Skip to content

hzeller/bant

Repository files navigation

License: GPL v2 Continuous Integration

bant - Build Analysis and Navigation Tool

Utility to support projects using the bazel build system, in particular C++ projects.

Bant

  • Helps cleaning up BUILD files by adding missing, and removing superfluous, dependencies (build_cleaner). Outputs buildozer script.
  • Helps finding targets with grep feature that emits the full surrounding target.
  • Extracts interesting project information such as the dependency graph, headers provided by which libraries etc., and presents them for easy post-proccessing (outputs simple tables for grep or awk, but as well CSV, JSON and S-Expressions).
  • Canonicalize targets.

Commands

Bant is invoked with a command and a bazel-like pattern such as ... or //foo/bar:all

bant [options] <command> [bazel-target-pattern]

See bant -h for general Synopsis and available commands. Commands can be shortened as long as they are unique, e.g. lib is sufficient to invoke lib-headers.

Play around with the various -f output formats of workspace, list-packages etc. for your preferred post-processing tools (default is just a plain table easy to process with e.g. awk and grep, but s-expressions and json are structured outputs interesting in other contexts).

Useful everyday commands

Print

print is useful to print a particular rule or rules matching a pattern bant print //foo/bar:baz (it is re-created from the AST, so you see exactly what bant sees). In general this is much faster than manually finding and opening the file, in particular when it is in some external project where it is harder to even find the right file; consider

bant print @googletest//:gtest

You see that gtest has some files glob()'d and other expressions that make that rule, which we want to see evaluated. With the -e option (for elaborate), bant can do some basic evaluation of these and print the final form:

bant print -e @googletest//:gtest

The -g option allows to 'grep' for targets where the rule-name or any of the strings in the AST of a rule matches a pattern.

bant print ... -g "scan.*test"

Evaluation of custom flags in select() also works, and similar to bazel, you pass these options with --//... to bant:

bant -e --//bant:static_linked_executables print bant:bant

print grep output

Workspace

workspace prints all the external projects found in the workspace.

bant workspace   # All external projects referenced in this workspace

If you give a pattern, it will only print external projects used by targets matching the pattern:

bant workspace ...        # Print projects referenced in your project
bant workspace @re2//...  # Print projects referenced by re2

list-targets

If you want to find the file quickly, bant list-target //foo/bar:baz will output the filename and exact line/column range where the target resides.

bant list-targets //...     # list all targets of current project
bant list-targets -r //...  # also recursively following all dependencies

lib-headers

Use lib-headers if you want to know the library to depend on for a particular header file. Or, automatically update your BUILD files with dwyu...

dwyu : Depend-on What You Use

dwyu Depend on What You Use (DWYU)1: Determine which dependencies are needed in cc_library(), cc_binary(), and cc_test() targets. Greps through their declared sources to find which headers they include. Uses the information from lib-headers to determine which libraries these sources thus need to depend on. Emit buildozer commands to 'add' or 'remove' dependencies. If unclear if a library can be removed, it is conservatively not suggested for removal.

You can use this to clean up existing builds, or keep your BUILD files up-to-date in development. I usually just don't even add deps = [...] manually anymore but just let bant dwyu do the work.

The following auto-fixes all dependencies of the project:

. <(bant dwyu ...)   # source the buildozer calls in shell
Features

If there is a dependency that bant would like to remove, but it is needed for some reason that can't be derived from the include files it provides (and you can't correctly mark it as alwayslink because it is coming from some external project for instance), you can add a comment that contains the word 'keep' in the line in question.

cc_library(
  deps = [
    ":foo"  # build_cleaner keep
  ]
)

Note, a well-defined project should not need this. This features provides an escape hatch if needed. With -k, bant will be strict and ignore # keep comments and emit the removal edit anyway.

Caveats
  • Does not understand package groups in visibility yet; these will be considered //visibility:public and might result in adding dependencies that bazel would not allow.
  • Will remove dependencies if they provide headers that are not needed by the sources of a target. If you want to keep them linked, you need to declare them as alwayslink (libraries that are linked to targets but do not provide any headerss are considered alwayslink implicitly). (this is not really a caveat, it just emphasizes that it is important to properly declare the intent in BUILD files).

The dwyu command is essentially a build_cleaner for C++ targets.

compilation-db

Creates a compilation db that can be used by clang-tidy or clangd. Experimental right now. It is not exactly emitting all the options bazel would use, and is missing virtual includes (provided by cc_library() with include_prefix) so less accurate than tools that make use of bazel actions.

  bant compilation-db > compile_commands.json

Canonicalize

canonicalize emits edits to canonicalize targets, e.g.

  • //foo/bar:baz when already in package //foo/bar becomes :baz
  • //foo:foo becomes //foo
  • @foo//:foo becomes @foo
  • foo without : prefix becomes :foo

Use

Note, bant can only find external projects if bazel has set up the workspace, and fetched, unpacked, and made visible these into bazel-out/../../../external.

Bant never does any fetching, it just observes the existing workspace. This means you need to run a bazel build ... before to have the workspace set up.

At the very minimum, do a bazel fetch and run all genrules; we can use bant itself to find genrules to be passed to bazel:

bazel fetch ...
bazel build $(bant genrule-outputs ... | awk '{print $2}' | sort | uniq)

(but of course there might be other bazel rules beside obvious genrules that create artifacts, so global bazel build ... will catch these as well)

Given that bazel adapts the visible external projecgts depending on what targets are used, it might be worthwhile running a bazel build that needs all of them, e.g. bazel test ...

With bant workspace, you can see what external projects bant is aware of.

In continuous integration

Examples how to use bant in GitHub CI you find for these projects that use bant already

Right now, these just report with the exit code of dwyu, that changes are needed. Nice-to-have would be an integration that sends actionable diffs right into PR comments. And in general a nicer action integration. PRs welcome.

Slow file system OS pre-warm

If you work on a slow network file system or operate on some cold storage in a CI, it might take some time for bant to follow directories in globbing patterns. It might be beneficial to pre-warm the OS file system cache with accesses bant remembers from last time it ran. If a ~/.cache/bant/ directory exists, bant will make use it for this purpose (if you're on a fast SSD, no need for it).

Synopsis

bant v0.1.x <http://bant.build/>
Copyright (c) 2024 Henner Zeller. This program is free software; license GPL 2.0.
Usage: bazel-bin/bant/bant [options] <command> [bazel-target-pattern...]
Options
    -C <directory> : Change to this project directory first (default = '.')
    -q             : Quiet: don't print info messages to stderr.
    -o <filename>  : Instead of stdout, emit command primary output to file.
    -f <format>    : Output format, support depends on command. One of
                   : native (default), s-expr, plist, json, csv
                     Unique prefix ok, so -fs , -fp, -fj or -fc is sufficient.
    -r             : Follow dependencies recursively starting from pattern.
                     Without parameter, follows dependencies to the end.
                     An optional parameter allows to limit the nesting depth,
                     e.g. -r2 just follows two levels after the toplevel
                     pattern. -r0 is equivalent to not providing -r.
    -v             : Verbose; print some stats. Multiple times: more verbose.
    -h             : This help.
    --//<option>   : configurable flag attribute to be used in select() and
                     picked up by elaboration (-e) (experimental; does not yet
                     read config_setting(), but flag value is used directly).

Commands (unique prefix sufficient):
    == Parsing ==
    print          : Print AST matching pattern. -E : only files w/ parse errors
                      -e : elaborate; light eval: expand variables, concat etc.
                      -g <regex> : 'grep' - only print targets where any string
                                    matches regex.
                      -i If '-g' is given: case insensitive
    parse          : Parse all BUILD files from pattern. Follow deps with -r
                     Emit parse errors. Silent otherwise: No news are good news.
                      -v : some stats.

    == Extract facts == (Use -f to choose output format) ==
    workspace      : Print external projects found in WORKSPACE.
                     Without pattern: All external projects.
                     With pattern   : Subset referenced by matching targets.
                     → 3 column table: (project, version, path)

    -- Given '-r', the following also follow dependencies recursively --
    list-packages  : List all BUILD files and the package they define
                     → 2 column table: (buildfile, package)
    list-targets   : List BUILD file locations of rules with matching targets
                     → 3 column table: (buildfile:location, ruletype, target)
    aliased-by     : List targets and the various aliases pointing to it.
                     → 2 column table: (actual, alias*)
    depends-on     : List cc library targets and the libraries they depend on
                     → 2 column table: (target, dependency*)
    has-dependent  : List cc library targets and the libraries that depend on it
                     → 2 column table: (target, dependent*)
    lib-headers    : Print headers provided by cc_library()s matching pattern.
                     → 2 column table: (header-filename, cc-library-target)
    genrule-outputs: Print generated files by genrule()s matching pattern.
                     → 2 column table: (filename, genrule-target)

    == Tools ==
    dwyu           : DWYU: Depend on What You Use (emit buildozer edit script)
                      -k strict: emit remove even if # keep comment in line.
    canonicalize   : Emit rename edits to canonicalize targets.
    compile-flags  : (experimental) Emit compile flags to stdout. Redirect to
                     compile_flags.txt
    compilation-db : (experimental) Emit compilation db to stdout. Redirect to
                     compile_commands.json

Usage examples

 bant parse -C ~/src/verible -v  # Read project in given directory.
 bant print ... -g scan.*cc   # print out all targets that contain given regex
 bant print @googletest//:gtest -e  # parse and expression-eval given target
                                    # (test without -e to see the difference)
 bant print //foo:bar   # Print specific target AST matching pattern
 bant print //foo/...   # Print all build files matching recursive pattern.
 bant workspace         # List all the external projects listed in workspace.
 bant list-packages -r  # List all the build files, follow dependencies
 bant list-targets ...  # List all targets in this project
 bant list-targets ... | grep cc_binary   # find all binaries build by project
 bant lib-headers       # For each header found in project, print exporting lib
 bant dwyu ...         # Look which headers are used and suggest add/remove deps
 bant print bant/tool:*_test  # Print all targets ending with _test
 . <(bant dwyu foo/...)  # YOLO oneliner: build_clean deps in package foo/...
                         # by sourcing the emitted buildozer edit script.

Compile/Installation

To compile

bazel build -c opt //bant

Resulting binary will be bazel-bin/bant/bant

To install, use your systems install command (or simply copy):

# To some writable directory that does not require root access
bazel build -c opt //bant && install -D --strip bazel-bin/bant/bant ~/bin/bant

# For a system directory that requires root-access
sudo install -D --strip bazel-bin/bant/bant /usr/local/bin/bant

Development

Environment

Compiled using bazel >= 6. Relevant dependencies are already in the shell.nix so you can set up your environment with that automatically.

To get a useful compilation database for clangd to be happy, run first

scripts/make-compilation-db.sh

Before submit, run scripts/before-submit.sh ... and fix potential clang-tidy issues (or update .clang-tidy if it is not useful).

Current status

Goals

  • Goal: Reading bazel-like BUILD files and doing useful things with content.
  • Non-Goal: parse full starlark (e.g. *.bzl files with rule def-initions)

Status

  • Parses BUILD/BUILD.bazel files correctly (tested for various projects), and builds an AST used by all bant features.
  • Some evaluation, like variable expansion, list and string concatenation and glob() calls. No list comprehension yet.
  • Very useful in daily life to navigate around a project (very useful bant print with -g and/or -e). As well as keeping projects clean with bant dwyu which reliably adds neccessary dependencies but also reliably don't mess them up.
  • Outputs in the list-* type features provide a neat interface for scripting.

Nice-to-have/next steps/TODO

  • expand list-comprehensions
  • language server for BUILD files.
  • ...

Footnotes

  1. Build-dependency analog to what Include What You Use is for source files.