Clone SOURCES repos outside container

This lets us always use speed-up options like --filter, even when we're
building inside a container.

We can also easily start using sapling for this in future by implementing the
new SCM methods.

This also helps with some annoying authentication-related issues, if running
inside --docker, such as in CI.

Fixes <https://its.cern.ch/jira/browse/O2-2439>.

When this is merged, we can also get rid of the `GIT_*` environment variables
used in CI, and stop installing an especially-new Git version in our
containers (since that was needed to understand the `GIT_*` variables).

alibuild_helpers/build.py

@@ -13,12 +13,12 @@
 from alibuild_helpers.utilities import Hasher
 from alibuild_helpers.utilities import yamlDump
 from alibuild_helpers.utilities import resolve_tag, resolve_version
-from alibuild_helpers.git import git, clone_speedup_options, Git
+from alibuild_helpers.git import Git, git
 from alibuild_helpers.sl import Sapling
 from alibuild_helpers.scm import SCMError
 from alibuild_helpers.sync import remote_from_url
 import yaml
-from alibuild_helpers.workarea import logged_scm, updateReferenceRepoSpec
+from alibuild_helpers.workarea import logged_scm, updateReferenceRepoSpec, checkout_sources
 from alibuild_helpers.log import ProgressPrint, log_current_package
 from glob import glob
 from textwrap import dedent
@@ -69,9 +69,7 @@ def update_repo(package, git_prompt):
             if exists(os.path.join(specs[package]["source"], ".sl")):
                 specs[package]["scm"] = Sapling()
         updateReferenceRepoSpec(args.referenceSources, package, specs[package],
-                                fetch=args.fetchRepos,
-                                usePartialClone=not args.docker,
-                                allowGitPrompt=git_prompt)
+                                fetch=args.fetchRepos, allowGitPrompt=git_prompt)
 
         # Retrieve git heads
         output = logged_scm(specs[package]["scm"], package, args.referenceSources,
@@ -967,10 +965,6 @@ def doBuild(args, parser):
             if spec["cachedTarball"] else "No cache tarballs found")
 
     # The actual build script.
-    referenceStatement = ""
-    if "reference" in spec:
-      referenceStatement = "export GIT_REFERENCE=${GIT_REFERENCE_OVERRIDE:-%s}/%s" % (dirname(spec["reference"]), basename(spec["reference"]))
-
     debug("spec = %r", spec)
 
     cmd_raw = ""
@@ -982,22 +976,14 @@ def doBuild(args, parser):
       from pkg_resources import resource_string
       cmd_raw = resource_string("alibuild_helpers", 'build_template.sh')
 
-    source = spec.get("source", "")
-    # Shortend the commit hash in case it's a real commit hash and not simply
-    # the tag.
-    commit_hash = spec["commit_hash"]
-    if spec["tag"] != spec["commit_hash"]:
-      commit_hash = spec["commit_hash"][0:10]
-
-    # Split the source in two parts, sourceDir and sourceName.  This is done so
-    # that when we use Docker we can replace sourceDir with the correct
-    # container path, if required.  No changes for what concerns the standard
-    # bash builds, though.
     if args.docker:
       cachedTarball = re.sub("^" + workDir, "/sw", spec["cachedTarball"])
     else:
       cachedTarball = spec["cachedTarball"]
 
+    if not cachedTarball:
+      checkout_sources(spec, workDir, args.referenceSources, args.docker)
+
     scriptDir = join(workDir, "SPECS", args.architecture, spec["package"],
                      spec["version"] + "-" + spec["revision"])
 
@@ -1011,11 +997,6 @@ def doBuild(args, parser):
       "workDir": workDir,
       "configDir": abspath(args.configDir),
       "incremental_recipe": spec.get("incremental_recipe", ":"),
-      "sourceDir": (dirname(source) + "/") if source else "",
-      "sourceName": basename(source) if source else "",
-      "referenceStatement": referenceStatement,
-      "gitOptionsStatement": "" if args.docker else
-      "export GIT_CLONE_SPEEDUP=" + quote(" ".join(clone_speedup_options())),
       "requires": " ".join(spec["requires"]),
       "build_requires": " ".join(spec["build_requires"]),
       "runtime_requires": " ".join(spec["runtime_requires"]),
@@ -1028,14 +1009,14 @@ def doBuild(args, parser):
       ("BUILD_REQUIRES", " ".join(spec["build_requires"])),
       ("CACHED_TARBALL", cachedTarball),
       ("CAN_DELETE", args.aggressiveCleanup and "1" or ""),
-      ("COMMIT_HASH", commit_hash),
+      # Shorten the commit hash if it's a real commit hash and not simply the tag.
+      ("COMMIT_HASH", spec["tag"] if spec["tag"] == spec["commit_hash"] else spec["commit_hash"][:10]),
       ("DEPS_HASH", spec.get("deps_hash", "")),
       ("DEVEL_HASH", spec.get("devel_hash", "")),
       ("DEVEL_PREFIX", develPrefix),
       ("BUILD_FAMILY", spec["build_family"]),
       ("GIT_COMMITTER_NAME", "unknown"),
       ("GIT_COMMITTER_EMAIL", "unknown"),
-      ("GIT_TAG", spec["tag"]),
       ("INCREMENTAL_BUILD_HASH", spec.get("incremental_hash", "0")),
       ("JOBS", str(args.jobs)),
       ("PKGHASH", spec["hash"]),
@@ -1048,7 +1029,6 @@ def doBuild(args, parser):
       ("FULL_RUNTIME_REQUIRES", " ".join(spec["full_runtime_requires"])),
       ("FULL_BUILD_REQUIRES", " ".join(spec["full_build_requires"])),
       ("FULL_REQUIRES", " ".join(spec["full_requires"])),
-      ("WRITE_REPO", spec.get("write_repo", source)),
     ]
     # Add the extra environment as passed from the command line.
     buildEnvironment += [e.partition('=')[::2] for e in args.environment]
@@ -1059,15 +1039,13 @@ def doBuild(args, parser):
       build_command = (
         "docker run --rm --entrypoint= --user $(id -u):$(id -g) "
         "-v {workdir}:/sw -v {scriptDir}/build.sh:/build.sh:ro "
-        "-e GIT_REFERENCE_OVERRIDE=/mirror -e WORK_DIR_OVERRIDE=/sw "
         "{mirrorVolume} {develVolumes} {additionalEnv} {additionalVolumes} "
-        "{overrideSource} {extraArgs} {image} bash -ex /build.sh"
+        "-e WORK_DIR_OVERRIDE=/sw {extraArgs} {image} bash -ex /build.sh"
       ).format(
         image=quote(args.dockerImage),
         workdir=quote(abspath(args.workDir)),
         scriptDir=quote(scriptDir),
         extraArgs=" ".join(map(quote, args.docker_extra_args)),
-        overrideSource="-e SOURCE0_DIR_OVERRIDE=/" if source.startswith("/") else "",
         additionalEnv=" ".join(
           "-e {}={}".format(var, quote(value)) for var, value in buildEnvironment),
         # Used e.g. by O2DPG-sim-tests to find the O2DPG repository.

alibuild_helpers/build_template.sh

@@ -27,7 +27,6 @@ export PATH=$WORK_DIR/wrapper-scripts:$PATH
 # - DEPS_HASH
 # - DEVEL_HASH
 # - DEVEL_PREFIX
-# - GIT_TAG
 # - INCREMENTAL_BUILD_HASH
 # - JOBS
 # - PKGHASH
@@ -36,36 +35,27 @@ export PATH=$WORK_DIR/wrapper-scripts:$PATH
 # - PKGVERSION
 # - REQUIRES
 # - RUNTIME_REQUIRES
-# - WRITE_REPO
 
 export PKG_NAME="$PKGNAME"
 export PKG_VERSION="$PKGVERSION"
 export PKG_BUILDNUM="$PKGREVISION"
 
-export SOURCE0="${SOURCE0_DIR_OVERRIDE:-%(sourceDir)s}%(sourceName)s"
 export PKGPATH=${ARCHITECTURE}/${PKGNAME}/${PKGVERSION}-${PKGREVISION}
 mkdir -p "$WORK_DIR/BUILD" "$WORK_DIR/SOURCES" "$WORK_DIR/TARS" \
          "$WORK_DIR/SPECS" "$WORK_DIR/INSTALLROOT"
 export BUILDROOT="$WORK_DIR/BUILD/$PKGHASH"
 
-# In case the repository is local, it means we are in development mode, so we
-# install directly in $WORK_DIR/$PKGPATH so that we can do make install
-# directly into BUILD/$PKGPATH and have changes being propagated.
-if [ "${SOURCE0:0:1}" == "/" ]; then
+# If we are in development mode, then install directly in $WORK_DIR/$PKGPATH,
+# so that we can do "make install" directly into BUILD/$PKGPATH and have
+# changes being propagated.
+if [ -n "$DEVEL_HASH" ]; then
   export INSTALLROOT="$WORK_DIR/$PKGPATH"
 else
   export INSTALLROOT="$WORK_DIR/INSTALLROOT/$PKGHASH/$PKGPATH"
 fi
 export SOURCEDIR="$WORK_DIR/SOURCES/$PKGNAME/$PKGVERSION/$COMMIT_HASH"
 export BUILDDIR="$BUILDROOT/$PKGNAME"
 
-SHORT_TAG=${GIT_TAG:0:10}
-mkdir -p $(dirname $SOURCEDIR)
-if [[ ${COMMIT_HASH} != ${GIT_TAG} && "${SHORT_TAG:-0}" != ${COMMIT_HASH} ]]; then
-  GIT_TAG_DIR=${GIT_TAG:-0}
-  GIT_TAG_DIR=${GIT_TAG_DIR//\//_}
-  ln -snf ${COMMIT_HASH} "$WORK_DIR/SOURCES/$PKGNAME/$PKGVERSION/${GIT_TAG_DIR}"
-fi
 rm -fr "$WORK_DIR/INSTALLROOT/$PKGHASH"
 # We remove the build directory only if we are not in incremental mode.
 if [[ "$INCREMENTAL_BUILD_HASH" == 0 ]] && ! rm -rf "$BUILDROOT"; then
@@ -114,35 +104,6 @@ if [[ $DEVEL_PREFIX ]]; then
   ln -snf $PKGHASH $WORK_DIR/BUILD/$PKGNAME-latest-$DEVEL_PREFIX
 fi
 
-# Reference statements
-%(referenceStatement)s
-%(gitOptionsStatement)s
-
-if [ -z "$CACHED_TARBALL" ]; then
-  case "$SOURCE0" in
-    '')  # SOURCE0 is empty, so just create an empty SOURCEDIR.
-      mkdir -p "$SOURCEDIR" ;;
-    /*)  # SOURCE0 is an absolute path, so just make a symlink there.
-      ln -snf "$SOURCE0" "$SOURCEDIR" ;;
-    *)   # SOURCE0 is a relative path or URL, so clone/checkout the git repo from there.
-      if cd "$SOURCEDIR" 2>/dev/null; then
-        # Folder is already present, but check that it is the right tag
-        if ! git checkout -f "$GIT_TAG"; then
-          # If we can't find the tag, it might be new. Fetch tags and try again.
-          git fetch -f "$SOURCE0" "refs/tags/$GIT_TAG:refs/tags/$GIT_TAG"
-          git checkout -f "$GIT_TAG"
-        fi
-      else
-        # In case there is a stale link / file, for whatever reason.
-        rm -rf "$SOURCEDIR"
-        git clone -n $GIT_CLONE_SPEEDUP ${GIT_REFERENCE:+--reference "$GIT_REFERENCE"} "$SOURCE0" "$SOURCEDIR"
-        cd "$SOURCEDIR"
-        git remote set-url --push origin "$WRITE_REPO"
-        git checkout -f "$GIT_TAG"
-      fi ;;
-  esac
-fi
-
 cd "$BUILDDIR"
 
 # Actual build script, as defined in the recipe

alibuild_helpers/git.py

@@ -18,34 +18,58 @@ def clone_speedup_options():
 
 class Git(SCM):
   name = "Git"
+
   def checkedOutCommitName(self, directory):
     return git(("rev-parse", "HEAD"), directory)
+
   def branchOrRef(self, directory):
     out = git(("rev-parse", "--abbrev-ref", "HEAD"), directory=directory)
     if out == "HEAD":
       out = git(("rev-parse", "HEAD"), directory)[:10]
     return out
+
   def exec(self, *args, **kwargs):
     return git(*args, **kwargs)
+
   def parseRefs(self, output):
     return {
       git_ref: git_hash for git_hash, sep, git_ref
       in (line.partition("\t") for line in output.splitlines()) if sep
     }
+
   def listRefsCmd(self, repository):
     return ["ls-remote", "--heads", "--tags", repository]
-  def cloneCmd(self, source, referenceRepo, usePartialClone):
+
+  def cloneReferenceCmd(self, source, referenceRepo, usePartialClone):
     cmd = ["clone", "--bare", source, referenceRepo]
     if usePartialClone:
       cmd.extend(clone_speedup_options())
     return cmd
-  def fetchCmd(self, source):
-    return ["fetch", "-f", "--tags", source, "+refs/heads/*:refs/heads/*"]
+
+  def cloneSourceCmd(self, source, destination, referenceRepo, usePartialClone):
+    cmd = ["clone", "-n", source, destination]
+    if referenceRepo:
+      cmd.extend(["--reference", referenceRepo])
+    if usePartialClone:
+      cmd.extend(clone_speedup_options())
+    return cmd
+
+  def checkoutCmd(self, ref):
+    return ["checkout", "-f", ref]
+
+  def fetchCmd(self, source, *refs):
+    return ["fetch", "-f", source, *refs]
+
+  def setWriteUrlCmd(self, url):
+    return ["remote", "set-url", "--push", "origin", url]
+
   def diffCmd(self, directory):
     return "cd %s && git diff -r HEAD && git status --porcelain" % directory
+
   def checkUntracked(self, line):
     return line.startswith("?? ")
 
+
 def git(args, directory=".", check=True, prompt=True):
   debug("Executing git %s (in directory %s)", " ".join(args), directory)
   # We can't use git --git-dir=%s/.git or git -C %s here as the former requires

alibuild_helpers/init.py

@@ -66,6 +66,7 @@ def doInit(args):
 
   for p in pkgs:
     spec = specs.get(p["name"])
+    spec["is_devel_pkg"] = False
     spec["scm"] = Git()
     dieOnError(spec is None, "cannot find recipe for package %s" % p["name"])
     dest = join(args.develPrefix, spec["package"])

alibuild_helpers/scm.py

@@ -15,7 +15,15 @@ def parseRefs(self, output):
     raise NotImplementedError
   def exec(self, *args, **kwargs):
     raise NotImplementedError
-  def cloneCmd(self, spec, referenceRepo, usePartialClone):
+  def checkoutCmd(self, tag):
+    raise NotImplementedError
+  def fetchCmd(self, remote, *refs):
+    raise NotImplementedError
+  def cloneReferenceCmd(self, spec, referenceRepo, usePartialClone):
+    raise NotImplementedError
+  def cloneSourceCmd(self, spec, referenceRepo, usePartialClone):
+    raise NotImplementedError
+  def setWriteUrlCmd(self, url):
     raise NotImplementedError
   def diffCmd(self, directory):
     raise NotImplementedError

alibuild_helpers/sl.py

@@ -6,6 +6,7 @@
 SL_COMMAND_TIMEOUT_SEC = 120
 """How many seconds to let any sl command execute before being terminated."""
 
+
 # Sapling is a novel SCM by Meta (i.e. Facebook) that is fully compatible with
 # git, but has a different command line interface. Among the reasons why it's
 # worth suporting it is the ability to handle unnamed branches, the ability to
@@ -14,26 +15,34 @@
 # command line from each commit of a branch.
 class Sapling(SCM):
   name = "Sapling"
+
   def checkedOutCommitName(self, directory):
     return sapling(("whereami", ), directory)
+
   def branchOrRef(self, directory):
     # Format is <hash>[+] <branch>
     identity = sapling(("identify", ), directory)
     return identity.split(" ")[-1]
+
   def exec(self, *args, **kwargs):
     return sapling(*args, **kwargs)
+
   def parseRefs(self, output):
     return {
       sl_ref: sl_hash for sl_ref, sep, sl_hash
       in (line.partition("\t") for line in output.splitlines()) if sep
     }
+
   def listRefsCmd(self, repository):
     return ["bookmark", "--list", "--remote", "-R", repository]
+
   def diffCmd(self, directory):
     return "cd %s && sl diff && sl status" % directory
+
   def checkUntracked(self, line):
     return line.startswith("? ")
 
+
 def sapling(args, directory=".", check=True, prompt=True):
   debug("Executing sl %s (in directory %s)", " ".join(args), directory)
   # We can't use git --git-dir=%s/.git or git -C %s here as the former requires