Don’t use .gitignore of a different git repository

[SimonSapin]

In a clone of https://github.com/rust-lang/rust with git submodules checked out, in the src/llvm submodule, ripgrep 0.8.1 seems to "miss" several results that are found by GNU grep or git-grep:

~/rust/src/llvm% rg __powisf2                   
lib/Target/WebAssembly/WebAssemblyRuntimeLibcallSignatures.cpp
643:/* POWI_F32 */ "__powisf2",

~/rust/src/llvm% grep -R __powisf2            
include/llvm/CodeGen/RuntimeLibcalls.def:HANDLE_LIBCALL(POWI_F32, "__powisf2")
test/CodeGen/Mips/msa/f16-llvm-ir.ll:; MIPS32-DAG:     lw $25, %call16(__powisf2)($gp)
test/CodeGen/Mips/msa/f16-llvm-ir.ll:; MIPS64-N32-DAG: lw $25, %call16(__powisf2)($gp)
test/CodeGen/Mips/msa/f16-llvm-ir.ll:; MIPS64-N64-DAG: ld $25, %call16(__powisf2)($gp)
test/CodeGen/Mips/pr36061.ll:; MIPSN64-NEXT:    jal __powisf2
test/CodeGen/Mips/pr36061.ll:; MIPSN32-NEXT:    jal __powisf2
test/CodeGen/AArch64/illegal-float-ops.ll:; CHECK: bl __powisf2
test/CodeGen/AArch64/f16-instructions.ll:; CHECK-COMMON-NEXT: bl {{_?}}__powisf2
test/CodeGen/AArch64/arm64-illegal-float-ops.ll:; CHECK: bl __powisf2
test/CodeGen/SystemZ/fp-libcall.ll:; CHECK: brasl %r14, __powisf2@PLT
test/CodeGen/Thumb2/intrinsics-cc.ll:; CHECK: b __powisf2
test/CodeGen/Thumb2/float-intrinsics-float.ll:; SOFT: bl __powisf2
test/CodeGen/Thumb2/float-intrinsics-float.ll:; HARD: b __powisf2
test/CodeGen/ARM/Windows/powi.ll:; CHECK-NOT: __powisf2
test/CodeGen/ARM/Windows/powi.ll:; CHECK-NOT: __powisf2
test/CodeGen/ARM/Windows/powi.ll:; CHECK-NOT: bl __powisf2
test/CodeGen/ARM/fp16-promote.ll:; CHECK-FP16: bl __powisf2
test/CodeGen/ARM/fp16-promote.ll:; CHECK-LIBCALL: bl __powisf2
test/CodeGen/XCore/float-intrinsics.ll:; CHECK: bl __powisf2
lib/Target/WebAssembly/WebAssemblyRuntimeLibcallSignatures.cpp:/* POWI_F32 */ "__powisf2",

This turns out to be because of the .gitignore file at the root of the Rust repository. Running with --no-ignore-parent makes it find those extra results.

These files are not ignored by git. If I modify ~/rust/src/llvm/include/llvm/CodeGen/RuntimeLibcalls.def for example, git status will show it as modified. My understanding is that ~/rust/.gitignore is not relevant to RuntimeLibcalls.def because it belongs to a different repository. This is detected by the presence of a .git directory (or file) at ~/rust/src/llvm/.git.

When going up the parent/ancestor directories, ripgrep should probably stop considering .gitignore files after it goes beyond the root of a git repository (as indicated by .git). Conversely, when descending into a directory that contains .git, it should maybe stop considering .gitignore files that are "above" this directory.

[BurntSushi]

Ah this is a great bug! ripgrep does actually know to stop respecting parent .gitignore once it crosses a repository boundary. The problem is that it's looking for a .git directory:

ripgrep/ignore/src/dir.rs

Line 257 in 6b15ce2

has_git: dir.join(".git").is_dir(),

But it looks like .git can also be a file, which ripgrep should obviously respect. I think fixing that should completely fix this.

[SimonSapin]

Indeed, it looks like .git is a file for submodules too. I hadn’t realized that.