Bug: Multiple potential bugs in source code

[terasakisatoshi]

Summary

Code review identified 12 potential bugs across the RustCall.jl source code.

1. Nested generics regex failure in _parse_fn_arg_types (generics.jl:755)

File: src/generics.jl:755

fn_pattern = Regex("fn\\s+$(func_name)\\s*(?:<[^{]*?>)?\s*\(")

<[^{]*?> uses non-greedy matching which fails on deeply nested generics like Vec<Option<T>>. The > inside Option< would incorrectly match before the closing > of Vec<. This violates the project's own documented rule: "Never use regex alone to match angle brackets (< >) in Rust code."

2. Incorrect bracket depth tracking for [ and > (structs.jl:64-68, julia_functions.jl:206-215, ruststr.jl:1003-1031)

Files: src/structs.jl:64-68, src/julia_functions.jl:206-215, src/ruststr.jl:1003-1031

elseif c == '>'
    angle_depth = max(0, angle_depth - 1)

elseif c == ']'
    bracket_depth = max(0, bracket_depth - 1)

Using max(0, ...) for bracket depth clamping causes [-delimited brackets and >-delimited brackets to be tracked independently of each other. This means a < inside an array literal [a < b] would still increment angle_depth, and a > inside [a < b] would decrement it, leading to incorrect depth tracking when arrays contain comparison operators.

3. Unusual return value from drop! fallback (types.jl:837-843)

File: src/types.jl:837-843

function drop!(x::Union{RustBox, RustRc, RustArc, RustVec})
    lock(x.drop_lock) do
        x.dropped = true
        x.ptr = C_NULL
    end
    return nothing
end

Julia convention for !-suffixed mutation functions is to return nothing (not the tuple (nothing)). While this works due to implicit tuple return, it is non-idiomatic.

4. String literal brace counting causes false positives (exceptions.jl:559-565)

File: src/exceptions.jl:559-565

open_braces = count(c -> c == '{', source_code)
close_braces = count(c -> c == '}', source_code)
if open_braces > close_braces
    push!(suggestions, "Found $(open_braces - close_braces) more opening brace(s)...")

Brace counting includes braces inside string literals like s = "{test}", causing false positive suggestions for mismatched braces when none exist.

5. RustStr incorrectly mapped to Cstring in codegen (codegen.jl:125)

File: src/codegen.jl:125

julia_to_c_type(::Type{RustStr}) = Cstring

RustStr is a fat pointer (ptr + len) representing &str. Mapping it to Cstring (null-terminated string) loses the length information and causes ABI mismatch for FFI calls expecting &str.

6. Generic return type inference is oversimplified (generics.jl:973-1003)

File: src/generics.jl:973-1003

ret_type = length(type_params) > 0 ? first(values(type_params)) : Any

The first type parameter is assumed as the return type. For functions like fn transform<T, U>(x: T) -> U, this returns T instead of U, causing type mismatches.

7. readdir vs walkdir inconsistency in cache size calculation (cache.jl:459-464, cache.jl:485)

File: src/cache.jl:485

for file in readdir(cache_dir)
    if endswith(file, lib_ext)

list_cached_libraries uses readdir (non-recursive), while get_cache_size uses walkdir (recursive). This means .dylib files in subdirectories of the cache would be counted by get_cache_size but missed by list_cached_libraries.

8. _count_trailing_backslashes uses unsafe indexing (dependencies.jl:288)

File: src/dependencies.jl:288

for i in lastindex(s):-1:firstindex(s)
    s[i] == '\\' || break

Using linear index iteration on multi-byte strings is unsafe. Should use prevind-based iteration or iterate bytes instead.

9. lib_name / cache key mismatch (ruststr.jl:239-244, ruststr.jl:273-278)

File: src/ruststr.jl:239-244

code_hash = stable_content_hash(wrapped_code)[1:16]
lib_name = "rust_$(code_hash)"

The library registration key uses the first 16 hex chars of a stable content hash. The cache key uses a different computation path. If the compilation pipeline changes between cache save/load, the lookup could fail while the cache file exists.

10. found_libs duplicate potential in fallback search (ruststr.jl:86-106)

File: src/ruststr.jl:86-106

The fallback search in get_function_pointer doesn't guard against the same library being queried multiple times if RUST_LIBRARIES contains duplicate handles (through different keys pointing to the same library).

Expected Behavior

• Regex-based angle bracket matching should be replaced with bracket-counting in all Rust code parsing
• String literals must be skipped when counting braces/special characters
• RustStr should map to a fat pointer type (Cstruct with ptr + len), not Cstring
• Generic return type inference should parse the actual return type from function signature, not use first type param
• readdir calls in cache code should use walkdir for recursive search
• String indexing should use character-safe iteration with prevind/nextind

Related Files

• src/structs.jl
• src/julia_functions.jl
• src/ruststr.jl
• src/codegen.jl
• src/generics.jl
• src/cache.jl
• src/exceptions.jl
• src/dependencies.jl
• src/types.jl

[terasakisatoshi]

査定結果を整理しました。現時点では以下を confirmed と判断しています。

• Item 1: 指摘の再現例そのもの（Vec<Option<T>>）は再現しませんでしたが、同じ箇所の regex は generic parameter list に {...} を含むと壊れます。追跡 issue: Bug: _parse_fn_arg_types fails when generic parameter lists contain {...} #232
• Item 2: parse_julia_functions の深さ追跡は const expression 中の < / > を generic bracket と誤認し、引数分割と return type 解析を壊します。追跡 issue: Bug: parse_julia_functions misparses const expressions containing < / > #233
• Item 4: suggest_fix_for_error は文字列リテラル中の { / } まで数えて誤誘導します。追跡 issue: Bug: suggest_fix_for_error counts braces inside string literals #235
• Item 8: _count_trailing_backslashes は UTF-8 文字列で StringIndexError を投げます。追跡 issue: Bug: _count_trailing_backslashes throws StringIndexError on UTF-8 input #234

現時点で confirmed していないもの:

• Item 3: drop! は nothing を返しており、指摘内容は現状コードと一致しません
• Item 5: julia_to_c_type(::Type{RustStr}) = Cstring は整合性に欠けますが、現行呼び出し経路では未使用で、実バグまでは確認できませんでした
• Item 6: 指摘された -> U のケースは specialization 後に具体型へ置換されるため、その repro は確認できませんでした
• Item 7: cache library は top-level 保存で、list_cached_libraries と get_cache_size の役割差として説明可能でした
• Item 9: lib_name と cache key は別用途ですが、現行の save/load 経路で不整合は再現できませんでした
• Item 10: fallback search の duplicate handle 問題は現行の registry 更新経路では再現できませんでした

必要なら次は #232/#233/#234/#235 の修正に入れます。

[terasakisatoshi]

Confirmed items have been addressed locally on branch fix/issues-232-235 in commit f721364.

Handled issues:

• Bug: _parse_fn_arg_types fails when generic parameter lists contain {...} #232
• Bug: parse_julia_functions misparses const expressions containing < / > #233
• Bug: _count_trailing_backslashes throws StringIndexError on UTF-8 input #234
• Bug: suggest_fix_for_error counts braces inside string literals #235

Verification run after the fixes:

• julia --project test/test_generics.jl
• julia --project test/test_julia_attribute.jl
• julia --project test/test_dependencies.jl
• julia --project test/test_error_handling.jl
• `julia --project -e 'using Pkg; Pkg.test()'

[terasakisatoshi]

Opened draft PR #236 for the confirmed fixes: #236

[terasakisatoshi]

Confirmed bugs were fixed in merged PR #236, and the remaining items from the original audit were not reproduced as actionable bugs.