backend returns it is removed in
the same function.

Re: [PATCH v2] mergetools/p4merge: Honor $TMPDIR for the /dev/null placeholder

[David Aguilar]

On Fri, Dec 21, 2012 at 8:08 AM, Junio C Hamano <gitster@pobox.com> wrote:
> David Aguilar <davvid@gmail.com> writes:
>
>> Use $TMPDIR when creating the /dev/null placeholder for p4merge.
>> This keeps it out of the current directory.
>
> The usual $REMOTE "this is theirs" and $LOCAL "this is ours" are
> still created in the current directory, no?  It is unclear why this
> "this side does not exist" case wants to be outside of the current
> directory in the first place.

I'll take this as a hint that I should improve the commit message.
As you alluded to in your earlier email, we are in feature freeze
so I will be holding off on sending it until things have settled.

I don't believe there is a strong reason why the file should be in one
place vs. another, and the explanation is different depending on who
is driving the scriptlet.

When difftool drives then $LOCAL and $REMOTE may point to
temporary files created by the GIT_EXTERNAL_DIFF machinery.
For that use case, this commit makes things consistent with
those location of paths.

When mergetool drives it creates $LOCAL and $REMOTE in
the current directory.  They contain the different stages
of the index and can be helpful to the user when resolving merges.

The temporary /dev/null placeholder is a workaround for p4merge
and from the user's point of view this file is never interesting, so
writing it into the worktree is distracting to the user.
I could also say that it makes it consistent because "/dev" is
also not in the current directory, but that's a stretch ;-)

> In other words, "This keeps it out of the current directory" only
> explains what this patch changes, without explaining why it is a
> good thing to do in the first place.

I'll try and write up a replacement patch but I'll hold off
on sending it out until after things have settled down.

FWIW I'm seeing a "Bus Error" when doing "git update-index --refresh"
in a repository with large files on a 32bit machine. I'm not sure if
that counts as a regression since the same error occurs in 1.7.10.4
(debian testing).

I'll start another thread on this topic in case anyone is interested.

>> +create_empty_file () {
>> +     empty_file="${TMPDIR:-/tmp}/git-difftool-p4merge-empty-file.$$"
>> +     >"$empty_file"
>> +
>> +     printf "$empty_file"
>> +}
>
> Assuming that it makes sense to create only the "this side doe not
> exist, and here is a placeholder empty file" in $TMPDIR, I think
> this is probably sufficient.

Yup.  I mulled over whether or not to embed "LOCAL" and "REMOTE"
in the name but eventually decided that it was not worth the effort.
It makes the code much simpler when they share the placeholder
since we no longer need to track them separately.
-- 
David

Re: [PATCH v2] mergetools/p4merge: Honor $TMPDIR for the /dev/null placeholder

[David Aguilar]

On Sat, Dec 22, 2012 at 1:56 PM, David Aguilar <davvid@gmail.com> wrote:
> FWIW I'm seeing a "Bus Error" when doing "git update-index --refresh"
> in a repository with large files on a 32bit machine. I'm not sure if
> that counts as a regression since the same error occurs in 1.7.10.4
> (debian testing).
>
> I'll start another thread on this topic in case anyone is interested.

Nevermind.  The machine is an old salvaged laptop and its disk
probably has some bad blocks.

I wasn't able to copy the bad repo since 'cp' kept failing to
copy the index file.  I/O errors..  After deleting the borked
index the problems went away, so I'll chalk this up to
failing hardware.  Sorry for the false alarm.
-- 
David