[PATCH v3] deb-dl-dir class rework to use faster ln -P or fallback to cp

[PATCH v3] deb-dl-dir class rework to use faster ln -P or fallback to cp

[roberto.foglietta]

From: "Roberto A. Foglietta" <roberto.foglietta@gmail.com>

deb-dl-dir, feature: faster when using ln -P otherwise fallback to cp

The original class functions deb_dl_dir_import/export were using cp to
copy debian package to the target rootfs but this approach is quite slow
while using hard link does not work if the destination and source dirs
are not lying on the same filesystem. Thus, ln -P should fallback to cp
when it does not work (which is different from complaining on stderr).

Moreover, these two functions have been reworked to reach a straight
forward and more compact form. In particular, export function was using
bashism to do some kind of comparison which after all is useless
because copying back without overwriting just fulfills that part.

More rework using sudo in a different way plus a corner case
addressingi, in case the spia file exists for some other reasons.

Rebased on the current next

Signed-off-by: Roberto A. Foglietta <roberto.foglietta@gmail.com>
---
 meta/classes/deb-dl-dir.bbclass | 53 +++++++++++++++++++--------------
 1 file changed, 30 insertions(+), 23 deletions(-)

diff --git a/meta/classes/deb-dl-dir.bbclass b/meta/classes/deb-dl-dir.bbclass
index 7db25251..1fe9d40c 100644
--- a/meta/classes/deb-dl-dir.bbclass
+++ b/meta/classes/deb-dl-dir.bbclass
@@ -78,39 +78,46 @@ debsrc_download() {
 
 deb_dl_dir_import() {
     export pc="${DEBDIR}/${2}"
-    export rootfs="${1}"
-    sudo mkdir -p "${rootfs}"/var/cache/apt/archives/
+    export sc="${1}/var/cache/apt/archives/"
+    sudo mkdir -p "${sc}"
     [ ! -d "${pc}" ] && return 0
-    flock -s "${pc}".lock -c '
+    export tf=$(ls -1 "${pc}"/*.deb | head -n1)
+    [ ! -e "${tf}" ] && return 0
+    flock -Fs "${pc}".lock sudo -Es << 'EOFSUDO'
         set -e
         printenv | grep -q BB_VERBOSE_LOGS && set -x
 
-        sudo find "${pc}" -type f -iname "*\.deb" -exec \
-            ln -Pf -t "${rootfs}"/var/cache/apt/archives/ {} +
-    '
+        rm -f "${sc}/"$(basename "${tf}")
+        ln -Pf -t "${sc}" "${tf}" 2>/dev/null ||:
+        if [ -r "${tf}" ]; then
+            find "${pc}" -type f -iname "*\.deb" -exec \
+                ln -Pf -t "${sc}" {} +
+        else
+            find "${pc}" -type f -iname "*\.deb" -exec \
+                cp -np owner --reflink=auto -t "${sc}" {} +
+        fi
+EOFSUDO
 }
 
 deb_dl_dir_export() {
     export pc="${DEBDIR}/${2}"
-    export rootfs="${1}"
+    export sc="${1}/var/cache/apt/archives/"
     mkdir -p "${pc}"
-    flock "${pc}".lock -c '
+    export tf=$(ls -1 "${sc}"/*.deb | head -n1)
+    [ ! -e "${tf}" ] && return 0
+    flock -F "${pc}".lock sudo -Es << 'EOFSUDO'
         set -e
         printenv | grep -q BB_VERBOSE_LOGS && set -x
 
-        find "${rootfs}"/var/cache/apt/archives/ \
-            -maxdepth 1 -type f -iname '*\.deb' |\
-        while read p; do
-            # skip files from a previous export
-            [ -f "${pc}/${p##*/}" ] && continue
-            # can not reuse bitbake function here, this is basically
-            # "repo_contains_package"
-            package=$(find "${REPO_ISAR_DIR}"/"${DISTRO}" -name ${p##*/})
-            if [ -n "$package" ]; then
-                cmp --silent "$package" "$p" && continue
-            fi
-            sudo ln -Pf "${p}" "${pc}"
-        done
-        sudo chown -R $(id -u):$(id -g) "${pc}"
-    '
+        rm -f "${pc}/"$(basename "${tf}")
+        ln -Pf -t "${pc}" "${tf}" 2>/dev/null ||:
+        if [ -r "${tf}" ]; then
+            find "${sc}" -maxdepth 1 -type f -iname '*\.deb' \
+                -exec ln -P -t "${pc}" {} + 2>/dev/null ||:
+        else
+            find "${sc}" -maxdepth 1 -type f -iname '*\.deb' \
+                -exec cp -n --reflink=auto -t "${pc}" {} +
+        fi
+        chown -R $(id -u):$(id -g) "${pc}"
+EOFSUDO
 }
-- 
2.34.1

Re: [PATCH v3] deb-dl-dir class rework to use faster ln -P or fallback to cp

[Uladzimir Bely]

In the email from Tuesday, 7 February 2023 10:52:22 +03 user 
roberto.foglietta@linuxteam.org wrote:
> From: "Roberto A. Foglietta" <roberto.foglietta@gmail.com>
> 
> deb-dl-dir, feature: faster when using ln -P otherwise fallback to cp
> 
> The original class functions deb_dl_dir_import/export were using cp to
> copy debian package to the target rootfs but this approach is quite slow
> while using hard link does not work if the destination and source dirs
> are not lying on the same filesystem. Thus, ln -P should fallback to cp
> when it does not work (which is different from complaining on stderr).
> 
> Moreover, these two functions have been reworked to reach a straight
> forward and more compact form. In particular, export function was using
> bashism to do some kind of comparison which after all is useless
> because copying back without overwriting just fulfills that part.
> 
> More rework using sudo in a different way plus a corner case
> addressingi, in case the spia file exists for some other reasons.
> 
> Rebased on the current next
> 
> Signed-off-by: Roberto A. Foglietta <roberto.foglietta@gmail.com>
> ---
>  meta/classes/deb-dl-dir.bbclass | 53 +++++++++++++++++++--------------
>  1 file changed, 30 insertions(+), 23 deletions(-)
> 


> diff --git a/meta/classes/deb-dl-dir.bbclass
> b/meta/classes/deb-dl-dir.bbclass index 7db25251..1fe9d40c 100644
> --- a/meta/classes/deb-dl-dir.bbclass
> +++ b/meta/classes/deb-dl-dir.bbclass
> @@ -78,39 +78,46 @@ debsrc_download() {
> 
>  deb_dl_dir_import() {
>      export pc="${DEBDIR}/${2}"
> -    export rootfs="${1}"
> -    sudo mkdir -p "${rootfs}"/var/cache/apt/archives/
> +    export sc="${1}/var/cache/apt/archives/"
> +    sudo mkdir -p "${sc}"
>      [ ! -d "${pc}" ] && return 0
> -    flock -s "${pc}".lock -c '
> +    export tf=$(ls -1 "${pc}"/*.deb | head -n1)
> +    [ ! -e "${tf}" ] && return 0
> +    flock -Fs "${pc}".lock sudo -Es << 'EOFSUDO'
>          set -e
>          printenv | grep -q BB_VERBOSE_LOGS && set -x
> 
> -        sudo find "${pc}" -type f -iname "*\.deb" -exec \
> -            ln -Pf -t "${rootfs}"/var/cache/apt/archives/ {} +
> -    '
> +        rm -f "${sc}/"$(basename "${tf}")
> +        ln -Pf -t "${sc}" "${tf}" 2>/dev/null ||:
> +        if [ -r "${tf}" ]; then
> +            find "${pc}" -type f -iname "*\.deb" -exec \
> +                ln -Pf -t "${sc}" {} +
> +        else
> +            find "${pc}" -type f -iname "*\.deb" -exec \
> +                cp -np owner --reflink=auto -t "${sc}" {} +
> +        fi
> +EOFSUDO
>  }
> 
>  deb_dl_dir_export() {
>      export pc="${DEBDIR}/${2}"
> -    export rootfs="${1}"
> +    export sc="${1}/var/cache/apt/archives/"
>      mkdir -p "${pc}"
> -    flock "${pc}".lock -c '
> +    export tf=$(ls -1 "${sc}"/*.deb | head -n1)
> +    [ ! -e "${tf}" ] && return 0
> +    flock -F "${pc}".lock sudo -Es << 'EOFSUDO'
>          set -e
>          printenv | grep -q BB_VERBOSE_LOGS && set -x
> 
> -        find "${rootfs}"/var/cache/apt/archives/ \
> -            -maxdepth 1 -type f -iname '*\.deb' |\
> -        while read p; do
> -            # skip files from a previous export
> -            [ -f "${pc}/${p##*/}" ] && continue
> -            # can not reuse bitbake function here, this is basically
> -            # "repo_contains_package"
> -            package=$(find "${REPO_ISAR_DIR}"/"${DISTRO}" -name ${p##*/})
> -            if [ -n "$package" ]; then
> -                cmp --silent "$package" "$p" && continue
> -            fi
> -            sudo ln -Pf "${p}" "${pc}"
> -        done
> -        sudo chown -R $(id -u):$(id -g) "${pc}"
> -    '
> +        rm -f "${pc}/"$(basename "${tf}")
> +        ln -Pf -t "${pc}" "${tf}" 2>/dev/null ||:
> +        if [ -r "${tf}" ]; then
> +            find "${sc}" -maxdepth 1 -type f -iname '*\.deb' \
> +                -exec ln -P -t "${pc}" {} + 2>/dev/null ||:
> +        else
> +            find "${sc}" -maxdepth 1 -type f -iname '*\.deb' \
> +                -exec cp -n --reflink=auto -t "${pc}" {} +
> +        fi
> +        chown -R $(id -u):$(id -g) "${pc}"
> +EOFSUDO
>  }

I was almost ready to merge it since it passes CI (positive case), but 
patchset appeared don't work (negative case) on different filesystems - I find 
downloads/deb/*/ empty after build. At least, exports don't work.

See the example of my build for "negative" case: https://dpaste.org/
MAUKv#L1,22,23,36,37,38,39,50,52

You could also easy test it by creating "downloads.img" with ext4 inside and 
mounting it to "downloads" to make DL_DIR and BUILD_DIR be on different 
filesystems.

I still tend to apply my "[v2] deb-dl-dir: Fallback to copying when import-
export" instead. But it is worth reworking a bit, regarding sudo using in 
loops (e.g., would prefer "sudo / while <cmd>" Instead of "while / sudo 
<cmd>")...

Re: [PATCH v3] deb-dl-dir class rework to use faster ln -P or fallback to cp

[Roberto A. Foglietta]

On Fri, 10 Feb 2023 at 11:07, Uladzimir Bely <ubely@ilbers.de> wrote:

> >
> >  deb_dl_dir_import() {
> >      export pc="${DEBDIR}/${2}"
> > -    export rootfs="${1}"
> > -    sudo mkdir -p "${rootfs}"/var/cache/apt/archives/
> > +    export sc="${1}/var/cache/apt/archives/"
> > +    sudo mkdir -p "${sc}"
> >      [ ! -d "${pc}" ] && return 0
> > -    flock -s "${pc}".lock -c '
> > +    export tf=$(ls -1 "${pc}"/*.deb | head -n1)
> > +    [ ! -e "${tf}" ] && return 0
> > +    flock -Fs "${pc}".lock sudo -Es << 'EOFSUDO'
> >          set -e
> >          printenv | grep -q BB_VERBOSE_LOGS && set -x
> >
> > -        sudo find "${pc}" -type f -iname "*\.deb" -exec \
> > -            ln -Pf -t "${rootfs}"/var/cache/apt/archives/ {} +
> > -    '
> > +        rm -f "${sc}/"$(basename "${tf}")
> > +        ln -Pf -t "${sc}" "${tf}" 2>/dev/null ||:
> > +        if [ -r "${tf}" ]; then

Because we need to check the destination, not the source.

if exists $sc/$(basename $tf) then

>
> I was almost ready to merge it since it passes CI (positive case), but
> patchset appeared don't work (negative case) on different filesystems - I find
> downloads/deb/*/ empty after build. At least, exports don't work.

Both do not work correctly. I am going to fix it. Sorry, R-