linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH -mm] Bisecting through -mm with quilt
@ 2005-09-23  0:32 Alexey Dobriyan
  2005-09-23  0:42 ` Andrew Morton
  0 siblings, 1 reply; 4+ messages in thread
From: Alexey Dobriyan @ 2005-09-23  0:32 UTC (permalink / raw)
  To: Andrew Morton; +Cc: linux-kernel

Quick, Dirty, Fragile, Should Work (TM).

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
---

 mm-bisect |   83 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 83 insertions(+)

diff -uprN linux-vanilla/mm-bisect linux-mm-bisect/mm-bisect
--- linux-vanilla/mm-bisect	1970-01-01 03:00:00.000000000 +0300
+++ linux-mm-bisect/mm-bisect	2005-09-23 04:25:26.000000000 +0400
@@ -0,0 +1,82 @@
+#!/bin/sh
+#
+# Bisecting through -mm.
+#
+# Assumptions
+# -----------
+# 1) X works
+# 2) X-mmY doesn't
+# 3) -mmY in broken-out form is in $QUILT_PATCHES, "quilt push -a"'d
+#
+# Usage cycle
+# -----------
+#	mm-bisect start
+#		...			<=== applying/reverting patches
+#	[recompile]
+#	[retest]
+# 	mm-bisect good			<=== if it works
+#		...
+#	[recompile]
+#	[retest]
+#	mm-bisect bad			<=== if it doesn't
+#		[...]
+#	Sucker is fix-typo.patch	<=== who to blame
+
+usage()
+{
+	echo >&2 'usage: mm-bisect [start | good | bad]'
+	exit 1
+}
+
+case "$#" in
+0)
+	usage
+	;;
+*)
+	CUR=$(quilt applied | wc -l)
+	case "$1" in
+	start)
+		echo 0 >.mm-bisect-good
+		echo $CUR >.mm-bisect-bad
+		;;
+	good)
+		if [ $(($(cat .mm-bisect-bad) - $CUR)) = 1 ]; then
+			quilt push -q >/dev/null
+			echo -n "Sucker is "
+			quilt top
+			rm -f .mm-bisect-bad .mm-bisect-good
+			exit 0
+		fi
+		echo $CUR >.mm-bisect-good
+		;;
+	bad)
+		if [ $(($CUR - $(cat .mm-bisect-good))) = 1 ]; then
+			echo -n "Sucker is "
+			quilt top
+			rm -f .mm-bisect-bad .mm-bisect-good
+			exit 0
+		fi
+		echo $CUR >.mm-bisect-bad
+		;;
+	*)
+		usage
+		;;
+	esac
+	GOOD=$(cat .mm-bisect-good)
+	BAD=$(cat .mm-bisect-bad)
+	MIDDLE=$((($BAD + $GOOD) / 2))
+	echo "[$GOOD .. $BAD] => $MIDDLE"
+	if [ $MIDDLE -lt $CUR ]; then
+		while [ $MIDDLE -lt $(quilt applied | wc -l) ]; do
+			quilt pop -q | sed -e "s/^Removing patch/-/" \
+						-e "/^Now at patch /d"
+		done
+	else
+		while [ $MIDDLE -gt $(quilt applied | wc -l) ]; do
+			quilt push -q | sed -e "s/^Applying patch/+/" \
+						-e "/^Now at patch /d"
+		done
+	fi
+	quilt top
+	echo "[$GOOD .. => $MIDDLE <= .. $BAD]"
+esac


^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH -mm] Bisecting through -mm with quilt
  2005-09-23  0:32 [PATCH -mm] Bisecting through -mm with quilt Alexey Dobriyan
@ 2005-09-23  0:42 ` Andrew Morton
  2005-09-23 15:20   ` [PATCH -mm v2] " Alexey Dobriyan
  0 siblings, 1 reply; 4+ messages in thread
From: Andrew Morton @ 2005-09-23  0:42 UTC (permalink / raw)
  To: Alexey Dobriyan; +Cc: linux-kernel

Alexey Dobriyan <adobriyan@gmail.com> wrote:
>
>  Quick, Dirty, Fragile, Should Work (TM).

hm, I wouldn't use it.  The problem is that a _lot_ of patches in -mm don't
fscking compile.

	bix:/usr/src/25> grep '[-]fix.patch' series | wc
	     72      72    2905

If your bisection happens to land you between foo.patch and foo-fix.patch,
you have a *known bad* kernel.   What's the point in testing it?

So I'd recommend the smarter approach: copy the series file to ~/hunt, edit
~/hunt and do the bisection by hand, marking the good and bad points in
~/hunt as you go.


^ permalink raw reply	[flat|nested] 4+ messages in thread

* [PATCH -mm v2] Bisecting through -mm with quilt
  2005-09-23  0:42 ` Andrew Morton
@ 2005-09-23 15:20   ` Alexey Dobriyan
  2005-09-23 15:33     ` Alexey Dobriyan
  0 siblings, 1 reply; 4+ messages in thread
From: Alexey Dobriyan @ 2005-09-23 15:20 UTC (permalink / raw)
  To: Andrew Morton; +Cc: linux-kernel

On Thu, Sep 22, 2005 at 05:42:50PM -0700, Andrew Morton wrote:
> Alexey Dobriyan <adobriyan@gmail.com> wrote:
> >  Quick, Dirty, Fragile, Should Work (TM).
> 
> hm, I wouldn't use it.  The problem is that a _lot_ of patches in -mm don't
> fscking compile.
> 
> 	bix:/usr/src/25> grep '[-]fix.patch' series | wc
> 	     72      72    2905
> 
> If your bisection happens to land you between foo.patch and foo-fix.patch,
> you have a *known bad* kernel.   What's the point in testing it?
> 
> So I'd recommend the smarter approach: copy the series file to ~/hunt, edit
> ~/hunt and do the bisection by hand, marking the good and bad points in
> ~/hunt as you go.

Done. If this isn't what you want, tell me so.
-----------------------------------------------------------------------
Marking patches as good or bad is separated from applying/reverting.
You can edit "series-bisect-mm" manually, then do "./bisect-mm apply"
and it will change tree to the new middle point.

Manual editing boils down to:
			    .--> "good fix-typo.patch"
	"fix-typo.patch" --<
			    `--> "bad fix-typo.patch"

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
---

 bisect-mm |  108 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 108 insertions(+)

--- a/bisect-mm
+++ b/bisect-mm
@@ -0,0 +1,108 @@
+#!/bin/sh
+#
+# Bisecting through -mm with quilt.
+#
+# Assumptions:
+#	* X + linus.patch works
+#	* X-mmY doesn't
+#	* broken-out -mmY is in $QUILT_PATCHES
+#
+# Usage:
+#	./bisect-mm start	mark linus.patch as good, last patch as bad
+#		|
+# +------>------+
+# |		|
+# |	[manually edit series-bisect-mm:		]
+# |	[mark known good patches as "good $GOOD.patch"	]
+# ^	[mark known bad patches as "bad $BAD.patch"	]
+# |		|
+# |	./bisect-mm apply	quilt will apply/revert stuff
+# |		|
+# +------<------+
+# |		|
+# |	     rebuild
+# |	      retest
+# ^		|
+# |	./bisect-mm {bad,good}
+# |		|
+# |		|
+# +------<------+---->	Sucker is fix-typo.patch
+
+usage()
+{
+	echo "usage: bisect-mm [start | good | bad | apply]" >&2
+	exit 1
+}
+
+SERIES=series-bisect-mm
+
+apply()
+{
+	quilt push -q | sed -e "s/^Applying patch/+/" -e "/^Now at patch/d"
+}
+
+revert()
+{
+	quilt pop -q | sed -e "s/^Removing patch/-/" -e "/^Now at patch/d"
+}
+
+case "$#" in
+0)
+	usage
+	;;
+*)
+	case "$1" in
+	start)
+		echo -n "good " >$SERIES
+		quilt series >>$SERIES
+		sed -i "s/$(tail -n1 $SERIES)/bad &/" $SERIES
+		echo "+ $SERIES"
+		;;
+	bad)
+		TOP=$(quilt top)
+		sed -i "s/.*$TOP/bad $TOP/" $SERIES
+		GOOD=$(cat -n $SERIES | grep "good " | tail -n1 | awk '{print $1}')
+		BAD=$(cat -n $SERIES | grep "bad " | head -n1 | awk '{print $1}')
+		if [ $(($BAD - $GOOD)) = 1 ]; then
+			echo -n "Sucker is "
+			quilt top
+			exit 0
+		fi
+		;;
+	good)
+		TOP=$(quilt top)
+		sed -i "s/.*$TOP/good $TOP/" $SERIES
+		GOOD=$(cat -n $SERIES | grep "good " | tail -n1 | awk '{print $1}')
+		BAD=$(cat -n $SERIES | grep "bad " | head -n1 | awk '{print $1}')
+		if [ $(($BAD - $GOOD)) = 1 ]; then
+			apply
+			echo -n "Sucker is "
+			quilt top
+			exit 0
+		fi
+		;;
+	apply)
+		GOOD=$(cat -n $SERIES | grep "good " | tail -n1 | awk '{print $1}')
+		BAD=$(cat -n $SERIES | grep "bad " | head -n1 | awk '{print $1}')
+		MIDDLE=$((($GOOD + $BAD) / 2))
+		TOP=$(quilt top)
+		CUR=$(quilt applied | wc -l)
+
+		if [ $MIDDLE -lt $CUR ]; then
+			while [ $MIDDLE -lt $(quilt applied | wc -l) ]; do
+				revert
+			done
+			quilt top
+		else
+			while [ $MIDDLE -gt $(quilt applied | wc -l) ]; do
+				apply
+			done
+		fi
+		echo "[$GOOD .. => $MIDDLE <= .. $BAD]"
+		;;
+	*)
+		usage
+		;;
+	esac
+	;;
+esac


^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH -mm v2] Bisecting through -mm with quilt
  2005-09-23 15:20   ` [PATCH -mm v2] " Alexey Dobriyan
@ 2005-09-23 15:33     ` Alexey Dobriyan
  0 siblings, 0 replies; 4+ messages in thread
From: Alexey Dobriyan @ 2005-09-23 15:33 UTC (permalink / raw)
  To: Andrew Morton; +Cc: linux-kernel

On Fri, Sep 23, 2005 at 07:20:25PM +0400, Alexey Dobriyan wrote:
> On Thu, Sep 22, 2005 at 05:42:50PM -0700, Andrew Morton wrote:
> > hm, I wouldn't use it.  The problem is that a _lot_ of patches in -mm don't
> > fscking compile.
> > 
> > 	bix:/usr/src/25> grep '[-]fix.patch' series | wc
> > 	     72      72    2905
> > 
> > If your bisection happens to land you between foo.patch and foo-fix.patch,
> > you have a *known bad* kernel.   What's the point in testing it?

If "./bisect-mm apply" landed you to obviously wrong kernel, with v2 you
can do a couple of "quilt {push,pop}" by hand. "./bisect {bad,good}"
marks _current_ patch as reported by "quilt top".


^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2005-09-23 15:23 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2005-09-23  0:32 [PATCH -mm] Bisecting through -mm with quilt Alexey Dobriyan
2005-09-23  0:42 ` Andrew Morton
2005-09-23 15:20   ` [PATCH -mm v2] " Alexey Dobriyan
2005-09-23 15:33     ` Alexey Dobriyan

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).