Skip to content

Commit

Permalink
Add a git merge driver for MODULE.bazel.lock
Browse files Browse the repository at this point in the history
Adds a `jq` script to `scripts/` that merges any number of `MODULE.bazel.lock` files without using Bazel or reading the corresponding `MODULE.bazel` files.

The lockfile docs now have a section explaining the steps needed to set up this script as a custom merger driver for Git, which means that merge conflicts in `MODULE.bazel.lock` files will always be resolved automatically. Note that resolution may emit lockfiles with redundant information that will be dropped by subsequent Bazel invocations.

When Bazel encounters an error during lockfile parsing that could be caused by a merge conflict, it emits a different error message with a link to the docs. This required fixing the following kind of server crash when a conflict marker occurs inside a `recordedFileInputs` object:
```
FATAL: bazel crashed due to an internal error. Printing stack trace:
java.lang.RuntimeException: Unrecoverable error while evaluating node 'com.google.devtools.build.lib.bazel.bzlmod.BazelLockFileValue$$Lambda/0x000000f8011da998@314cd9ee' (requested by nodes 'RegistryKey{url=https://bcr.bazel.build/}')
	at com.google.devtools.build.skyframe.AbstractParallelEvaluator$Evaluate.run(AbstractParallelEvaluator.java:557)
	at com.google.devtools.build.lib.concurrent.AbstractQueueVisitor$WrappedRunnable.run(AbstractQueueVisitor.java:426)
	at java.base/java.util.concurrent.ForkJoinTask$AdaptedRunnableAction.exec(ForkJoinTask.java:1403)
	at java.base/java.util.concurrent.ForkJoinTask.doExec(ForkJoinTask.java:387)
	at java.base/java.util.concurrent.ForkJoinPool$WorkQueue.topLevelExec(ForkJoinPool.java:1312)
	at java.base/java.util.concurrent.ForkJoinPool.scan(ForkJoinPool.java:1843)
	at java.base/java.util.concurrent.ForkJoinPool.runWorker(ForkJoinPool.java:1808)
	at java.base/java.util.concurrent.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:188)
Caused by: java.lang.IllegalArgumentException: the provided path should be absolute in the filesystem
	at com.google.common.base.Preconditions.checkArgument(Preconditions.java:143)
	at com.google.devtools.build.lib.rules.repository.RepoRecordedInput$RepoCacheFriendlyPath.createOutsideWorkspace(RepoRecordedInput.java:202)
	at com.google.devtools.build.lib.rules.repository.RepoRecordedInput$RepoCacheFriendlyPath.parse(RepoRecordedInput.java:222)
	at com.google.devtools.build.lib.rules.repository.RepoRecordedInput$File$1.parse(RepoRecordedInput.java:265)
	at com.google.devtools.build.lib.bazel.bzlmod.GsonTypeAdapterUtil$11.read(GsonTypeAdapterUtil.java:376)
	at com.google.devtools.build.lib.bazel.bzlmod.GsonTypeAdapterUtil$11.read(GsonTypeAdapterUtil.java:367)
	at com.google.gson.internal.bind.TypeAdapterRuntimeTypeWrapper.read(TypeAdapterRuntimeTypeWrapper.java:41)
	at com.google.gson.internal.bind.MapTypeAdapterFactory$Adapter.read(MapTypeAdapterFactory.java:186)
	at com.google.gson.internal.bind.MapTypeAdapterFactory$Adapter.read(MapTypeAdapterFactory.java:145)
	at com.google.devtools.build.lib.bazel.bzlmod.DelegateTypeAdapterFactory$1.read(DelegateTypeAdapterFactory.java:133)
	at com.google.devtools.build.lib.bazel.bzlmod.LockFileModuleExtension_GsonTypeAdapter.read(LockFileModuleExtension_GsonTypeAdapter.java:171)
	at com.google.devtools.build.lib.bazel.bzlmod.LockFileModuleExtension_GsonTypeAdapter.read(LockFileModuleExtension_GsonTypeAdapter.java:17)
	at com.google.gson.internal.bind.TypeAdapterRuntimeTypeWrapper.read(TypeAdapterRuntimeTypeWrapper.java:41)
	at com.google.gson.internal.bind.MapTypeAdapterFactory$Adapter.read(MapTypeAdapterFactory.java:187)
	at com.google.gson.internal.bind.MapTypeAdapterFactory$Adapter.read(MapTypeAdapterFactory.java:145)
	at com.google.devtools.build.lib.bazel.bzlmod.DelegateTypeAdapterFactory$1.read(DelegateTypeAdapterFactory.java:133)
	at com.google.gson.internal.bind.TypeAdapterRuntimeTypeWrapper.read(TypeAdapterRuntimeTypeWrapper.java:41)
	at com.google.gson.internal.bind.MapTypeAdapterFactory$Adapter.read(MapTypeAdapterFactory.java:187)
	at com.google.gson.internal.bind.MapTypeAdapterFactory$Adapter.read(MapTypeAdapterFactory.java:145)
	at com.google.devtools.build.lib.bazel.bzlmod.DelegateTypeAdapterFactory$1.read(DelegateTypeAdapterFactory.java:133)
	at com.google.devtools.build.lib.bazel.bzlmod.BazelLockFileValue_GsonTypeAdapter.read(BazelLockFileValue_GsonTypeAdapter.java:129)
	at com.google.devtools.build.lib.bazel.bzlmod.BazelLockFileValue_GsonTypeAdapter.read(BazelLockFileValue_GsonTypeAdapter.java:15)
	at com.google.gson.Gson.fromJson(Gson.java:991)
	at com.google.gson.Gson.fromJson(Gson.java:956)
	at com.google.gson.Gson.fromJson(Gson.java:905)
	at com.google.gson.Gson.fromJson(Gson.java:876)
	at com.google.devtools.build.lib.bazel.bzlmod.BazelLockFileFunction.getLockfileValue(BazelLockFileFunction.java:93)
	at com.google.devtools.build.lib.bazel.bzlmod.BazelLockFileFunction.compute(BazelLockFileFunction.java:73)
	at com.google.devtools.build.skyframe.AbstractParallelEvaluator$Evaluate.run(AbstractParallelEvaluator.java:468)
	... 7 more
```

Alternatives considered:
* Letting Bazel resolve the conflict would require building knowledge about particular version control systems and their conflict style into Bazel. It would also either require the user to resolve conflicts in `MODULE.bazel` first or deviate from the current behavior that the lockfile is not updated when any Bzlmod error is encountered. The jq script can be used as is by every VCS with merge driver support and resolves the conflict in `MODULE.bazel.lock` independently of `MODULE.bazel`.
* Implementing the git merge driver as a `bazel mod` subcommand. This could be the source of intransparent slowdowns during regular git operations, which may even be triggered by other tools such as IDEs. The jq script is very fast.
* Implementing the merger as a Go binary in buildtools would replace the ubiquitous jq tool with a special purpose binary while also not solving the problem that per-user action is required once to register a custom merge driver.

Implements https://docs.google.com/document/d/1TjA7-M5njkI1F38IC0pm305S9EOmxcUwaCIvaSmansg/edit#heading=h.5mcn15i0e1ch

RELNOTES: Git merge conflicts in `MODULE.bazel.lock` files can be resolved automatically. See https://bazel.build/external/lockfile#automatic-resolution for the required setup.

Closes #22428.

PiperOrigin-RevId: 640596606
Change-Id: I20659e3e53a7d8f2529f2ad5a3e7f258d7af026d
  • Loading branch information
fmeum authored and copybara-github committed Jun 5, 2024
1 parent 3c47388 commit 3187250
Show file tree
Hide file tree
Showing 8 changed files with 524 additions and 16 deletions.
24 changes: 24 additions & 0 deletions MODULE.bazel
Original file line number Diff line number Diff line change
Expand Up @@ -324,6 +324,30 @@ gvm.graalvm(
)
use_repo(gvm, "graalvm_toolchains")

http_file = use_repo_rule("@bazel_tools//tools/build_defs/repo:http.bzl", "http_file")

# DO NOT UPDATE the jq version, this is used to verify compatibility with old versions.
http_file(
name = "jq_linux_amd64",
executable = True,
integrity = "sha256-xrOn19PntwxvUbcGo7kL0BgzhGxU0yyjLwAn8AIm/20=",
urls = ["https://github.com/jqlang/jq/releases/download/jq-1.5/jq-linux64"],
)

http_file(
name = "jq_macos_amd64",
executable = True,
integrity = "sha256-OG6SyYKlb+SFFGjXqTHfyilWDO4wag5mxqG9QGXT2sU=",
urls = ["https://github.com/jqlang/jq/releases/download/jq-1.5/jq-osx-amd64"],
)

http_file(
name = "jq_windows_amd64",
executable = True,
integrity = "sha256-6+zYQLpH779mgihoF4zHIaFRBgk396xAbj0xvQFb3pQ=",
urls = ["https://github.com/jqlang/jq/releases/download/jq-1.5/jq-win64.exe"],
)

# =========================================
# Other Bazel testing dependencies
# =========================================
Expand Down
89 changes: 80 additions & 9 deletions MODULE.bazel.lock

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

26 changes: 26 additions & 0 deletions scripts/BUILD
Original file line number Diff line number Diff line change
Expand Up @@ -40,6 +40,32 @@ sh_test(
],
)

filegroup(
name = "jq",
srcs = select({
"@platforms//os:linux": ["@jq_linux_amd64//file"],
"@platforms//os:macos": ["@jq_macos_amd64//file"],
"@platforms//os:windows": ["@jq_windows_amd64//file"],
}),
)

sh_test(
name = "bazel_lockfile_merge_test",
size = "small",
srcs = ["bazel_lockfile_merge_test.sh"],
data = [
"bazel-lockfile-merge.jq",
"testenv.sh",
":jq",
"//src/test/shell:bashunit",
"//src/test/tools/bzlmod:MODULE.bazel.lock",
"@bazel_tools//tools/bash/runfiles",
],
env = {
"JQ_RLOCATIONPATH": "$(rlocationpath :jq)",
},
)

filegroup(
name = "srcs",
srcs = glob(["**"]) + [
Expand Down
54 changes: 54 additions & 0 deletions scripts/bazel-lockfile-merge.jq
Original file line number Diff line number Diff line change
@@ -0,0 +1,54 @@
# Merges an arbitrary number of MODULE.bazel.lock files.
#
# Input: an array of MODULE.bazel.lock JSON objects (as produced by `jq -s`).
# Output: a single MODULE.bazel.lock JSON object.
#
# This script assumes that all files are valid JSON and have a numeric
# "lockFileVersion" field. It will not fail on any such files, but only
# preserves information for files with a version of 10 or higher.
#
# The first file is considered to be the base when deciding which values to
# keep in case of conflicts.

# Like unique, but preserves the order of the first occurrence of each element.
def stable_unique:
reduce .[] as $item ([]; if index($item) == null then . + [$item] else . end);

# Given an array of objects, shallowly merges the result of applying f to each
# object into a single object, with a few special properties:
# 1. Values are uniquified before merging and then merged with last-wins
# semantics. Assuming that the first value is the base, this ensures that
# later occurrences of the base value do not override other values. For
# example, when this is called with B A1 A2 and A1 contains changes to a
# field but A2 does not (compared to B), the changes in A1 will be preserved.
# 2. Object keys on the top level are sorted lexicographically after merging,
# but are additionally split on ":". This ensures that module extension IDs,
# which start with labels, sort as strings in the same way as they due as
# structured objects in Bazel (that is, //python/extensions:python.bzl
# sorts before //python/extensions/private:internal_deps.bzl).
def shallow_merge(f):
map(f) | stable_unique | add | to_entries | sort_by(.key | split(":")) | from_entries;

(
# Ignore all MODULE.bazel.lock files that do not have the maximum
# lockFileVersion.
(map(.lockFileVersion) | max) as $maxVersion
| map(select(.lockFileVersion == $maxVersion))
| {
lockFileVersion: $maxVersion,
registryFileHashes: shallow_merge(.registryFileHashes),
selectedYankedVersions: shallow_merge(.selectedYankedVersions),
# Group extension results by extension ID across all lockfiles with
# shallowly merged factors map, then shallowly merge the results.
moduleExtensions: (map(.moduleExtensions | to_entries)
| flatten
| group_by(.key)
| shallow_merge({(.[0].key): shallow_merge(.value)}))
}
)? //
# We get here if the lockfiles with the highest lockFileVersion could not be
# processed, for example because all lockfiles have lockFileVersion < 10.
# In this case Bazel 7.2.0+ would ignore all lockfiles, so we might as well
# return the first lockfile for the proper "mismatched version" error
# message.
.[0]
Loading

0 comments on commit 3187250

Please sign in to comment.