convert_error() assumes VerboseError saved inputs are substrings of the same overall input

[douglas-raillard-arm]

Prerequisites

• Rust version : rustc -V: rustc 1.66.1 (90743e729 2023-01-10)
• nom version :

  [[package]]
  name = "nom"
  version = "7.1.2"
  source = "registry+https://github.com/rust-lang/crates.io-index"
  checksum = "e5507769c4919c998e69e49c839d9dc6e693ede4cc4290d6ad8b41d4f09c548c"

• nom compilation features used: none

Test case

I'm currently using an input based on &[u8] (coming from a binary file) but I want to display the VerboseError as a string since it's actually working on source code snippets. Consequently, I mapped over VerboseError{ error } member to convert the input of each level to a String before handing it to convert_error().

Problem: convert_error() assumes that the saved input in VerboseError { error } is simply a pointer inside the overall input, but in this case it's not at all true (overall input got converted to String, and so is every saved input in the stack). This results in a panic!() as the offset computation results in a negative usize:

   3: <str as nom::traits::Offset>::offset
             at /home/dourai01/.cargo/registry/src/github.com-1ecc6299db9ec823/nom-7.1.2/src/traits.rs:76:5
   4: nom::error::convert_error
             at /home/dourai01/.cargo/registry/src/github.com-1ecc6299db9ec823/nom-7.1.2/src/error.rs:261:18
   5: traceevent::parser::tests::test_parser
             at ./src/parser.rs:286:27
   6: traceevent::cparser::tests::declaration_test::test
             at ./src/cparser.rs:1683:13

Would it be possible to use a content-based offset detection rather than assuming all pointers point inside the same object ? It may be possible to do that as a fallback while the fast path still does pointer arithmetic, since it should be easy to check if the substring ptr is within the address range of the overall input.

[douglas-raillard-arm]

As a workaround, I've used std::str::from_utf8(buf).unwrap() but I was initially hoping to use std::string::String::from_utf8_lossy(buf).to_string()