anyhow integration for gix-error

[Byron]

Make it possible to follow gix-error types with source() chains, and
provide a conversion to anyhow (beyind a feature toggle).

Tasks

• anyhow conversion from Exn
• source-chain for Exn
  • make it possible to keep the location.
• source-chain for Error
• A way to turn on auto-source-chain for Error so gix errors will be compatible automatically

[Copilot]

Pull request overview

This pull request adds anyhow integration to gix-error and implements error source chains for the Exn and Error types. The main enhancement is the addition of an auto-chain-error feature flag that automatically flattens error trees into chains, making them compatible with anyhow and other error-handling libraries.

Changes:

• Added new concrete error types (Message, ParseError, ChainedError) in gix-error
• Implemented std::error::Error::source() chains for Exn and Error types
• Added auto-chain-error feature toggle with tree-error override, making auto-chain-error the default
• Refactored message!() macro usage throughout the codebase to use format string capture syntax
• Added anyhow conversion support via the anyhow feature flag

Reviewed changes

Copilot reviewed 24 out of 27 changed files in this pull request and generated 16 comments.

Show a summary per file

File	Description
justfile	Added test execution for auto-chain-error feature
gix/src/revision/spec/parse/delegate/*.rs	Updated message! macro calls to use format string capture syntax
gix/Cargo.toml	Added auto-chain-error to default features, tree-error to dev features
gix-error/tests/error/*.rs	Updated tests to work with Message type and new error APIs
gix-error/tests/auto_chain_error.rs	New test file validating auto-chain-error behavior
gix-error/src/lib.rs	Added feature documentation, write_location helper, and concrete error module
gix-error/src/exn/impls.rs	Implemented anyhow conversion, ChainedError conversion, and renamed iter() to iter_frames()
gix-error/src/error.rs	Split implementation based on feature flags for tree vs chain error handling
gix-error/src/concrete/*.rs	New concrete error types (Message, ParseError, ChainedError)
gix-error/Cargo.toml	Added feature flags and dependencies for anyhow integration
gix-commitgraph/src/*.rs	Updated error message formatting to use capture syntax
gitoxide-core/src/*.rs	Removed manual error conversions now handled automatically
gitoxide-core/Cargo.toml	Added gix-error dependency with anyhow feature
Cargo.lock	Updated dependency graph

[EliahKagan]

The test-32bit test failures (on arm32v7) were:

        FAIL [   0.003s] (1196/2983) gix-error::auto-chain-error from_exn_error
  stdout ───

    running 1 test
    test from_exn_error ... FAILED

    failures:

    failures:
        from_exn_error

    test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 2 filtered out; finished in 0.00s
    
  stderr ───

    thread 'from_exn_error' (19568) panicked at gix-error/tests/auto_chain_error.rs:7:5:
    assertion `left == right` failed
      left: "Message(\"one\")"
     right: "one"
    note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

        FAIL [   0.003s] (1197/2983) gix-error::auto-chain-error from_exn_error_tree
  stdout ───

    running 1 test
    test from_exn_error_tree ... FAILED

    failures:

    failures:
        from_exn_error_tree

    test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 2 filtered out; finished in 0.00s
    
  stderr ───

    thread 'from_exn_error_tree' (19570) panicked at gix-error/tests/auto_chain_error.rs:20:5:
    assertion `left == right` failed
      left: "Message(\"topmost\")\n|\n└─ Message(\"E6\")\n    |\n    └─ Message(\"E5\")\n    |   |\n    |   └─ Message(\"E3\")\n    |   |   |\n    |   |   └─ Message(\"E1\")\n    |   |\n    |   └─ Message(\"E10\")\n    |   |   |\n    |   |   └─ Message(\"E9\")\n    |   |\n    |   └─ Message(\"E12\")\n    |       |\n    |       └─ Message(\"E11\")\n    |\n    └─ Message(\"E4\")\n    |   |\n    |   └─ Message(\"E2\")\n    |\n    └─ Message(\"E8\")\n        |\n        └─ Message(\"E7\")"
     right: "topmost"
    note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

Given that no tests fail in the test-fast jobs that have completed so far, I don't know why this would happen. The error seems somehow specific to 32-bit targets, but I don't know why that would be.

[Byron]

I don't think it is specific to 32bits, I hope it will resolve itself. In any case, I will see tomorrow 😅.

[EliahKagan]

Sounds good! It seemed weird enough that I figured I'd mention it. (Also, sorry about not saying what commit, fd8c038, those CI results were from--a new commit was pushed around the same time as I posted that comment.)

But actually it wasn't as weird as I thought:

• In ci.yml, it looks like some relevant tests were excluded here in the test-fast job. That might be sufficient to explain the combination of passing and failing jobs. If not...

• The failures seem like the usually happened when GIX_TEST_IGNORE_ARCHIVES=1 is set: the text-fixtures-windows job also failed, in the same way, it just hadn't finished yet when I commented. This is set in test-32bit, too, but not in test-fast. This is to say that, other than the "full" test job, they always happened in the jobs that rerun all fixture scripts. I mention this because, if that turns out to be relevant, then it suggest there may be less than the intended amout of feature coverage--for example, some missing nextest commands--in the justfile recipes used by the test job.

In any case, more pass since the subsequent commit a435a29.

Edit: Those two tests still fail in test-fixtures-windows though.

[Byron]

Thanks @EliahKagan. I have to confess that I did something quite messy: there is gix-error with a positive and a negative feature toggle. The positive one turns on a feature which disturbs gix-error's own tests which expect it to be off. cargo nextest doesn't handle feature resolution correctly by building everything at once seemingly, so the normal nextest run now needs gix-error barred from running, whenever it is called.

Maybe this knowledge should be put into the justfile.

The origin of this issue is cargo test -p gix which must disable default features in gix, but instead now has to use that negative feature toggle to do it.

Maybe a better way to fix this is to change the gix-error tests to expect either this or that depending on how they are run, and, I think that would be a good follow-up. Keeping note of that.