How to properly use bazel transitions for multiarch build - bazel

I'm trying to define a bazel rule that will build 2 different cc_binaries for 2 different platforms with just 1 bazel build invocation. I'm struggling with how to define the transition properly and attach it.
I would like ultimately to be able to do something like:
cc_binary(
name = "binary_platform_a"
...
)
cc_binary(
name = "binary_platform_b"
...
)
my_custom_multi_arch_rule(
name = "multiarch_build",
binary_a = ":binary_platform_a",
binary_b = ":binary_platform_b",
...
)
I have deduced from bazel documents: [https://bazel.build/rules/config#user-defined-transitions] that I need to do something like the following in a defs.bzl:
def _impl(settings, attr):
_ignore = (settings, attr)
return {
"Platform A": {"//command_line_option:platform": "platform_a"},
"Platform B": {"//command_line_option:platform": "platform_b"},
}
multi_arch_transition = transition(
implementation = _impl,
inputs = [],
outputs = ["//command_line_option:platform"]
)
def _rule_impl(ctx):
# How to implement this?
my_custom_multi_arch_rule = rule(
implementation = _rule_impl,
attrs = {
"binary_a": attr.label(cfg = multi_arch_transition)
"binary_b": attr.label(cfg = multi_arch_transition)
...
})
The best-case final scenario would be able to issue:
bazel build //path/to/my:multiarch_build
and it successfully builds my 2 separate binaries for their respective platforms.

Use ctx.split_attr.<attr name>[<transition key>] to get the configured Target object representing a particular arch configuration of a binary.
def _rule_impl(ctx):
binary_a_platform_a = ctx.split_attr.binary_a["Platform A"]
binary_a_platform_b = ctx.split_attr.binary_b["Platform B"]
# ...
return [DefaultInfo(files = depset([binary_a_platform_a, binary_b_platform_b, ...]))]
https://bazel.build/rules/config#accessing-attributes-with-transitions

Related

Is there a way to pass the 'bazel run' result file to the next rule file in bazel?

I created a first rule file to geneate a scirpt using the ctx.actions.expand_template and ran it. and I wanted that pass the result of bazel run with first rule to the next 2nd rule file as a resource. But, I could't get the result which is generated by bazel run 1st rule in the 2nd rule.
It does not mean the script created by the first rule, but the file created when the script is executed.
Below example is what I tested.
This is bazel rule file
def _1st_rule_impl(ctx):
...
out = ctx.outputs.executable
template = ctx.file.my_template
ctx.actions.expand_template(
output = out,
template = template,
substitutions = {
"{ARG1}: ctx.attr.my_file_name,
"{ARG2}: ctx.attr.my_file_content,
}
return [
DefaultInfo(
files = depset([out]),
runfiles = ctx.runfiles(files = [out])
)
]
1st_rule = rule(
implementation = _1st_rule_impl,
attrs = {
"my_template":attr.label(
allow_single_file = True,
default = Label("#my_test//my_rules:my_script.sh.template"),
),
"my_file_name": attr.string(
default = "myfile.txt",
),
"my_file_content": attr.string(
default = "hello world",
),
},
executable = True,
)
def _2nd_rule_impl(ctx):
...
for dep in ctx.attr.deps:
//
// HOW CAN I GET THE RESULT OF `bazel run` THE '1st_rule'?
// I WANT TO GET THE `myfile.txt` WHICH IS GENERATED BY '1st_rule'
//
return [
DefaultInfo(
files = depset([out]),
runfiles = ctx.runfiles(files = [out]),
)
]
2nd_rule = rule(
implementation = _2nd_rule_impl,
attrs = {
"dep":attr.label_list(
mandatory= True,
),
...
},
executable = True,
)
This is BUILD file
1st_rule(
name = "my_1st_rule",
my_file_name = "myfile.txt",
my_file_content = "hello world",
)
2nd_rule(
name = "my_2nd_rule",
dep = [
":my_1st_rule",
],
)
This is template shell script
...
function create_file() {
echo "{ARG2}" > "{ARG1}"
}
...
I tested with the example described above.
There are several ways to access the outputs of dependencies. Perhaps, the simplest is ctx.files. For example, print(ctx.files.dep) in _2nd_rule_impl should show the output of my_1st_rule when my_2nd_rule is analyzed.

How do I get my custom header template rule to pass it's output downstream cc_binary/cc_library dependency?

I'm trying to build a rule for bazel which emulates the CMake *.in template system.
This has two challenges, the first is generate the template output. The second is make the output available to both genrules, filegroups and cc_* rules. The third is to get that dependency to transitively be passed to further downstream rules.
I have it generating the output file version.hpp in genfiles (or bazel-bin), and I can get the initial library rule to include it, but I can't seem to figure out how to make my cc_binary rule, which depends on the cc_library and transitively the header_template rule to find the header file.
I have the following .bzl rule:
def _header_template_impl(ctx):
# this generates the output from the template
ctx.actions.expand_template(
template = ctx.file.template,
output = ctx.outputs.out,
substitutions = ctx.attr.vars,
)
return [
# create a provider which says that this
# out file should be made available as a header
CcInfo(compilation_context=cc_common.create_compilation_context(
headers=depset([ctx.outputs.out])
)),
# Also create a provider referencing this header ???
DefaultInfo(files=depset(
[ctx.outputs.out]
))
]
header_template = rule(
implementation = _header_template_impl,
attrs = {
"vars": attr.string_dict(
mandatory = True
),
"extension": attr.string(default=".hpp"),
"template": attr.label(
mandatory = True,
allow_single_file = True,
),
},
outputs = {
"out": "%{name}%{extension}",
},
output_to_genfiles = True,
)
elsewhere I have a cc_library rule:
load("//:tools/header_template.bzl", "header_template")
# version control
BONSAI_MAJOR_VERSION = '2'
BONSAI_MINOR_VERSION = '0'
BONSAI_PATCH_VERSION = '9'
BONSAI_VERSION = \
BONSAI_MAJOR_VERSION + '.' + \
BONSAI_MINOR_VERSION + '.' + \
BONSAI_PATCH_VERSION
header_template(
name = "bonsai_version",
extension = ".hpp",
template = "version.hpp.in",
vars = {
"#BONSAI_MAJOR_VERSION#": BONSAI_MAJOR_VERSION,
"#BONSAI_MINOR_VERSION#": BONSAI_MINOR_VERSION,
"#BONSAI_PATCH_VERSION#": BONSAI_PATCH_VERSION,
"#BONSAI_VERSION#": BONSAI_VERSION,
},
)
# ...
private = glob([
"src/**/*.hpp",
"src/**/*.cpp",
"proto/**/*.hpp",
])
public = glob([
"include/*.hpp",
":bonsai_version",
])
cc_library(
# target name matches directory name so you can call:
# bazel build .
name = "bonsai",
srcs = private,
hdrs = public,
# public headers
includes = [
"include",
],
# ...
deps = [
":bonsai_version",
# ...
],
# ...
)
When I build, my source files need to be able to:
#include "bonsai_version.hpp"
I think the answer involves CcInfo but I'm grasping in the dark as to how it should be constructed.
I've already tried add "-I$(GENDIR)/" + package_name() to the copts, to no avail. The generated header still isn't available.
My expectation is that I should be able to return some kind of Info object that would allow me to add the dependency in srcs. Maybe it should be a DefaultInfo.
I've dug through the bazel rules examples and the source, but I'm missing something fundamental, and I can't find documentation that discuss this particular.
I'd like to be able to do the following:
header_template(
name = "some_header",
extension = ".hpp",
template = "some_header.hpp.in",
vars = {
"#SOMEVAR#": "value",
"{ANOTHERVAR}": "another_value",
},
)
cc_library(
name = "foo",
srcs = ["foo.src", ":some_header"],
...
)
cc_binary(
name = "bar",
srcs = ["bar.cpp"],
deps = [":foo"],
)
and include the generated header like so:
#include "some_header.hpp"
void bar(){
}
The answer looks like it is:
def _header_template_impl(ctx):
# this generates the output from the template
ctx.actions.expand_template(
template = ctx.file.template,
output = ctx.outputs.out,
substitutions = ctx.attr.vars,
)
return [
# create a provider which says that this
# out file should be made available as a header
CcInfo(compilation_context=cc_common.create_compilation_context(
# pass out the include path for finding this header
includes=depset([ctx.outputs.out.dirname]),
# and the actual header here.
headers=depset([ctx.outputs.out])
))
]
elsewhere:
header_template(
name = "some_header",
extension = ".hpp",
template = "some_header.hpp.in",
vars = {
"#SOMEVAR#": "value",
"{ANOTHERVAR}": "another_value",
},
)
cc_library(
name = "foo",
srcs = ["foo.cpp"],
deps = [":some_header"],
...
)
cc_binary(
name = "bar",
srcs = ["bar.cpp"],
deps = [":foo"],
)
If your header has a generic name (eg config.h) and you want it to be private (ie srcs instead of hdrs), you might need a different approach. I've seen this problem for gflags, which "leaked" config.h and affected libraries that depended on it (issue).
Of course, in both cases, the easiest solution is to generate and commit header files for the platforms you target.
Alternatively, you can set copts for the cc_library rule that uses the generated private header:
cc_library(
name = "foo",
srcs = ["foo.cpp", "some_header.hpp"],
copts = ["-I$(GENDIR)/my/package/name"],
...
)
If you want this to work when your repository is included as an external repository, you have a bit more work cut out for you due to bazel issue #4463.
PS. You might want to see if cc_fix_config from https://github.com/antonovvk/bazel_rules works for you. It's just a wrapper around perl but I found it useful.

Custom C++ rule with the cc_common API

I'm trying to write a custom rule to compile C++ code using the cc_common API. Here's my current attempt at an implementation:
load("#bazel_tools//tools/cpp:toolchain_utils.bzl", "find_cpp_toolchain")
load("#bazel_tools//tools/build_defs/cc:action_names.bzl", "C_COMPILE_ACTION_NAME")
def _impl(ctx):
cc_toolchain = find_cpp_toolchain(ctx)
feature_configuration = cc_common.configure_features(
cc_toolchain = cc_toolchain,
unsupported_features = ctx.disabled_features,
)
compiler = cc_common.get_tool_for_action(
feature_configuration=feature_configuration,
action_name=C_COMPILE_ACTION_NAME
)
compile_variables = cc_common.create_compile_variables(
feature_configuration = feature_configuration,
cc_toolchain = cc_toolchain,
)
compiler_options = cc_common.get_memory_inefficient_command_line(
feature_configuration = feature_configuration,
action_name = C_COMPILE_ACTION_NAME,
variables = compile_variables,
)
outfile = ctx.actions.declare_file("test.o")
args = ctx.actions.args()
args.add_all(compiler_options)
ctx.actions.run(
outputs = [outfile],
inputs = ctx.files.srcs,
executable = compiler,
arguments = [args],
)
return [DefaultInfo(files = depset([outfile]))]
However, this fails with the error "execvp(external/local_config_cc/wrapped_clang, ...)": No such file or directory. I assume this is because get_tool_for_action returns a string representing a path, not a File object, so Bazel doesn't add wrapped_clang to the sandbox. Executing the rule with sandboxing disabled seems to confirm this, as it completes successfully.
Is there a way to implement this custom rule without disabling the sandbox?
If you use ctx.actions.run_shell you can add the files associated with the toolchain to the input (ctx.attr._cc_toolchain.files). Also, you'll want to add the compiler environment variables. E.g.
srcs = depset(ctx.files.srcs)
tools = ctx.attr._cc_toolchain.files
...
compiler_env = cc_common.get_environment_variables(
feature_configuration = feature_configuration,
action_name = C_COMPILE_ACTION_NAME,
variables = compiler_variables,
)
...
args = ctx.actions.args()
args.add_all(compiler_options)
ctx.actions.run_shell(
outputs = [outfile],
inputs = depset(transitive = [srcs, tools]), # Merge src and tools depsets
command = "{compiler} $*".format(compiler = compiler),
arguments = [args],
env = compiler_env,
)
Bazel doesn't add files as action inputs automatically, you have to do it explicitly, as you did in your second approach (ctx.attr._cc_toolchain.files). With that, ctx.actions.run should work just fine.

Filter source files for custom rule

I used a bazel macro to run a python test on a subset of source files. Similar to this:
def report(name, srcs):
source_labels = [file for file in srcs if file.startswith("a")]
if len(source_labels) == 0:
return;
source_filenames = ["$(location %s)" % x for x in source_labels]
native.py_test(
name = name + "_report",
srcs = ["report_tool"],
data = source_labels,
main = "report_tool.py",
args = source_filenames,
)
report("foo", ["foo.hpp", "afoo.hpp"])
This worked fine until one of my source files started using a select and now I get the error:
File "/home/david/foo/report.bzl", line 47, in report
[file for file in srcs if file.startswith("a")]
type 'select' is not iterable
I tried to move the code to a bazel rule, but then I get a different error that py_test can not be used in the analysis phase.
The reason that the select is causing the error is that macros are evaluated during the loading phase, whereas selectss are not evaluated until the analysis phase (see Extension Overview).
Similarly, py_test can't be used in a rule implementation because the rule implementation is evaluated in the analysis phase, whereas the py_test would need to have been loaded in the loading phase.
One way past this is to create a separate Starlark rule that takes a list of labels and just creates a file with each filename from the label. Then the py_test takes that file as data and loads the other files from there. Something like this:
def report(name, srcs):
file_locations_label = "_" + name + "_file_locations"
_generate_file_locations(
name = file_locations_label,
labels = srcs
)
native.py_test(
name = name + "_report",
srcs = ["report_tool.py"],
data = srcs + [file_locations_label],
main = "report_tool.py",
args = ["$(location %s)" % file_locations_label]
)
def _generate_file_locations_impl(ctx):
paths = []
for l in ctx.attr.labels:
f = l.files.to_list()[0]
if f.basename.startswith("a"):
paths.append(f.short_path)
ctx.actions.write(ctx.outputs.file_paths, "\n".join(paths))
return DefaultInfo(runfiles = ctx.runfiles(files = [ctx.outputs.file_paths]))
_generate_file_locations = rule(
implementation = _generate_file_locations_impl,
attrs = { "labels": attr.label_list(allow_files = True) },
outputs = { "file_paths": "%{name}_files" },
)
This has one disadvantage: Because the py_test has to depend on all the sources, the py_test will get rerun even if the only files that have changed are the ignored files. (If this is a significant drawback, then there is at least one way around this, which is to have _generate_file_locations filter the files too, and have the py_test depend on only _generate_file_locations. This could maybe be accomplished through runfiles symlinks)
Update:
Since the test report tool comes from an external repository and can't be easily modified, here's another approach that might work better. Rather than create a rule that creates a params file (a file containing the paths to process) as above, the Starlark rule can itself be a test rule that uses the report tool as the test executable:
def _report_test_impl(ctx):
filtered_srcs = []
for f in ctx.attr.srcs:
f = f.files.to_list()[0]
if f.basename.startswith("a"):
filtered_srcs.append(f)
report_tool = ctx.attr._report_test_tool
ctx.actions.write(
output = ctx.outputs.executable,
content = "{report_tool} {paths}".format(
report_tool = report_tool.files_to_run.executable.short_path,
paths = " ".join([f.short_path for f in filtered_srcs]))
)
runfiles = ctx.runfiles(files = filtered_srcs).merge(
report_tool.default_runfiles)
return DefaultInfo(runfiles = runfiles)
report_test = rule(
implementation = _report_test_impl,
attrs = {
"srcs": attr.label_list(allow_files = True),
"_report_test_tool": attr.label(default="//:report_test_tool"),
},
test = True,
)
This requires that the test report tool be a py_binary somewhere so that the test rule above can depend on it:
py_binary(
name = "report_test_tool",
srcs = ["report_tool.py"],
main = "report_tool.py",
)

Idiomatic retrieval of the Bazel execution path

I'm working on my first custom Bazel rules. The rules allow the running of bats command line tests.
I've included the rule definition below verbatim. I'm pretty happy with it so far but there's one part which feels really ugly and non-standard. If the rule user adds a binary dependency to the rule then I make sure that the binary appears on the PATH so that it can be tested. At the moment I do this by making a list of the binary paths and then appending them with $PWD which is expanded inside the script to the complete execution path. This feels hacky and error prone.
Is there a more idiomatic way to do this? I don't believe I can access the execution path in the rule due to it not being created until the execution phase.
Thanks for your help!
BATS_REPOSITORY_BUILD_FILE = """
package(default_visibility = [ "//visibility:public" ])
sh_binary(
name = "bats",
srcs = ["libexec/bats"],
data = [
"libexec/bats-exec-suite",
"libexec/bats-exec-test",
"libexec/bats-format-tap-stream",
"libexec/bats-preprocess",
],
)
"""
def bats_repositories(version="v0.4.0"):
native.new_git_repository(
name = "bats",
remote = "https://github.com/sstephenson/bats",
tag = version,
build_file_content = BATS_REPOSITORY_BUILD_FILE
)
BASH_TEMPLATE = """
#!/usr/bin/env bash
set -e
export TMPDIR="$TEST_TMPDIR"
export PATH="{bats_bins_path}":$PATH
"{bats}" "{test_paths}"
"""
def _dirname(path):
prefix, _, _ = path.rpartition("/")
return prefix.rstrip("/")
def _bats_test_impl(ctx):
runfiles = ctx.runfiles(
files = ctx.files.srcs,
collect_data = True,
)
tests = [f.short_path for f in ctx.files.srcs]
path = ["$PWD/" + _dirname(b.short_path) for b in ctx.files.deps]
sep = ctx.configuration.host_path_separator
ctx.file_action(
output = ctx.outputs.executable,
executable = True,
content = BASH_TEMPLATE.format(
bats = ctx.executable._bats.short_path,
test_paths = " ".join(tests),
bats_bins_path = sep.join(path),
),
)
runfiles = runfiles.merge(ctx.attr._bats.default_runfiles)
return DefaultInfo(
runfiles = runfiles,
)
bats_test = rule(
attrs = {
"srcs": attr.label_list(
allow_files = True,
),
"deps": attr.label_list(),
"_bats": attr.label(
default = Label("#bats//:bats"),
executable = True,
cfg = "host",
),
},
test = True,
implementation = _bats_test_impl,
)
This should be easy to support from Bazel 0.8.0 which will be released in ~2 weeks.
In your skylark implementation you should do ctx.expand_location(binary) where binary should be something like $(execpath :some-label) so you might want to just format the label you got from the user with the $(execpath) and bazel will make sure to give you the execution location of that label.
Some relevant resources:
$location expansion in Bazel
https://github.com/bazelbuild/bazel/issues/2475
https://github.com/bazelbuild/bazel/commit/cff0dc94f6a8e16492adf54c88d0b26abe903d4c

Resources