Skip to contentSkip to conversation

Use of tac in git-rebase--interactive

PATCH In review Brian Gernhardt 13 participants 30 messages Jun 15, 2016

lore
Share & tools
lore
#0 lore raw

Commit d481bcc9: "Do rebase with preserve merges with advanced TODO list" uses the command tac, apparently to reverse the TODO command list. (I don't use rebase -i much, if you can't tell.)

The problem is that tac doesn't exist on my OS X system. I do appear to have a rev command which does the same thing. Simply posting a patch that does 472s/tac/rev/ would fix the problem on my system, but the fact that this is an issue raises the question of the relative portability of the two commands. I don't think we have a declared dependancy on GNU's coreutils, which is where Debian lists tac as coming from.

~~ Brian

#1 lore raw
Quoted 6 lines · Brian Gernhardt

The problem is that tac doesn't exist on my OS X system. I do appear to have a rev command which does the same thing. Simply posting a patch that does 472s/tac/rev/ would fix the problem on my system, but the fact that this is an issue raises the question of the relative portability of the two commands. I don't think we have a declared dependancy on GNU's coreutils, which is where Debian lists tac as coming from.

I know the list will be shocked to hear that Solaris has neither.

An easy perl replacement is:

perl -e 'print reverse <>'

which should work fine for small-ish input (since it puts the whole thing in memory).

-Peff

#2 lore raw

PATCH 1/1 Use perl instead of tac

Subject: [PATCH] Use perl instead of tac

tac is part of GNU coreutils and not portable. Use perl's reverse function instead, since we already require it.

---

Quoted 1 line · Jeff King

I know the list will be shocked to hear that Solaris has neither.

I am shocked and horrified. But then again, that's mostly just my reaction to Solaris in general. ;-D

Quoted 6 lines

An easy perl replacement is:

perl -e 'print reverse <>'

which should work fine for small-ish input (since it puts the whole thing in memory).

Something like this?

And I'm having problems with t3404.13 now (mark :0 invalid). And it's too late for me to track it down.

 git-rebase--interactive.sh |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
git-rebase--interactive.sh +0 -0
index 1751b08..303b754 100755
@@ -469,7 +469,7 @@ create_extended_todo_list () {
 	test -n "${last_parent:-}" -a "${last_parent:-}" != $SHORTUPSTREAM  

&& \
echo reset $last_parent
) | \
- tac | \
+ perl -e 'print reverse <>' | \
while read cmd args
do
: ${commit_mark_list:=} ${last_commit:=000}

Signature · 1 line
1.5.5.111.g180d
#3 lore raw

PATCH 1/1 rebase--interactive: Replace unportable 'tac' by a sed script.

From: Johannes Sixt <[email protected]>

---

Review

Quoted 2 lines · Brian Gernhardt
(quoted diff) +1 -1
-    tac | \
+    perl -e 'print reverse <>' | \

Here's my try, which avoids the perl hammer. ;)

Sorry, I can't test this at the moment due to an unrelated breakage that I first have to chase down.

-- Hannes

PS: I picked the sed script from this patch by Simon 'corecode' Schubert:

http://article.gmane.org/gmane.comp.version-control.git/37074

 git-rebase--interactive.sh |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
git-rebase--interactive.sh +1 -1
index 1751b08..a9ac332 100755
@@ -469,7 +469,7 @@ create_extended_todo_list () {
 	test -n "${last_parent:-}" -a "${last_parent:-}" != $SHORTUPSTREAM && \
 		echo reset $last_parent
 	) | \
-	tac | \
+	sed -ne '1!G;$p;h' | \
 	while read cmd args
 	do
 		: ${commit_mark_list:=} ${last_commit:=000}
Signature · 1 line
1.5.5.1.930.g66f94.dirty
#4 lore raw

Re: [PATCH] rebase--interactive: Replace unportable 'tac' by a sed script.

git-rebase--interactive.sh create_extended_todo_list () { · line 472

Quoted 3 lines · Johannes Sixt
Quoted 1 line

+ perl -e 'print reverse <>' | \

[...]

git-rebase--interactive.sh +1 -0
+	sed -ne '1!G;$p;h' | \

Wow, and people complain about perl being unreadable. ;)

-Peff

#5 lore raw

Re: [PATCH] rebase--interactive: Replace unportable 'tac' by a sed script.

Quoted 8 lines · Jeff King

git-rebase--interactive.sh create_extended_todo_list () { · line 472

Quoted 3 lines · Johannes Sixt
Quoted 1 line

+ perl -e 'print reverse <>' | \

[...]

git-rebase--interactive.sh +1 -0
+	sed -ne '1!G;$p;h' | \

Wow, and people complain about perl being unreadable. ;)

Well, the tmtowtdi dogma has its drawbacks. Here's another sed-script that does the exact same thing:

	sed '1!G;h;$!d'

That being said, perl borrows most of its regex notation from sed, so it's not as if perl is easier. This is just the top-end of the scale that things like "sed s/foo/bar/" is at the bottom of ;-)

Signature · 3 lines
Andreas Ericsson                   [email protected]
OP5 AB                             www.op5.se
Tel: +46 8-230225                  Fax: +46 8-230231
#6 lore raw

Re: [PATCH] rebase--interactive: Replace unportable 'tac' by a sed script.

In reply to Johannes Sixt

git-rebase--interactive.sh create_extended_todo_list () { · line 472

Quoted 12 lines · Johannes Sixt

From: Johannes Sixt <[email protected]>

---

Review

Quoted 2 lines · Brian Gernhardt
(quoted diff) +1 -1
-    tac | \
+    perl -e 'print reverse <>' | \

Here's my try, which avoids the perl hammer. ;)
...

git-rebase--interactive.sh +1 -1
-	tac | \
+	sed -ne '1!G;$p;h' | \

Thanks for trying, but frankly, I'd prefer the Perl hammer, as any advanced sed scripting tend to be far less portable.

#7 lore raw

Re: [PATCH] rebase--interactive: Replace unportable 'tac' by a sed script.

Quoted 8 lines

git-rebase--interactive.sh create_extended_todo_list () { · line 472

Quoted 5 lines

Here's my try, which avoids the perl hammer. ;)
...

git-rebase--interactive.sh +1 -1
-	tac | \
+	sed -ne '1!G;$p;h' | \

Thanks for trying, but frankly, I'd prefer the Perl hammer, as any advanced sed scripting tend to be far less portable.

In fact this script is not portable. :-( Not so much because of its "advanced-ness", but because it loads the entire file in memory, and thus hits the limit on the buffer length of many seds.

Paolo

#8 lore raw

Re: [PATCH] Use perl instead of tac

In reply to Brian Gernhardt

Hi,

git-rebase--interactive.sh create_extended_todo_list () { · line 473

Quoted 11 lines · Brian Gernhardt
git-rebase--interactive.sh +1 -1
index 1751b08..303b754 100755
@@ -469,7 +469,7 @@ create_extended_todo_list () {
  	test -n "${last_parent:-}" -a "${last_parent:-}" != $SHORTUPSTREAM  
 && \
  		echo reset $last_parent
  	) | \
-	tac | \
+	perl -e 'print reverse <>' | \

What about using a shell function and a *big* variable or an intermediate file?

tac()
{
while read line
do
reversed="$line
$reversed"

	done
	echo "${reversed%
}"

}

Bye, Jörg.

Signature · 3 lines
Ich halte ihn zwar für einen Schurken und das was er sagt für
falsch – aber ich bin bereit mein Leben dafür einzusetzen, daß
er seine Meinung sagen kann.		(Voletair)
#9 lore raw

Re: [PATCH] Use perl instead of tac

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Jörg" == Jörg Sommer <[email protected]> writes:

Jörg> What about using a shell function and a *big* variable or an intermediate Jörg> file?

What makes you think that's any more faster or efficient than calling Perl at this point?

Signature · 4 lines
Randal L. Schwartz - Stonehenge Consulting Services, Inc. - +1 503 777 0095
<[email protected]> <URL:http://www.stonehenge.com/merlyn/>
Perl/Unix/security consulting, Technical writing, Comedy, etc. etc.
See PerlTraining.Stonehenge.com for onsite and open-enrollment Perl training!
#10 lore raw

Re: [PATCH] Use perl instead of tac

Quoted 7 lines · Randal L. Schwartz
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Jörg" == Jörg Sommer <[email protected]> writes:

Jörg> What about using a shell function and a *big* variable or an intermediate Jörg> file?

What makes you think that's any more faster or efficient than calling Perl at this point?

I doubt Jörg suggested it for its speed, but it removes the dependency on Perl.

Bit ugly, still.

Dave.

#11 lore raw

Re: [PATCH] Use perl instead of tac

Quoted 6 lines · David Symonds
Quoted 4 lines · Randal L. Schwartz
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Jörg" == Jörg Sommer <[email protected]> writes:

Jörg> What about using a shell function and a *big* variable or an intermediate Jörg> file?

Quoted 6 lines
Quoted 2 lines

What makes you think that's any more faster or efficient than calling Perl at this point?

I doubt Jörg suggested it for its speed, but it removes the dependency on Perl.

Bit ugly, still.

And quadratic, unlike Perl (and unlike sed, if it worked portably).

Paolo

#12 lore raw

Re: [PATCH] Use perl instead of tac

In reply to David Symonds
Quoted 2 lines · David Symonds

I doubt Jörg suggested it for its speed, but it removes the dependency on Perl.

We already depend on perl for git-add--interactive.perl in core git, and for git-svn, git-cvs*, and git-send-email. I don't think this one line is going to make a big difference in our Perl dependency.

~~ Brian

#13 lore raw

Re: [PATCH] Use perl instead of tac

Hi,

Quoted 8 lines · Brian Gernhardt
Quoted 2 lines · David Symonds

I doubt Jörg suggested it for its speed, but it removes the dependency on Perl.

We already depend on perl for git-add--interactive.perl in core git, and for git-svn, git-cvs*, and git-send-email. I don't think this one line is going to make a big difference in our Perl dependency.

It does. It goes the wrong way. And work on building in git-add--interactive.perl has already begun, so that is not a strong argument.

As for git-svn and git-cvs* I agree, they will probably remain perl scripts. But I think that they are not that important, since they are not strictly _core_ parts of Git, being importers/exporters that not everybody needs.

Ciao,
Dscho

#14 lore raw

Re: [PATCH] Use perl instead of tac

In reply to Randal L. Schwartz

Hallo Randal,

Quoted 7 lines · Randal L. Schwartz
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Jörg" == Jörg Sommer <[email protected]> writes:

Jörg> What about using a shell function and a *big* variable or an intermediate Jörg> file?

What makes you think that's any more faster or efficient than calling Perl at this point?

Nothing. My intention was not the speed, but the dependency on Perl. But your are right. Except from the point that my suggestion is broken, it's *much* slower:

% time dash -c 'while IFS= read -r line; do rev="$line
$rev"; done; printf "%s" "$rev"' < gitk-git/gitk-wish DN
dash -c < gitk-git/gitk-wish > /dev/null 2> /dev/null 17,89s user 2,68s system 96% cpu 21,352 total

% time tac < gitk-git/gitk-wish DN
tac < gitk-git/gitk-wish > /dev/null 2> /dev/null 0,01s user 0,01s system 91% cpu 0,017 total

% time perl -e 'print reverse <>' < gitk-git/gitk-wish DN
perl -e 'print reverse <>' < gitk-git/gitk-wish > /dev/null 2> /dev/null 0,07s user 0,01s system 59% cpu 0,141 total

But I doubt this hurts, because we don't have such a big input.

% wc -l git.c
390 git.c

% time dash -c 'while IFS= read -r line; do rev="$line
$rev"; done; printf "%s" "$rev"' < git.c DN
dash -c < git.c > /dev/null 2> /dev/null 0,04s user 0,04s system 62% cpu 0,115 total

And what about something like this: 'tac || rev || perl …'

Bye, Jörg.

Signature · 3 lines
Was der Bauer nicht kennt, das frisst er nicht. Würde der Städter kennen,
was er frisst, er würde umgehend Bauer werden.
                                                       Oliver Hassencamp
#15 lore raw

Re: [PATCH] Use perl instead of tac

Quoted 5 lines · Jörg Sommer

Nothing. My intention was not the speed, but the dependency on Perl. But
your are right. Except from the point that my suggestion is broken, it's
*much* slower:

[snip]

Quoted 1 line

And what about something like this: 'tac || rev || perl …'

This would actually be spelled something like:

----- 8< -----
tac=""

tac < /dev/null > /dev/null 2>&1
if test $? != 127; then
tac=tac
fi

if test -z "$tac"; then

    perl < /dev/null > /dev/null 2>&1
    if test $? != 127; then
       tac="perl -e 'print reverse <>'"
    fi

fi

if test -z "$tac"; then
die "Couldn't find tac or perl."
fi

(subshell) | $tac | while loop
----- 8< -----

Ugly, no? Plus it adds a dependency on tac OR doesn't solve the perl dependency. I personally think using perl is better than this approach. Even if we're trying to reduce the perl dependency in core, it's there right now so this has zero impact now. The ideal solution is to re-write the loop so that it doesn't need to be reversed in the first place. We can use perl until that rewrite is done.

This loop-reverse-loop construct is _extremely_ ugly, and I'm having problems following it. None of what this function is doing is immediately obvious. It's a good candidate for a comment or two.

Looking over it, we should be able to do this with one loop over the list of commits, doing this for each of them:

if more than one parent
other_parents=all but first, comma separated
print merge command (without marks)
else
print pick command
fi

if commit is tagged
print tag command
fi

I also dislike the large lists this is carrying around in shell variables. If I'm reading it correctly, the tag list could be replaced by invocations of "git describe --exact-match". The mark list appears to be unavoidable, but significantly smaller than the tag list.

Now that I think about it, the generation of marks could be done by a second loop over the list. Notice what commits need to be marked in loop 1, then add the mark commands in loop 2. Both of these loops would function in the same direction, removing the need for either tac or perl.

~~ Brian

#16 lore raw

Re: [PATCH] Use perl instead of tac

Hi Brian,

Brian Gernhardt schrieb am Mon 28. Apr, 15:13 (-0400):

Quoted 3 lines

This loop-reverse-loop construct is _extremely_ ugly, and I'm having problems following it. None of what this function is doing is immediately obvious. It's a good candidate for a comment or two.

I write one.

Quoted 3 lines

I also dislike the large lists this is carrying around in shell variables. If I'm reading it correctly, the tag list could be replaced by invocations of "git describe --exact-match".

Yes. How to get all tags of a commit?

% git tag foo v1.5.5
% git describe --exact-match 9d831805195ba40b62f632acc6bb6e53d3
warning: tag 'v1.5.5' is really 'foo' here
v1.5.5

Quoted 4 lines

Now that I think about it, the generation of marks could be done by a second loop over the list. Notice what commits need to be marked in loop 1, then add the mark commands in loop 2. Both of these loops would function in the same direction, removing the need for either tac or perl.

You are right.

Thanks for your comments.

Bye, Jörg.

Signature · 2 lines
Damit das Mögliche entsteht, muß immer wieder das Unmögliche versucht
werden.                                       (Hermann Hesse)
#17 lore raw

Re: [PATCH] Use perl instead of tac

Hallo Brian,

Jörg Sommer schrieb am Wed 30. Apr, 11:02 (+0200):

Quoted 7 lines

Brian Gernhardt schrieb am Mon 28. Apr, 15:13 (-0400):

Quoted 4 lines

Now that I think about it, the generation of marks could be done by a second loop over the list. Notice what commits need to be marked in loop 1, then add the mark commands in loop 2. Both of these loops would function in the same direction, removing the need for either tac or perl.

You are right.

… but I found the problem. I don't know how to pass the list from one subshell to the other.

% { echo pick; echo merge; a=12 } | { echo +$a+; while read line; do echo $line; done; } ++
pick
merge

The only idea I have is a file. Do you have a better idea?

% { echo pick; echo merge; a=12 } >/tmp/tmp; \
{ echo +$a+; while read line; do echo $line; done; } < /tmp/tmp
+12+
pick
merge

Bye, Jörg.

Signature · 2 lines
Viele Leute glauben, dass sie denken, wenn sie lediglich
ihre Vorurteile neu ordnen.
#18 lore raw

Re: [PATCH] Use perl instead of tac

Quoted 9 lines · Jörg Sommer

… but I found the problem. I don't know how to pass the list from one subshell to the other.

% { echo pick; echo merge; a=12 } | { echo +$a+; while read line; do echo $line; done; }
++
pick
merge

Quoted 7 lines

The only idea I have is a file. Do you have a better idea?

% { echo pick; echo merge; a=12 } >/tmp/tmp; \
{ echo +$a+; while read line; do echo $line; done; } < /tmp/tmp
+12+
pick
merge

No, I don't. But I think the resulting code will be easier to read, even with the file.

~~ Brian

#19 lore raw

Re: [PATCH] Use perl instead of tac

In reply to Jörg Sommer
Quoted 11 lines · Jörg Sommer
Quoted 4 lines

I also dislike the large lists this is carrying around in shell variables. If I'm reading it correctly, the tag list could be replaced
by invocations of "git describe --exact-match".

Yes. How to get all tags of a commit?

% git tag foo v1.5.5
% git describe --exact-match 9d831805195ba40b62f632acc6bb6e53d3
warning: tag 'v1.5.5' is really 'foo' here
v1.5.5

I wish I could be clever and say I pointed this out as an obviously wrong answer or similar. But, no, I simply didn't think of that. The long list may be required, despite my concerns about it.. :-(

Those concerns being: overrunning the length of a shell variable, the speed of constructing and searching the list, over-complexity of the code. But, of course, if there isn't another way to do it right then the list stays.

~~ Brian

#20 lore raw

Re: [PATCH] Use perl instead of tac

Hallo Brian,

Brian Gernhardt schrieb am Wed 30. Apr, 11:25 (-0400):

Quoted 7 lines · Jörg Sommer
Quoted 6 lines
Quoted 5 lines

I also dislike the large lists this is carrying around in shell variables. If I'm reading it correctly, the tag list could be replaced
by invocations of "git describe --exact-match".

Quoted 6 lines
Quoted 6 lines

Yes. How to get all tags of a commit?

% git tag foo v1.5.5
% git describe --exact-match 9d831805195ba40b62f632acc6bb6e53d3
warning: tag 'v1.5.5' is really 'foo' here
v1.5.5

And how can I get only tags no annotated tags? We can't recreate annotated tags with git rebase.

Quoted 1 line

Those concerns being: overrunning the length of a shell variable,

Are you shure there is any such bounding? I didn't saw anything about the size of a variables in IEEE 1003.1-2004, but didn't look very carfully. All my shells (bash, dash, zsh) can handle more than 16777217 characters:

dash -c 'a=a; while :; do a=$a$a; echo $a | wc -c; done'

This would be a new version of create_extended_todo_list() without tac. What do you think about it? Is it better readable? But there's a bug.

create_extended_todo_list () {
	# The idea of this function is to
	# 1. build a list of tags
	# 2. go through the list and
	#    * issue a reset command to the parent of the commit, if the last
	#      commit was not the parent of the current commit,
	#    * issue a pick command for simple commits,
	#    * issue for each merge commit a merge command with the hashs of
	#      the parent commits,
	#    * register each parent of a merge and issue a mark command
	#      (without an ID) after the commit for each registered commit and
	#    * issue a tag command, if the commit is in the tag list.
	# 3. Then go through the created list and
	#    * add an ID to each mark command and
	#    * replace all occurences of the hash in reset and merge commands
	#      by the mark ID
	test -e "$DOTEST"/cetl.tmp \
		&& die "Someone else uses our filename cetl.tmp." \
			"That's not nice"
	if test t = "${PRESERVE_TAGS:-}"
	then
		tag_list=$(git show-ref --abbrev=7 --tags | \
			(
			while read sha1 tag
			do
				tag=${tag#refs/tags/}
				if test ${last_sha1:-0000} = $sha1
				then
					saved_tags="$saved_tags:$tag"
				else
					printf "%s" "${last_sha1:+ $last_sha1#$saved_tags}"
					last_sha1=$sha1
					saved_tags=$tag
				fi
			done
			echo "${last_sha1:+ $last_sha1:$saved_tags}"
			) )
	else
		tag_list=
	fi
	mark_these_commits=
	while IFS=_ read commit parents subject
	do
		first_parent=${parents%% *}
		if test "${last_commit:-$SHORTUPSTREAM}" != $first_parent
		then
			test "$first_parent" = $SHORTUPSTREAM &&
				first_parent=$SHORTONTO
			echo reset $first_parent
		fi
		unset first_parent
		last_commit=$commit

		case "$parents" in
		*' '*)
			new_parents=
			for p in $parents
			do
				mark_these_commits=$(insert_value_at_key_into_list \
					"$commit" "$p" "$mark_these_commits")
				if test "$p" = $SHORTUPSTREAM
				then
					new_parents="$new_parents $SHORTONTO"
				else
					new_parents="$new_parents $p"
				fi
			done
			unset p
			echo merge $commit ${new_parents# * }
			unset new_parents
			;;
		*)
			echo "pick $commit $subject"
			;;
		esac

		if tmp=$(get_value_from_list $commit "$tag_list")
		then
			for t in $(echo $tmp | tr : ' ')
			do
				echo tag $t
			done
		fi
	done > "$DOTEST"/cetl.tmp
	unset commit parents subject

	commit_mark_list=
	next_mark=0
	last_commit=
	while read cmd args
	do
		case "$cmd" in
		pick)
			this_commit="${args%% *}"
			;;
		reset)
			this_commit=$args
			if tmp=$(get_value_from_list $args "$commit_mark_list")
			then
				args=":$tmp"
			fi
			;;
		merge)
			new_args=
			for i in ${args#* }
			do
				if tmp=$(get_value_from_list $i \
					"$commit_mark_list")
				then
					new_args="$new_args :$tmp"
				else
					new_args="$new_args $i"
				fi
			done
			this_commit="${args%% *}"
			args="$this_commit ${new_args# }"
			;;
		esac

		if tmp=$(get_value_from_list "$last_commit" \
			"$mark_these_commits") && \
			test "${this_commit:-$last_commit}" != $tmp
		then
			if tmp=$(get_value_from_list "$last_commit" \
				"$commit_mark_list")
			then
				test "$last_cmd" = reset -o "$last_cmd" = tag \
					|| echo mark ":$tmp"
			else
				commit_mark_list=$(insert_value_at_key_into_list \
					$next_mark $last_commit "$commit_mark_list")
				echo mark ":$next_mark"
				next_mark=$(($next_mark + 1))
			fi
		fi
		last_commit=${this_commit:-$last_commit}
		unset this_commit

		echo "$cmd $args"
		last_cmd=$cmd
	done < "$DOTEST"/cetl.tmp
	rm "$DOTEST"/cetl.tmp
	unset last_cmd last_commit next_mark cmd args tmp commit_mark_list \
		mark_these_commits
}

The problem is the mark command. If you walk from ONTO to HEAD trough the list, you must know for a commit, if it is used _later_. But don't create a mark if it is used immediately, e.g. pick; merge not pick; mark; merge.

If you walk from HEAD to ONTO, this is much easier. I delayed the mark for the first head of a merge and checked if the next commit is this commit. This way I keep the todo list clean and don't get something like this for --first-parent:

pick abc
pick def
mark :0
merge 012 foreign-branch

Bye, Jörg.

Signature · 4 lines
Es ist außerdem ein weit verbreiteter Irrtum das USENET ‚helfen‘ soll.
Tatsächlich wurde USENET nachweislich zur persönlichen Belustigung
seiner Erfinder geschaffen.
Jörg Klemenz <[email protected]>, <[email protected]>
#21 lore raw

Re: [PATCH] Use perl instead of tac

Quoted 6 lines · Jörg Sommer

Hallo Brian,

Brian Gernhardt schrieb am Wed 30. Apr, 11:25 (-0400):

Quoted 3 lines · Jörg Sommer
Quoted 2 lines

Yes. How to get all tags of a commit?

Quoted 7 lines
Quoted 4 lines
Quoted 4 lines

% git tag foo v1.5.5
% git describe --exact-match 9d831805195ba40b62f632acc6bb6e53d3
warning: tag 'v1.5.5' is really 'foo' here
v1.5.5

And how can I get only tags no annotated tags? We can't recreate annotated tags with git rebase.

Yes, the tag list has to stay. I had missed several bits of what it was trying to do.

Quoted 10 lines
Quoted 1 line

Those concerns being: overrunning the length of a shell variable,

Are you shure there is any such bounding? I didn't saw anything about the
size of a variables in IEEE 1003.1-2004, but didn't look very carfully.
All my shells (bash, dash, zsh) can handle more than 16777217 characters:

dash -c 'a=a; while :; do a=$a$a; echo $a | wc -c; done'

My concern is that some shell somewhere (like the ever problematic Solaris) would have a limit and it would silently fail. It may be an unfounded concern, but I thought I'd mention it to the list and see if anyone agreed.

Quoted 3 lines

This would be a new version of create_extended_todo_list() without tac.
What do you think about it? Is it better readable? But there's a bug.

I find it more readable and am pleased about the lack of perl and tac.

Quoted 9 lines

create_extended_todo_list () {
# The idea of this function is to
# 1. build a list of tags
# 2. go through the list and
# * issue a reset command to the parent of the commit, if the last
# commit was not the parent of the current commit,
# * issue a pick command for simple commits,
# * issue for each merge commit a merge command with the hashs of
# the parent commits,

Quoted 8 lines

# * register each parent of a merge and issue a mark command
# (without an ID) after the commit for each registered commit
and
# * issue a tag command, if the commit is in the tag list.
# 3. Then go through the created list and
# * add an ID to each mark command and
# * replace all occurences of the hash in reset and merge commands
# by the mark ID

Comment is excellent, although I'd prefer that the three parts were simply next to the code they're referring to.

Quoted 3 lines
	test -e "$DOTEST"/cetl.tmp \
		&& die "Someone else uses our filename cetl.tmp." \
			"That's not nice"

Is this just to catch a bug where another part of rebase -i uses the same file? Or is there a way this could be triggered by user actions? (I am under the impression that we'd never get this far if $DOTEST already existed.)

Quoted 10 lines
	mark_these_commits=
	while IFS=_ read commit parents subject
	do
		first_parent=${parents%% *}
		if test "${last_commit:-$SHORTUPSTREAM}" != $first_parent
		then
			test "$first_parent" = $SHORTUPSTREAM &&
				first_parent=$SHORTONTO
			echo reset $first_parent
		fi
Quoted 8 lines
		unset first_parent
		last_commit=$commit

		case "$parents" in
		*' '*)
			new_parents=
			for p in $parents
			do

To avoid unneeded marks, you need one more variable and an if statement. Instead of unconditionally marking every parent, keep track of the previous commit and add a line like this:

                                if test "$p" != "$prev_commit"
                                then
Quoted 2 lines
				mark_these_commits=$(insert_value_at_key_into_list \
					"$commit" "$p" "$mark_these_commits")

fi

Quoted 25 lines
				if test "$p" = $SHORTUPSTREAM
				then
					new_parents="$new_parents $SHORTONTO"
				else
					new_parents="$new_parents $p"
				fi
			done
			unset p
			echo merge $commit ${new_parents# * }
			unset new_parents
			;;
		*)
			echo "pick $commit $subject"
			;;
		esac

		if tmp=$(get_value_from_list $commit "$tag_list")
		then
			for t in $(echo $tmp | tr : ' ')
			do
				echo tag $t
			done
		fi
Quoted 7 lines
	done > "$DOTEST"/cetl.tmp
	unset commit parents subject

	commit_mark_list=
	next_mark=0
	last_commit=
	while read cmd args

I'd personally do "cat cetl.tmp | while" instead of "while .... done < cetl.tmp", just so that where we're reading from is obvious from the start.

Quoted 13 lines

The problem is the mark command. If you walk from ONTO to HEAD trough the
list, you must know for a commit, if it is used _later_. But don't create
a mark if it is used immediately, e.g. pick; merge not pick; mark; merge.

If you walk from HEAD to ONTO, this is much easier. I delayed the mark for the first head of a merge and checked if the next commit is this commit. This way I keep the todo list clean and don't get something like
this for --first-parent:

Quoted 4 lines

pick abc
pick def
mark :0
merge 012 foreign-branch

(See above.)

~~ Brian

#22 lore raw

Re: [PATCH] Use perl instead of tac

In reply to Jörg Sommer
Quoted 1 line · Jörg Sommer

And what about something like this: 'tac || rev || perl …'

Or we could just add a git-tac.c and use that :)

Have fun,

Avery

#23 lore raw

Re: [PATCH] Use perl instead of tac

In reply to Jörg Sommer
Quoted 1 line · Jörg Sommer

And what about something like this: 'tac || rev || perl …'

Be careful: rev and tac are different. tac reverses lines, while rev reverses chars inside lines.

My 2 cents,

Signature · 1 line
Matthieu
#24 lore raw

Re: [PATCH] Use perl instead of tac

Quoted 6 lines · Matthieu Moy
Quoted 1 line · Jörg Sommer

And what about something like this: 'tac || rev || perl …'

Be careful: rev and tac are different. tac reverses lines, while rev reverses chars inside lines.

Indeed you are correct. This is a mistake I made in my original email, which people have been parroting. My mistake, sorry.

~~ Brian

#25 lore raw

Re: [PATCH] Use perl instead of tac

In reply to Jörg Sommer

Hi,

Quoted 15 lines · Jörg Sommer

git-rebase--interactive.sh create_extended_todo_list () { · line 473

Quoted 11 lines · Brian Gernhardt
git-rebase--interactive.sh +1 -1
index 1751b08..303b754 100755
@@ -469,7 +469,7 @@ create_extended_todo_list () {
  	test -n "${last_parent:-}" -a "${last_parent:-}" != $SHORTUPSTREAM  
 && \
  		echo reset $last_parent
  	) | \
-	tac | \
+	perl -e 'print reverse <>' | \

What about using a shell function and a *big* variable or an intermediate file?

How about fixing the code to not need tac instead? We went to great lengths to introduce the --reverse option to the rev-list command, in order to avoid tac in the original version of rebase -i.

Ciao,
Dscho

#26 lore raw

Re: [PATCH] Use perl instead of tac

Quoted 3 lines · Johannes Schindelin

How about fixing the code to not need tac instead? We went to great lengths to introduce the --reverse option to the rev-list command, in order to avoid tac in the original version of rebase -i.

Because this is reversing the output of a sub-shell (ll.395-471), not rev-list. And making the shell code insert commands before the lines the produce it would make the code complex at best. (And it's not simple code either.)

Or at least that's why I didn't do it. Someone who understands it better can feel free to make it work in reverse in the first place.

~~ Brian

#27 lore raw

Re: [PATCH] Use perl instead of tac

Hi,

Quoted 10 lines · Brian Gernhardt
Quoted 3 lines · Johannes Schindelin

How about fixing the code to not need tac instead? We went to great lengths to introduce the --reverse option to the rev-list command, in order to avoid tac in the original version of rebase -i.

Because this is reversing the output of a sub-shell (ll.395-471), not rev-list. And making the shell code insert commands before the lines the produce it would make the code complex at best. (And it's not simple code either.)

That's my _point_: the code is way too complex for what it tries to accomplish, namely implement a sane syntax for rebase -i -p.

Ciao,
Dscho

#28 lore raw
In reply to Brian Gernhardt
Quoted 3 lines · Brian Gernhardt

Commit d481bcc9: "Do rebase with preserve merges with advanced TODO list" uses the command tac, apparently to reverse the TODO command list. (I don't use rebase -i much, if you can't tell.)

That is very sloppy job on Junio's part. He usually is very picky about shell portability, and I remember that he earlier rewrote somebody else's commit that originally used tac command.

Signature · 6 lines
Nanako Shiraishi
http://ivory.ap.teacup.com/nanako3/

----------------------------------------------------------------------
Get a free email address with REAL anti-spam protection.
http://www.bluebottle.com/tag/1
#29 lore raw
Quoted 7 lines · しらいしななこ
Quoted 3 lines · Brian Gernhardt

Commit d481bcc9: "Do rebase with preserve merges with advanced TODO list" uses the command tac, apparently to reverse the TODO command list. (I don't use rebase -i much, if you can't tell.)

That is very sloppy job on Junio's part. He usually is very picky about shell portability, and I remember that he earlier rewrote somebody else's commit that originally used tac command.

Heh, sorry about sloppiness. I've been swamped lately outside git...