Skip to content

Keep blocks of code (more generally text) in sync across different files in your codebase.

Notifications You must be signed in to change notification settings

aksiksi/onchg-rs

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

95 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

onchg

Crates.io docs.rs codecov test

A tool that allows you to keep blocks in sync across different files in your codebase.

Install

- repo: https://github.com/aksiksi/onchg-rs
  rev: v0.1.6
  hooks:
    - id: onchg

CLI

cargo install onchg

Quickstart

Video

https://www.loom.com/share/4018aea2378f4e4e8fcd403a70749cde?sid=19f4c8ec-87b6-4eac-a448-f326695189ee

Setup

Create an empty directory:

mkdir -p /tmp/onchg/quickstart && cd /tmp/onchg/quickstart

Create two files - docs.md and header.h:

docs.md:

cat >docs.md <<EOL
# Docs

## Supported Services

<!--- LINT.OnChange(supported-services) --->
* Main
* Primary
* Other
<!--- LINT.ThenChange(header.h:supported-services) --->

EOL

header.h:

cat >header.h <<EOL

// LINT.OnChange(supported-services)
typedef enum {
    INVALID = 0,
    MAIN = 1,
    PRIMARY = 2,
    OTHER = 3,
} supported_services_t;
// LINT.ThenChange(docs.md:supported-services)

EOL

Initialize a Git repo and commit both files:

git init . && git add . && git commit -m "first commit"

pre-commit

Create a pre-commit config and install the hook:

cat >.pre-commit-config.yaml <<EOL
repos:
  - repo: https://github.com/aksiksi/onchg-rs
    rev: v0.1.6
    hooks:
      - id: onchg
EOL

pre-commit install

Change header.h:

--- a/header.h
+++ b/header.h
@@ -5,6 +5,7 @@ typedef enum {
     MAIN = 1,
     PRIMARY = 2,
     OTHER = 3,
+    NEW = 4,
 } supported_services_t;
 // LINT.ThenChange(docs.md:supported-services)

Stage and commit:

$ git add . && git commit -m "my commit"
onchg....................................................................Failed
- hook id: onchg
- exit code: 1

Root path: /home/aksiksi/onchg/quickstart

Parsed 2 files (2 blocks total):
  * /home/aksiksi/onchg/quickstart/docs.md
  * /home/aksiksi/onchg/quickstart/header.h

Violations:
  * block "supported-services" at /home/aksiksi/onchg/quickstart/docs.md:5 (due to block "supported-services" at /home/aksiksi/onchg/quickstart/header.h:2)

CLI

Run onchg on the directory:

$ onchg directory
Root path: /home/aksiksi/onchg/quickstart

Parsed 2 files (2 blocks total):
  * /home/aksiksi/onchg/quickstart/docs.md
  * /home/aksiksi/onchg/quickstart/header.h

OK.

Make a change to the enum in header.h:

--- a/header.h
+++ b/header.h
@@ -5,6 +5,7 @@ typedef enum {
     MAIN = 1,
     PRIMARY = 2,
     OTHER = 3,
+    NEW = 4,
 } supported_services_t;
 // LINT.ThenChange(docs.md:supported-services)

Stage the change & run onchg in repo mode:

$ git add header.h && onchg repo
Root path: /home/aksiksi/onchg/quickstart

Parsed 2 files (2 blocks total):
  * /home/aksiksi/onchg/quickstart/docs.md
  * /home/aksiksi/onchg/quickstart/header.h

Violations:
  * block "supported-services" at /home/aksiksi/onchg/quickstart/docs.md:5 (due to block "supported-services" at /home/aksiksi/onchg/quickstart/header.h:2)

Change docs.md:

--- a/docs.md
+++ b/docs.md
@@ -6,5 +6,6 @@
 * Main
 * Primary
 * Other
+* New
 <!-- LINT.ThenChange(header.h:supported-services) -->

Stage the change & re-run onchg:

$ git add docs.md && onchg repo
Root path: /home/aksiksi/onchg/quickstart

Parsed 2 files (2 blocks total):
  * /home/aksiksi/onchg/quickstart/docs.md
  * /home/aksiksi/onchg/quickstart/header.h

OK.

Documentation

Examples

Two-way Dependency

alpha.txt:

OnChange(my-block)

ThenChange(beta.txt:their-block)

beta.txt:

OnChange(their-block)

ThenChange(alpha.txt:my-block)

Relative Paths

alpha.txt:

OnChange(my-block)

ThenChange(subdir/beta.txt:their-block)

subdir/beta.txt:

OnChange(their-block)

ThenChange(../alpha.txt:my-block)

Root Paths

alpha.txt:

OnChange(my-block)

ThenChange(subdir/beta.txt:their-block)

subdir/beta.txt:

OnChange(their-block)

ThenChange(//alpha.txt:my-block)

One-way OnChange and ThenChange

alpha.txt:

OnChange()

ThenChange(beta.txt:their-block)

beta.txt:

OnChange(their-block)

ThenChange()

Multiple Dependencies

alpha.txt:

OnChange(my-block)

ThenChange(beta.txt:their-block, gamma.txt:another)

beta.txt:

OnChange(their-block)

ThenChange()

gamma.txt:

OnChange(another)

ThenChange(alpha.txt:my-block, beta.txt:their-block)

Nested Blocks

alpha.txt:

OnChange(my-block)

OnChange(inner-block)

ThenChange(beta.txt:their-block)

ThenChange()

beta.txt:

OnChange(their-block)

ThenChange()

Details

onchg uses blocks to capture depdendencies between sections of code (or more generally text) across different files.

A block looks like this:

OnChange( [name] )

ThenChange( [<target>[, ...]] )

The OnChange and ThenChange sections can exist anywhere on a line. This allows you to place the sections inside any type of code comment.

OnChange accepts an optional name. If a block does not specify a name, it cannot be used as a target by other blocks. This is useful in cases where you want one-way dependencies - i.e., if this block changes, other blocks should change, but not vice-versa.

ThenChange accepts zero or more targets. A block target has the following syntax:

[file][:[block]]

Just like OnChange, ThenChange allows for one-way dependencies if the target list is empty.

If a target is specified, it can either be a file or a block in a file. The block is just the block name. The file path must be one of the following:

  1. Relative: The path is relative to the current file's path (e.g., abc/hello.txt).
  2. Relative to the root: The path starts with // to indicate that the path is relative to the root directory. This is the path you specify when running onchg. Typically, the root would be the Git repo root.

Benchmarks

Synthetic

Setup:

  • OS: Ubuntu 22.04 VM
  • CPU: 10-core AMD 3900x equivalent (virtualized)
  • Disk: Corsair Force MP510 PCIe Gen3 NVMe drive

150 and 1000 in the bench names refer to the number of files analyzed.

When compared to grep, in addition to finding matches in all files, onchg needs to:

  1. Load the state of all blocks into memory.
  2. Parse and extract the capture group content to ensure that blocks are valid.
  3. Run a validation step across all parsed blocks.

Note

All benchmarks are seeded to allow for reproducibility.

Sparse

Note

This is the more realistic benchmark.

0-10 blocks per file with up to 100 lines per block. Each line varies from 0-100 characters long.

~2x slower than grep:

directory-sparse/150    time:   [2.5643 ms 2.5974 ms 2.6338 ms]
grep-sparse/150         time:   [1.6161 ms 1.6241 ms 1.6328 ms]
ripgrep-sparse/150      time:   [4.9077 ms 4.9354 ms 4.9640 ms]

directory-sparse/1000   time:   [15.186 ms 15.271 ms 15.359 ms]
grep-sparse/1000        time:   [6.4750 ms 6.5380 ms 6.6048 ms]
ripgrep-sparse/1000     time:   [7.6132 ms 7.6550 ms 7.6980 ms]

Dense

Note

This is more of a pathological worst-case benchmark.

50-100 blocks per file. Same line count and line length settings as sparse bench.

5-6x slower than grep:

directory-dense/150     time:   [11.388 ms 11.469 ms 11.554 ms]
grep-dense/150          time:   [3.1692 ms 3.1907 ms 3.2138 ms]
ripgrep-dense/150       time:   [6.7027 ms 6.7731 ms 6.8621 ms]

directory-dense/1000    time:   [83.987 ms 84.581 ms 85.224 ms]
grep-dense/1000         time:   [15.269 ms 15.349 ms 15.430 ms]
ripgrep-dense/1000      time:   [15.800 ms 15.901 ms 16.004 ms]

Git Repo

Note

This is also a pathological worst-case benchmark. Blocks randomly depend on other blocks and the graph is large even with relatively few changed blocks.

This is the same as the dense bench above, but we instead randomly modify 200 blocks, stage them, and run onchg repo. The bench ends up parsing ~9500 files; the degree of block/file connectivity is quite high as noted above. Note that the bench generates a total of ~75000 blocks across 1000 files.

git-repo/200            time:   [339.91 ms 340.70 ms 341.53 ms]
Why so much slower??

Two reasons:

  1. The file walk is single-threaded.
  2. It takes a whopping 250ms just to render the staged diff to stdout!

One interesting finding: when using libgit2 via the git feature, the bench takes ~250ms longer. After digging into it a bit, it seems that the source of the delay is the diff line iterator in libgit2. Somehow, it takes 2x longer than rendering the diff to stdout and parsing it!

Real Codebases

Linux

Warning

This clones the full Linux kernel tree (~2GB) to the current working directory.

$ ./benches/linux.sh
Number of lines in Linux kernel: 38566988
Root path: /home/aksiksi/repos/onchg/linux

Parsed 82259 files (2 blocks total):
  * /home/aksiksi/repos/onchg/linux/.clang-format
  * /home/aksiksi/repos/onchg/linux/.cocciconfig
  * /home/aksiksi/repos/onchg/linux/.get_maintainer.ignore
  * /home/aksiksi/repos/onchg/linux/.gitattributes
  * /home/aksiksi/repos/onchg/linux/.gitignore
  * /home/aksiksi/repos/onchg/linux/.mailmap
  * /home/aksiksi/repos/onchg/linux/.rustfmt.toml
  * /home/aksiksi/repos/onchg/linux/COPYING
  * /home/aksiksi/repos/onchg/linux/CREDITS
  * /home/aksiksi/repos/onchg/linux/Documentation/.gitignore
  * /home/aksiksi/repos/onchg/linux/Documentation/ABI/README
  * /home/aksiksi/repos/onchg/linux/Documentation/ABI/obsolete/o2cb
  * /home/aksiksi/repos/onchg/linux/Documentation/ABI/obsolete/procfs-i8k
  * /home/aksiksi/repos/onchg/linux/Documentation/ABI/obsolete/sysfs-bus-iio
  * /home/aksiksi/repos/onchg/linux/Documentation/ABI/obsolete/sysfs-bus-usb
  ... 82244 files omitted

OK.

real  0m0.628s
user  0m0.779s
sys   0m0.974s

About

Keep blocks of code (more generally text) in sync across different files in your codebase.

Resources

Stars

Watchers

Forks

Packages

No packages published

Languages