paths wildcard validation seems off

[bplessis-swi]

Want to help get this issue implemented? Donate to the action-validator code fund.

---

Hi,

Sorry if i missed something but when following the "filter pattern cheat cheet" of github action, the validator fail the file with the following message:
Glob "**.md" in on.push.paths is invalid: Pattern syntax error near position 2: recursive wildcards must form a single path component

Documentation source

https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#patterns-to-match-file-paths

Sample use

on:
  push:
    branches:
      - main
    paths:
      - '**.md'

Version

action-validator 0.3.0

[mpalmer]

Thanks for this solid report. I'd say this comes down to a difference of opinion between the globbing library that action-validator uses, and what GitHub specifies. Might not be trivial to solve, but I'll get it figured out one way or another.

[bplessis-swi]

Yeah the rust glob() function seem more "limited" in it's patterns, the equivalent of '**.md' seem to be '**/*.md' (that apparently github also understand apparently, given the 'docs/**/*.md' sample a few lines down in the same documentation.

[jhberges]

FWIW: The tool will also error on exclusion patterns.

Example:

on:
  push:
    branches:
      - main
    paths:
      - '!generated/**'

Adding the extra /* (that is '!generated/**/*') does not work around this.

I like the idea though :-)

[kevinlul]

I encountered this behaviour as well: https://github.com/DawnbrandBots/yaml-yugi/actions/runs/5626963939/job/15248759792

Claims the directory/** pattern is invalid, but these workflows trigger and run just fine:

• https://github.com/DawnbrandBots/yaml-yugi/blob/f55d872968ec5b400fa29a88aef2dd5865e97302/.github/workflows/validate-assignments.yaml#L5-L14
• https://github.com/DawnbrandBots/yaml-yugi/blob/f55d872968ec5b400fa29a88aef2dd5865e97302/.github/workflows/validate-data.yaml#L16

[vitalii-buchyn-exa]

Hey guys,

Experiencing a similar issue with somedir/** path.

Error example:

errors: [
        NoFilesMatchingGlob {
            code: "glob_not_matched",
            detail: Some(
                "Glob \"terraform/dev/**\" in /on/pull_request/paths does not match any files",
            ),
            path: "/on/pull_request/paths",
            title: "Glob does not match any files",
        },
    ],

Any workaround for that?
Thanks.

action-validator version 0.5.4
installed using asdf-vm/actions/install@v3

[mpalmer]

No workaround available, action-validator needs to be modified to use a different glob expansion mechanism in order to support these patterns. I don't have the available time to fix it myself, so as the tag says, "PR welcome"!

[DilepDev]

Experiencing a similar issue for below ci.yml

name: CI Workflow

on:
  push:
    paths-ignore:
      - '.gitattributes'
      - '**.MD'
      - '.devcontainer/**'

Treating ci.yml as a Workflow definition
Fatal error validating .github/workflows/ci.yml
Validation failed: ValidationState {
    action_type: Some(
        Workflow,
    ),
    file_path: Some(
        ".github/workflows/ci.yml",
    ),
    errors: [
        InvalidGlob {
            code: "invalid_glob",
            detail: Some(
                "Glob \"**.MD\" in /on/push/paths-ignore is invalid: Pattern syntax error near position 2: recursive wildcards must form a single path component",
            ),
            path: "/on/push/paths-ignore",
            title: "Glob does not match any files",
        },
        NoFilesMatchingGlob {
            code: "glob_not_matched",
            detail: Some(
                "Glob \".devcontainer/**\" in /on/push/paths-ignore does not match any files",
            ),
            path: "/on/push/paths-ignore",
            title: "Glob does not match any files",
        },
    ],
}

But when .devcontainer/** is replaced with .github/** glob_not_matched error related to this disappears

name: CI Workflow

on:
  push:
    paths-ignore:
      - '.gitattributes'
      - '**.MD'
      - '.github/**'

Treating ci.yml as a Workflow definition
Fatal error validating .github/workflows/ci.yml
Validation failed: ValidationState {
    action_type: Some(
        Workflow,
    ),
    file_path: Some(
        ".github/workflows/ci.yml",
    ),
    errors: [
        InvalidGlob {
            code: "invalid_glob",
            detail: Some(
                "Glob \"**.MD\" in /on/push/paths-ignore is invalid: Pattern syntax error near position 2: recursive wildcards must form a single path component",
            ),
            path: "/on/push/paths-ignore",
            title: "Glob does not match any files",
        },
    ],
}

[dabrahams]

IMO the right thing to do is disable glob validation until it can be made to match what GitHub uses.

[ferdnyc]

No workaround available, action-validator needs to be modified to use a different glob expansion mechanism in order to support these patterns.

Indeed, the globbing package currently used (the creatively named "glob") is explicitly documented as offering:

$ cargo info glob
glob
Support for matching file paths against Unix shell style patterns.

Shell patterns don't treat ** as a "match anything" wildcard (a superset of *), which is what GitHub documents their ** as meaning.

A ** in the middle of a shell path (.../**/) will recursively expand to all directories at or below that location. A ** at the end of a path just becomes *. It NON-recursively matches everything at that level, but nothing below.

(In zsh...)

$ cd .../action-validator

$ ls src
config.rs  lib.rs   schemas.rs   system    validation_error.rs
js         main.rs  schemastore  utils.rs  validation_state.rs

$ ls -d src/*
src/config.rs  src/schemas.rs    src/validation_error.rs
src/js/        src/schemastore/  src/validation_state.rs
src/lib.rs     src/system/
src/main.rs    src/utils.rs

# Identical to previous (non-recursive)
$ ls -d src/**
src/config.rs  src/schemas.rs    src/validation_error.rs
src/js/        src/schemastore/  src/validation_state.rs
src/lib.rs     src/system/
src/main.rs    src/utils.rs

# Actually recursive
$ ls -d src/**/*
src/config.rs     src/schemas.rs         src/system/mod.rs
src/js/           src/schemastore/       src/system/process.rs
src/js/system.js  src/system/            src/utils.rs
src/lib.rs        src/system/console.rs  src/validation_error.rs
src/main.rs       src/system/fs.rs       src/validation_state.rs

(In bash, which interestingly doesn't allow **/* to match directories directly alongside the **; IOW, /a/b/**/* doesn't decay to include /a/b/* like in zsh, the ** has to match at least one directory level...)

$ cd .../action-validator

$ ls src
config.rs  lib.rs   schemas.rs   system    validation_error.rs
js         main.rs  schemastore  utils.rs  validation_state.rs

$ ls -d src/*
src/config.rs  src/schemas.rs   src/validation_error.rs
src/js         src/schemastore  src/validation_state.rs
src/lib.rs     src/system
src/main.rs    src/utils.rs

# Identical to previous (non-recursive)
$ ls -d src/**
src/config.rs  src/schemas.rs   src/validation_error.rs
src/js         src/schemastore  src/validation_state.rs
src/lib.rs     src/system
src/main.rs    src/utils.rs

# Actually recursive, but consumes a path level
$ ls -d src/**/*
src/js/system.js       src/system/fs.rs   src/system/process.rs
src/system/console.rs  src/system/mod.rs

I think the Rust glob package is still buggy, because for example glob validation will claim that a glob like src/system/** matches no files in this repo, when clearly even the shells think it does:

$ ls src/system/**
src/system/console.rs  src/system/mod.rs
src/system/fs.rs       src/system/process.rs

...But that's a moot point, because the package's goal is to replicate shell globbing. Even if it did that perfectly, GitHub's globbing is different and more flexible.