[
https://issues.apache.org/jira/browse/CASSANDRA-21704?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Michael Semb Wever updated CASSANDRA-21704:
-------------------------------------------
Complexity: Low Hanging Fruit
Fix Version/s: 6.0.x
7.x
Test and Documentation Plan:
No JDK build ran; the change touches no Java and every site is shell, make or
rpm text.
{{.build/sh/test/stub-ant-cases.sh}} was run before and after the change, and
the same 19 of 64 cases fail either way. Those 19 fail on this machine for an
unrelated reason, {{run-tests.sh}} exiting 126, so the suite proves no
regression rather than proving the new code path.
The derivation was exercised directly against a {{build.xml}} renamed to
{{dse-db}}, in each of its three languages: the shell expression, the {{$(shell
…)}} in a makefile, and the classpath block of {{bin/cassandra.in.sh}} sourced
against a tree holding {{build/dse-db-5.1-SNAPSHOT.jar}},
{{build/dse-db-5.1-SNAPSHOT-sources.jar}} and {{build/dtest-5.1.jar}}. All
three answer {{dse-db}}, and the classpath block selects the one jar and
ignores the sources jar and the dtest jar.
Two negative cases were checked. With no {{build.xml}} present the classpath
block falls back to {{apache-cassandra}} and selects the same jar it selects
today. And {{ls -1 build/*-[0-9]*.jar}} on that same tree returns both
{{apache-cassandra-5.1-SNAPSHOT.jar}} and {{dtest-5.1.jar}}, which is why the
globs derive the name instead of widening.
The debian and redhat package builds need {{ant artifacts}} in the build
containers, so CI on this branch is what proves them.
Description:
The jar and the source tarball are named after the {{<project name="…">}}
attribute in {{build.xml}}, because ant builds
{{build/${ant.project.name}-${version}.jar}}. Nine sites in the build and
packaging scripts spell {{apache-cassandra}} instead:
{code:none}
.build/docker/_build-redhat.sh rm -rf ${DIST_DIR}/apache-cassandra-*.jar
cp ${DIST_DIR}/apache-cassandra-*-src.tar.gz …
.build/run-tests.sh [[ -f
"${DIST_DIR}/apache-cassandra-${version}.jar" ]] || …
.build/run-python-dtests.sh the same check
.build/sh/test/stub-ant-cases.sh touch
"${DIST_DIR}/apache-cassandra-${version}-SNAPSHOT.jar"
debian/rules dh_install
$(BUILD_DIR)/apache-cassandra-$(VERSION).jar
redhat/cassandra.spec %define relname
apache-cassandra-%{upstream_version}
bin/cassandra.in.sh ls -1
$CASSANDRA_HOME/build/apache-cassandra*.jar
tools/bin/cassandra.in.sh the same glob
redhat/cassandra.in.sh the same glob
{code}
Rename the project and every one of them misses. The debian and redhat
packaging cannot find the jar it just built, {{run-tests.sh}} reports {{Project
must be built first}} against a tree that is built, and {{bin/cassandra}}
constructs a classpath with no Cassandra jar on it.
The patch reads the name at each site, where that site already reads
{{build.xml}}: beside the {{base.version}} line in the two test runners, beside
{{buildxml_version}} in {{_build-redhat.sh}}, as a {{$(shell …)}} beside
{{VERSION}} in {{debian/rules}}, and from {{$CASSANDRA_HOME/build.xml}} in the
three {{cassandra.in.sh}} files, whose glob sits inside a {{[ -d
$CASSANDRA_HOME/build ]}} test and therefore runs against a source checkout.
One expression is used everywhere, {{grep '<project[[:space:]]*basedir='
build.xml | sed -ne 's/.*name="\([^"]*\)".*/\1/p'}}, with {{[[:space:]]}}
rather than {{\s}} because {{bin/cassandra.in.sh}} also runs under the BSD grep
on macOS.
{{redhat/cassandra.spec}} is the exception. It takes the name from
{{_build-redhat.sh}} as an {{rpmbuild --define}}, beside the {{version}} and
{{revision}} defines that call already passes, so the derivation stays in shell
and the spec carries no second copy of the regular expression. The spec
defaults to {{apache-cassandra}}, so {{rpmbuild -ba redhat/cassandra.spec}} run
by hand behaves as it does today.
Three limits are deliberate. The debian symlink keeps the name
{{apache-cassandra.jar}}: it is a published path, and only its target has to
follow the rename, so {{debian/init}} needs no change either. The classpath
globs derive the name rather than widening from {{apache-cassandra*.jar}} to
{{*-[0-9]*.jar}}, which would also match a built {{dtest-<version>.jar}} and
make the script refuse to start on "JAR artifacts for multiple versions". And
{{_build-redhat.sh}} alone refuses an empty name, because it is the one site
that hands the value to {{rm -rf}}.
Three sites are left alone: the sub-build names in {{.build/build-*.xml}},
which name ant projects rather than artifacts; the {{Conflicts}} and
{{Replaces}} on {{apache-cassandra1}} in {{debian/control}}, which name a
historical package; and {{toCopy}} in {{.jenkins/Jenkinsfile}}, which no code
in the tree reads.
Nothing changes for a build that keeps the name {{apache-cassandra}}, which is
every Apache build.
Patch:
[mck/CASSANDRA-21704/trunk|https://github.com/thelastpickle/cassandra/tree/mck/CASSANDRA-21704/trunk]
Provenance:
[2315d6f359|https://github.com/datastax/cassandra/commit/2315d6f359b963bb55c629c2416dc6f66903bf8f]
by [~mck], in
[datastax/cassandra#2193|https://github.com/datastax/cassandra/pull/2193],
where the fork's project name is {{dse-db}}. That pull request covers five of
these nine sites, and reaches the remaining fork needs by hardcoding {{dse-db}}
in the two {{cassandra.in.sh}} files and by widening the redhat glob. This
patch derives the name at all nine instead. The JDK 22 support and the docker
back-ports in that pull request are fork-only and are not here.
was:
The jar and the source tarball are named after the {{<project name="…">}}
attribute in {{build.xml}}, because ant builds
{{build/${ant.project.name}-${version}.jar}}. Nine sites in the build and
packaging scripts spell {{apache-cassandra}} instead:
{code:none}
.build/docker/_build-redhat.sh rm -rf ${DIST_DIR}/apache-cassandra-*.jar
cp ${DIST_DIR}/apache-cassandra-*-src.tar.gz …
.build/run-tests.sh [[ -f
"${DIST_DIR}/apache-cassandra-${version}.jar" ]] || …
.build/run-python-dtests.sh the same check
.build/sh/test/stub-ant-cases.sh touch
"${DIST_DIR}/apache-cassandra-${version}-SNAPSHOT.jar"
debian/rules dh_install
$(BUILD_DIR)/apache-cassandra-$(VERSION).jar
redhat/cassandra.spec %define relname
apache-cassandra-%{upstream_version}
bin/cassandra.in.sh ls -1
$CASSANDRA_HOME/build/apache-cassandra*.jar
tools/bin/cassandra.in.sh the same glob
redhat/cassandra.in.sh the same glob
{code}
Rename the project and every one of them misses. The debian and redhat
packaging cannot find the jar it just built, {{run-tests.sh}} reports {{Project
must be built first}} against a tree that is built, and {{bin/cassandra}}
constructs a classpath with no Cassandra jar on it.
The patch reads the name at each site, where that site already reads
{{build.xml}}: beside the {{base.version}} line in the two test runners, beside
{{buildxml_version}} in {{_build-redhat.sh}}, as a {{$(shell …)}} beside
{{VERSION}} in {{debian/rules}}, and from {{$CASSANDRA_HOME/build.xml}} in the
three {{cassandra.in.sh}} files, whose glob sits inside a {{[ -d
$CASSANDRA_HOME/build ]}} test and therefore runs against a source checkout.
One expression is used everywhere, {{grep '<project[[:space:]]*basedir='
build.xml | sed -ne 's/.*name="\([^"]*\)".*/\1/p'}}, with {{[[:space:]]}}
rather than {{\s}} because {{bin/cassandra.in.sh}} also runs under the BSD grep
on macOS.
{{redhat/cassandra.spec}} is the exception. It takes the name from
{{_build-redhat.sh}} as an {{rpmbuild --define}}, beside the {{version}} and
{{revision}} defines that call already passes, so the derivation stays in shell
and the spec carries no second copy of the regular expression. The spec
defaults to {{apache-cassandra}}, so {{rpmbuild -ba redhat/cassandra.spec}} run
by hand behaves as it does today.
Three limits are deliberate. The debian symlink keeps the name
{{apache-cassandra.jar}}: it is a published path, and only its target has to
follow the rename, so {{debian/init}} needs no change either. The classpath
globs derive the name rather than widening from {{apache-cassandra*.jar}} to
{{*-[0-9]*.jar}}, which would also match a built {{dtest-<version>.jar}} and
make the script refuse to start on "JAR artifacts for multiple versions". And
{{_build-redhat.sh}} alone refuses an empty name, because it is the one site
that hands the value to {{rm -rf}}.
Three sites are left alone: the sub-build names in {{.build/build-*.xml}},
which name ant projects rather than artifacts; the {{Conflicts}} and
{{Replaces}} on {{apache-cassandra1}} in {{debian/control}}, which name a
historical package; and {{toCopy}} in {{.jenkins/Jenkinsfile}}, which no code
in the tree reads.
Nothing changes for a build that keeps the name {{apache-cassandra}}, which is
every Apache build.
Patch:
[mck/upstream/packaging-project-name/trunk|https://github.com/thelastpickle/cassandra/tree/mck/upstream/packaging-project-name/trunk]
Provenance:
[2315d6f359|https://github.com/datastax/cassandra/commit/2315d6f359b963bb55c629c2416dc6f66903bf8f]
by [~mck], in
[datastax/cassandra#2193|https://github.com/datastax/cassandra/pull/2193],
where the fork's project name is {{dse-db}}. That pull request covers five of
these nine sites, and reaches the remaining fork needs by hardcoding {{dse-db}}
in the two {{cassandra.in.sh}} files and by widening the redhat glob. This
patch derives the name at all nine instead. The JDK 22 support and the docker
back-ports in that pull request are fork-only and are not here.
Priority: Low (was: Normal)
> Make the packaging and build scripts follow the ant project name
> ----------------------------------------------------------------
>
> Key: CASSANDRA-21704
> URL: https://issues.apache.org/jira/browse/CASSANDRA-21704
> Project: Apache Cassandra
> Issue Type: Improvement
> Components: Build, Packaging
> Reporter: Michael Semb Wever
> Priority: Low
> Fix For: 6.0.x, 7.x
>
>
> The jar and the source tarball are named after the {{<project name="…">}}
> attribute in {{build.xml}}, because ant builds
> {{build/${ant.project.name}-${version}.jar}}. Nine sites in the build and
> packaging scripts spell {{apache-cassandra}} instead:
> {code:none}
> .build/docker/_build-redhat.sh rm -rf ${DIST_DIR}/apache-cassandra-*.jar
> cp ${DIST_DIR}/apache-cassandra-*-src.tar.gz
> …
> .build/run-tests.sh [[ -f
> "${DIST_DIR}/apache-cassandra-${version}.jar" ]] || …
> .build/run-python-dtests.sh the same check
> .build/sh/test/stub-ant-cases.sh touch
> "${DIST_DIR}/apache-cassandra-${version}-SNAPSHOT.jar"
> debian/rules dh_install
> $(BUILD_DIR)/apache-cassandra-$(VERSION).jar
> redhat/cassandra.spec %define relname
> apache-cassandra-%{upstream_version}
> bin/cassandra.in.sh ls -1
> $CASSANDRA_HOME/build/apache-cassandra*.jar
> tools/bin/cassandra.in.sh the same glob
> redhat/cassandra.in.sh the same glob
> {code}
> Rename the project and every one of them misses. The debian and redhat
> packaging cannot find the jar it just built, {{run-tests.sh}} reports
> {{Project must be built first}} against a tree that is built, and
> {{bin/cassandra}} constructs a classpath with no Cassandra jar on it.
> The patch reads the name at each site, where that site already reads
> {{build.xml}}: beside the {{base.version}} line in the two test runners,
> beside {{buildxml_version}} in {{_build-redhat.sh}}, as a {{$(shell …)}}
> beside {{VERSION}} in {{debian/rules}}, and from
> {{$CASSANDRA_HOME/build.xml}} in the three {{cassandra.in.sh}} files, whose
> glob sits inside a {{[ -d $CASSANDRA_HOME/build ]}} test and therefore runs
> against a source checkout. One expression is used everywhere, {{grep
> '<project[[:space:]]*basedir=' build.xml | sed -ne
> 's/.*name="\([^"]*\)".*/\1/p'}}, with {{[[:space:]]}} rather than {{\s}}
> because {{bin/cassandra.in.sh}} also runs under the BSD grep on macOS.
> {{redhat/cassandra.spec}} is the exception. It takes the name from
> {{_build-redhat.sh}} as an {{rpmbuild --define}}, beside the {{version}} and
> {{revision}} defines that call already passes, so the derivation stays in
> shell and the spec carries no second copy of the regular expression. The spec
> defaults to {{apache-cassandra}}, so {{rpmbuild -ba redhat/cassandra.spec}}
> run by hand behaves as it does today.
> Three limits are deliberate. The debian symlink keeps the name
> {{apache-cassandra.jar}}: it is a published path, and only its target has to
> follow the rename, so {{debian/init}} needs no change either. The classpath
> globs derive the name rather than widening from {{apache-cassandra*.jar}} to
> {{*-[0-9]*.jar}}, which would also match a built {{dtest-<version>.jar}} and
> make the script refuse to start on "JAR artifacts for multiple versions". And
> {{_build-redhat.sh}} alone refuses an empty name, because it is the one site
> that hands the value to {{rm -rf}}.
> Three sites are left alone: the sub-build names in {{.build/build-*.xml}},
> which name ant projects rather than artifacts; the {{Conflicts}} and
> {{Replaces}} on {{apache-cassandra1}} in {{debian/control}}, which name a
> historical package; and {{toCopy}} in {{.jenkins/Jenkinsfile}}, which no code
> in the tree reads.
> Nothing changes for a build that keeps the name {{apache-cassandra}}, which
> is every Apache build.
> Patch:
> [mck/CASSANDRA-21704/trunk|https://github.com/thelastpickle/cassandra/tree/mck/CASSANDRA-21704/trunk]
> Provenance:
> [2315d6f359|https://github.com/datastax/cassandra/commit/2315d6f359b963bb55c629c2416dc6f66903bf8f]
> by [~mck], in
> [datastax/cassandra#2193|https://github.com/datastax/cassandra/pull/2193],
> where the fork's project name is {{dse-db}}. That pull request covers five of
> these nine sites, and reaches the remaining fork needs by hardcoding
> {{dse-db}} in the two {{cassandra.in.sh}} files and by widening the redhat
> glob. This patch derives the name at all nine instead. The JDK 22 support and
> the docker back-ports in that pull request are fork-only and are not here.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]