Hi - I am new to the logos ecosystem.
I don't want to cast any doubt on the correctness of logos or that anyone should choose to use it - it looks like its a remarkable project that is clearly a great choice for many users.
I am upstreaming a crate which uses logos to generate a lexer ( bytecodealliance/wasmtime#8872 ) into wasmtime. As part of accepting logos to be used as a transitive dependency in the wasmtime project, I need to certify that it meets cargo-vet's safe-to-deploy criteria: https://mozilla.github.io/cargo-vet/built-in-criteria.html#safe-to-deploy .
The lexer will be handling untrusted input, and after spending some time examining the way logos codegen works, I don't feel that I can certify that the use of unsafe rust is sound: while I don't have any evidence that it is unsound, the code generator, and the code it generates, is too complex for me to reasonably declare that any use of the logos derive macro is fully sound, as would be implied by marking it as safe-to-deploy.
My rough understanding, from reading the code generator and the author's blog, is that performance is a huge goal of logos, and it has achieved very high performance. In my use case, I don't really care about performance: the inputs to the lexer are small and infrequent, and if the lexer was one or two orders of magnitude slower, that would be fine. However, I do care about correctness, to the point of being very conservative in what dependencies I can accept.
One path forward might be to have an alternative code generator for the logos macros that uses entirely safe rust. Have the logos authors ever considered this approach? Otherwise, I will have to rewrite my lexer by hand, and in doing so I will lose composability with other logos lexers in our ecosystem.
Hi - I am new to the
logosecosystem.I don't want to cast any doubt on the correctness of logos or that anyone should choose to use it - it looks like its a remarkable project that is clearly a great choice for many users.
I am upstreaming a crate which uses logos to generate a lexer ( bytecodealliance/wasmtime#8872 ) into wasmtime. As part of accepting
logosto be used as a transitive dependency in the wasmtime project, I need to certify that it meetscargo-vet'ssafe-to-deploycriteria: https://mozilla.github.io/cargo-vet/built-in-criteria.html#safe-to-deploy .The lexer will be handling untrusted input, and after spending some time examining the way logos codegen works, I don't feel that I can certify that the use of
unsaferust is sound: while I don't have any evidence that it is unsound, the code generator, and the code it generates, is too complex for me to reasonably declare that any use of thelogosderive macro is fully sound, as would be implied by marking it assafe-to-deploy.My rough understanding, from reading the code generator and the author's blog, is that performance is a huge goal of
logos, and it has achieved very high performance. In my use case, I don't really care about performance: the inputs to the lexer are small and infrequent, and if the lexer was one or two orders of magnitude slower, that would be fine. However, I do care about correctness, to the point of being very conservative in what dependencies I can accept.One path forward might be to have an alternative code generator for the logos macros that uses entirely safe rust. Have the logos authors ever considered this approach? Otherwise, I will have to rewrite my lexer by hand, and in doing so I will lose composability with other logos lexers in our ecosystem.