[v3] setlocalversion: work around "git describe" performance

Commit Message

Rasmus Villemoes Nov. 18, 2024, 11:01 a.m. UTC

Contrary to expectations, passing a single candidate tag to "git
describe" is slower than not passing any --match options.

  $ time git describe --debug
  ...
  traversed 10619 commits
  ...
  v6.12-rc5-63-g0fc810ae3ae1

  real    0m0.169s

  $ time git describe --match=v6.12-rc5 --debug
  ...
  traversed 1310024 commits
  v6.12-rc5-63-g0fc810ae3ae1

  real    0m1.281s

In fact, the --debug output shows that git traverses all or most of
history. For some repositories and/or git versions, those 1.3s are
actually 10-15 seconds.

This has been acknowledged as a performance bug in git [1], and a fix
is on its way [2]. However, no solution is yet in git.git, and even
when one lands, it will take quite a while before it finds its way to
a release and for $random_kernel_developer to pick that up.

So rewrite the logic to use plumbing commands. For each of the
candidate values of $tag, we ask: (1) is $tag even an annotated
tag? (2) Is it eligible to describe HEAD, i.e. an ancestor of
HEAD? (3) If so, how many commits are in $tag..HEAD?

I have tested that this produces the same output as the current script
for ~700 random commits between v6.9..v6.10. For those 700 commits,
and in my git repo, the 'make -s kernelrelease' command is on average
~4 times faster with this patch applied (geometric mean of ratios).

For the commit mentioned in Josh's original report [3], the
time-consuming part of setlocalversion goes from

$ time git describe --match=v6.12-rc5 c1e939a21eb1
v6.12-rc5-44-gc1e939a21eb1

real    0m1.210s

to

$ time git rev-list --count --left-right v6.12-rc5..c1e939a21eb1
0       44

real    0m0.037s

[1] https://lore.kernel.org/git/[e-mail]/
[2] https://lore.kernel.org/git/[e-mail]/
[3] https://lore.kernel.org/lkml/[e-mail]/

Reported-by: Sean Christopherson <[e-mail]>
Closes: https://lore.kernel.org/lkml/[e-mail]/
Reported-by: Josh Poimboeuf <[e-mail]>
Tested-by: Josh Poimboeuf <[e-mail]>
Closes: https://lore.kernel.org/lkml/[e-mail]/
Signed-off-by: Rasmus Villemoes <[e-mail]>
---
v3:

- Update trailer tag list, per Masahiro.
- Drop redundant quoutes around the word tag
- Add a shellcheck disable directive

Masahiro, I decided to keep the changes minimal, in particular not to
change anything around the logic or the (unused) return values, in
order not to invalidate Josh' T-b tag. I think it's more important for
this to make it to 6.13-rc1 (if that is even still possible, given
that the MW is already open).

Since you mentioned shellcheck, which I myself am a big fan of, I
added a directive in the new code, but didn't want to try to make the
whole script shellcheck clean. But I don't know if we want such linter
directives; if not, feel free to just remove that line.

v2: https://lore.kernel.org/lkml/[e-mail]/


 scripts/setlocalversion | 54 +++++++++++++++++++++++++++++------------
 1 file changed, 38 insertions(+), 16 deletions(-)

Comments

Masahiro Yamada Nov. 18, 2024, 11:19 p.m. UTC | #1

On Mon, Nov 18, 2024 at 8:01 PM Rasmus Villemoes
<[e-mail]> wrote:
>
> Contrary to expectations, passing a single candidate tag to "git
> describe" is slower than not passing any --match options.
>
>   $ time git describe --debug
>   ...
>   traversed 10619 commits
>   ...
>   v6.12-rc5-63-g0fc810ae3ae1
>
>   real    0m0.169s
>
>   $ time git describe --match=v6.12-rc5 --debug
>   ...
>   traversed 1310024 commits
>   v6.12-rc5-63-g0fc810ae3ae1
>
>   real    0m1.281s
>
> In fact, the --debug output shows that git traverses all or most of
> history. For some repositories and/or git versions, those 1.3s are
> actually 10-15 seconds.
>
> This has been acknowledged as a performance bug in git [1], and a fix
> is on its way [2]. However, no solution is yet in git.git, and even
> when one lands, it will take quite a while before it finds its way to
> a release and for $random_kernel_developer to pick that up.
>
> So rewrite the logic to use plumbing commands. For each of the
> candidate values of $tag, we ask: (1) is $tag even an annotated
> tag? (2) Is it eligible to describe HEAD, i.e. an ancestor of
> HEAD? (3) If so, how many commits are in $tag..HEAD?
>
> I have tested that this produces the same output as the current script
> for ~700 random commits between v6.9..v6.10. For those 700 commits,
> and in my git repo, the 'make -s kernelrelease' command is on average
> ~4 times faster with this patch applied (geometric mean of ratios).
>
> For the commit mentioned in Josh's original report [3], the
> time-consuming part of setlocalversion goes from
>
> $ time git describe --match=v6.12-rc5 c1e939a21eb1
> v6.12-rc5-44-gc1e939a21eb1
>
> real    0m1.210s
>
> to
>
> $ time git rev-list --count --left-right v6.12-rc5..c1e939a21eb1
> 0       44
>
> real    0m0.037s
>
> [1] https://lore.kernel.org/git/[e-mail]/
> [2] https://lore.kernel.org/git/[e-mail]/
> [3] https://lore.kernel.org/lkml/[e-mail]/
>
> Reported-by: Sean Christopherson <[e-mail]>
> Closes: https://lore.kernel.org/lkml/[e-mail]/
> Reported-by: Josh Poimboeuf <[e-mail]>
> Tested-by: Josh Poimboeuf <[e-mail]>
> Closes: https://lore.kernel.org/lkml/[e-mail]/
> Signed-off-by: Rasmus Villemoes <[e-mail]>
> ---
> v3:
>
> - Update trailer tag list, per Masahiro.
> - Drop redundant quoutes around the word tag
> - Add a shellcheck disable directive
>
> Masahiro, I decided to keep the changes minimal, in particular not to
> change anything around the logic or the (unused) return values, in
> order not to invalidate Josh' T-b tag. I think it's more important for
> this to make it to 6.13-rc1 (if that is even still possible, given
> that the MW is already open).


This is not urgent because it has been broken more than one year.

Your "|| return 1" may not live long.
https://lore.kernel.org/linux-kbuild/[e-mail]/T/#u


If you write try_tag() like you wrote, my patch can become even simpler.




> +
> +       # $2 is the number of commits in the range $tag..HEAD, possibly 0.
> +       count="$2"

count=$2

is enough because double-quotes are not required on the RHS
of an assignment.




> -       # If we are at the tagged commit, we ignore it because the version is
> -       # well-defined.
> -       if [ "${tag}" != "${desc}" ]; then
> +       # If we are at the tagged commit, we ignore it because the
> +       # version is well-defined. If none of the attempted tags exist
> +       # or were usable, $count is still empty.
> +       if [ -z "${count}" ] || [ "${count}" -gt 0 ]; then

Is this code equivalent to the following?

         if [ "${count}" != 0 ]; then

Masahiro Yamada Nov. 26, 2024, 1:10 p.m. UTC | #2

On Mon, Nov 18, 2024 at 8:01 PM Rasmus Villemoes
<[e-mail]> wrote:
>
> Contrary to expectations, passing a single candidate tag to "git
> describe" is slower than not passing any --match options.
>
>   $ time git describe --debug
>   ...
>   traversed 10619 commits
>   ...
>   v6.12-rc5-63-g0fc810ae3ae1
>
>   real    0m0.169s
>
>   $ time git describe --match=v6.12-rc5 --debug
>   ...
>   traversed 1310024 commits
>   v6.12-rc5-63-g0fc810ae3ae1
>
>   real    0m1.281s
>
> In fact, the --debug output shows that git traverses all or most of
> history. For some repositories and/or git versions, those 1.3s are
> actually 10-15 seconds.
>
> This has been acknowledged as a performance bug in git [1], and a fix
> is on its way [2]. However, no solution is yet in git.git, and even
> when one lands, it will take quite a while before it finds its way to
> a release and for $random_kernel_developer to pick that up.
>
> So rewrite the logic to use plumbing commands. For each of the
> candidate values of $tag, we ask: (1) is $tag even an annotated
> tag? (2) Is it eligible to describe HEAD, i.e. an ancestor of
> HEAD? (3) If so, how many commits are in $tag..HEAD?
>
> I have tested that this produces the same output as the current script
> for ~700 random commits between v6.9..v6.10. For those 700 commits,
> and in my git repo, the 'make -s kernelrelease' command is on average
> ~4 times faster with this patch applied (geometric mean of ratios).
>
> For the commit mentioned in Josh's original report [3], the
> time-consuming part of setlocalversion goes from
>
> $ time git describe --match=v6.12-rc5 c1e939a21eb1
> v6.12-rc5-44-gc1e939a21eb1
>
> real    0m1.210s
>
> to
>
> $ time git rev-list --count --left-right v6.12-rc5..c1e939a21eb1
> 0       44
>
> real    0m0.037s
>
> [1] https://lore.kernel.org/git/[e-mail]/
> [2] https://lore.kernel.org/git/[e-mail]/
> [3] https://lore.kernel.org/lkml/[e-mail]/
>
> Reported-by: Sean Christopherson <[e-mail]>
> Closes: https://lore.kernel.org/lkml/[e-mail]/
> Reported-by: Josh Poimboeuf <[e-mail]>
> Tested-by: Josh Poimboeuf <[e-mail]>
> Closes: https://lore.kernel.org/lkml/[e-mail]/
> Signed-off-by: Rasmus Villemoes <[e-mail]>
> ---
> v3:

Applied to linux-kbuild because this is better than v4 at least.

Thanks.

Patch

diff --git a/scripts/setlocalversion b/scripts/setlocalversion
index 38b96c6797f4..5818465abba9 100755
--- a/scripts/setlocalversion
+++ b/scripts/setlocalversion
@@ -30,6 +30,27 @@  if test $# -gt 0 -o ! -d "$srctree"; then
 	usage
 fi
 
+try_tag() {
+	tag="$1"
+
+	# Is $tag an annotated tag?
+	[ "$(git cat-file -t "$tag" 2> /dev/null)" = tag ] || return 1
+
+	# Is it an ancestor of HEAD, and if so, how many commits are in $tag..HEAD?
+	# shellcheck disable=SC2046 # word splitting is the point here
+	set -- $(git rev-list --count --left-right "$tag"...HEAD 2> /dev/null)
+
+	# $1 is 0 if and only if $tag is an ancestor of HEAD. Use
+	# string comparison, because $1 is empty if the 'git rev-list'
+	# command somehow failed.
+	[ "$1" = 0 ] || return 1
+
+	# $2 is the number of commits in the range $tag..HEAD, possibly 0.
+	count="$2"
+
+	return 0
+}
+
 scm_version()
 {
 	local short=false
@@ -61,33 +82,33 @@  scm_version()
 	# stable kernel:    6.1.7      ->  v6.1.7
 	version_tag=v$(echo "${KERNELVERSION}" | sed -E 's/^([0-9]+\.[0-9]+)\.0(.*)$/\1\2/')
 
+	# try_tag initializes count if the tag is usable.
+	count=
+
 	# If a localversion* file exists, and the corresponding
 	# annotated tag exists and is an ancestor of HEAD, use
 	# it. This is the case in linux-next.
-	tag=${file_localversion#-}
-	desc=
-	if [ -n "${tag}" ]; then
-		desc=$(git describe --match=$tag 2>/dev/null)
+	if [ -n "${file_localversion#-}" ] ; then
+		try_tag "${file_localversion#-}"
 	fi
 
 	# Otherwise, if a localversion* file exists, and the tag
 	# obtained by appending it to the tag derived from
 	# KERNELVERSION exists and is an ancestor of HEAD, use
 	# it. This is e.g. the case in linux-rt.
-	if [ -z "${desc}" ] && [ -n "${file_localversion}" ]; then
-		tag="${version_tag}${file_localversion}"
-		desc=$(git describe --match=$tag 2>/dev/null)
+	if [ -z "${count}" ] && [ -n "${file_localversion}" ]; then
+		try_tag "${version_tag}${file_localversion}"
 	fi
 
 	# Otherwise, default to the annotated tag derived from KERNELVERSION.
-	if [ -z "${desc}" ]; then
-		tag="${version_tag}"
-		desc=$(git describe --match=$tag 2>/dev/null)
+	if [ -z "${count}" ]; then
+		try_tag "${version_tag}"
 	fi
 
-	# If we are at the tagged commit, we ignore it because the version is
-	# well-defined.
-	if [ "${tag}" != "${desc}" ]; then
+	# If we are at the tagged commit, we ignore it because the
+	# version is well-defined. If none of the attempted tags exist
+	# or were usable, $count is still empty.
+	if [ -z "${count}" ] || [ "${count}" -gt 0 ]; then
 
 		# If only the short version is requested, don't bother
 		# running further git commands
@@ -95,14 +116,15 @@  scm_version()
 			echo "+"
 			return
 		fi
+
 		# If we are past the tagged commit, we pretty print it.
 		# (like 6.1.0-14595-g292a089d78d3)
-		if [ -n "${desc}" ]; then
-			echo "${desc}" | awk -F- '{printf("-%05d", $(NF-1))}'
+		if [ -n "${count}" ]; then
+			printf "%s%05d" "-" "${count}"
 		fi
 
 		# Add -g and exactly 12 hex chars.
-		printf '%s%s' -g "$(echo $head | cut -c1-12)"
+		printf '%s%.12s' -g "$head"
 	fi
 
 	if ${no_dirty}; then