--query-driver validation check does not understand shorthands like PROGRA~1

[ReasonJinMing]

In Windows 10, It looks like clangd can't cope with spaces or some special symbol in the --query-driver path for system include extraction. I use mingw in windows, and the path to mingw is set under the Program Files folder. Here is a comparison, the gcc paths are the same, but one has spaces and the other does not.
I hope clangd can do the path replacement automatically in windows.

Logs
--query-driver with spaces clangd report System include extraction: not allowed driver

C:\Program Files\clangd_15.0.6\bin>clangd --check=C:\Users\lenovo\helloc\main.c --query-driver="C:\Program Files\RedPanda-Cpp\MinGW64\bin\gcc.exe" --log=verbose
I[09:52:18.655] clangd version 15.0.6 (https://github.com/llvm/llvm-project 088f33605d8a61ff519c580a71b1dd57d16a03f8)
I[09:52:18.657] Features: windows+grpc
I[09:52:18.658] PID: 6416
I[09:52:18.659] Working directory: C:\Program Files\clangd_15.0.6\bin
I[09:52:18.660] argv[0]: clangd
I[09:52:18.661] argv[1]: --check=C:\Users\lenovo\helloc\main.c
I[09:52:18.661] argv[2]: --query-driver=C:\Program Files\RedPanda-Cpp\MinGW64\bin\gcc.exe
I[09:52:18.662] argv[3]: --log=verbose
V[09:52:18.670] User config file is C:\Users\lenovo\AppData\Local\clangd\config.yaml
I[09:52:18.672] Entering check mode (no LSP server)
I[09:52:18.672] Testing on source file C:\Users\lenovo\helloc\main.c
I[09:52:18.687] Loading compilation database...
I[09:52:18.692] Loaded compilation database from C:\Users\lenovo\helloc\build/compile_commands.json
V[09:52:18.693] Broadcasting compilation database from C:\Users\lenovo\helloc
V[09:52:18.694] System include extraction: not allowed driver C:\PROGRA~1\REDPAN~1\MinGW64\bin\gcc.exe
I[09:52:18.696] Compile command from CDB is: "C:\\Program Files\\RedPanda-Cpp\\MinGW64\\bin\\gcc.exe" -g -std=gnu99 -o "CMakeFiles\\helloc.dir\\main.c.obj" -c "-resource-dir=C:\\Program Files\\clangd_15.0.6\\lib\\clang\\15.0.6" -- "C:\\Users\\lenovo\\helloc\\main.c"

--query-driver without spaces clangd report System includes extractor: successfully executed

C:\Program Files\clangd_15.0.6\bin>clangd --check=C:\Users\lenovo\helloc\main.c --query-driver="C:\PROGRA~1\REDPAN~1\MinGW64\bin\gcc.exe" --log=verbose
I[10:00:32.309] clangd version 15.0.6 (https://github.com/llvm/llvm-project 088f33605d8a61ff519c580a71b1dd57d16a03f8)
I[10:00:32.311] Features: windows+grpc
I[10:00:32.312] PID: 1116
I[10:00:32.313] Working directory: C:\Program Files\clangd_15.0.6\bin
I[10:00:32.314] argv[0]: clangd
I[10:00:32.314] argv[1]: --check=C:\Users\lenovo\helloc\main.c
I[10:00:32.315] argv[2]: --query-driver=C:\PROGRA~1\REDPAN~1\MinGW64\bin\gcc.exe
I[10:00:32.316] argv[3]: --log=verbose
V[10:00:32.328] User config file is C:\Users\lenovo\AppData\Local\clangd\config.yaml
I[10:00:32.330] Entering check mode (no LSP server)
I[10:00:32.331] Testing on source file C:\Users\lenovo\helloc\main.c
I[10:00:32.335] Loading compilation database...
I[10:00:32.340] Loaded compilation database from C:\Users\lenovo\helloc\build/compile_commands.json
V[10:00:32.341] Broadcasting compilation database from C:\Users\lenovo\helloc
V[10:00:32.372] System include extraction: target extracted: "x86_64-w64-mingw32"
V[10:00:32.373] System include extraction: adding  C:/PROGRA~1/REDPAN~1/MinGW64/bin/../lib/gcc/x86_64-w64-mingw32/11.2.0/include
V[10:00:32.375] System include extraction: adding  C:/PROGRA~1/REDPAN~1/MinGW64/bin/../lib/gcc/x86_64-w64-mingw32/11.2.0/include-fixed
V[10:00:32.377] System include extraction: adding  C:/PROGRA~1/REDPAN~1/MinGW64/bin/../lib/gcc/x86_64-w64-mingw32/11.2.0/../../../../x86_64-w64-mingw32/include
I[10:00:32.378] System includes extractor: successfully executed C:\PROGRA~1\REDPAN~1\MinGW64\bin\gcc.exe
        got includes: "C:/PROGRA~1/REDPAN~1/MinGW64/bin/../lib/gcc/x86_64-w64-mingw32/11.2.0/include, C:/PROGRA~1/REDPAN~1/MinGW64/bin/../lib/gcc/x86_64-w64-mingw32/11.2.0/include-fixed, C:/PROGRA~1/REDPAN~1/MinGW64/bin/../lib/gcc/x86_64-w64-mingw32/11.2.0/../../../../x86_64-w64-mingw32/include"
        got target: "x86_64-w64-mingw32"
I[10:00:32.383] Compile command from CDB is: "C:\\Program Files\\RedPanda-Cpp\\MinGW64\\bin\\gcc.exe" -g -std=gnu99 -o "CMakeFiles\\helloc.dir\\main.c.obj" -c -isystem C:/PROGRA~1/REDPAN~1/MinGW64/bin/../lib/gcc/x86_64-w64-mingw32/11.2.0/include -isystem C:/PROGRA~1/REDPAN~1/MinGW64/bin/../lib/gcc/x86_64-w64-mingw32/11.2.0/include-fixed -isystem C:/PROGRA~1/REDPAN~1/MinGW64/bin/../lib/gcc/x86_64-w64-mingw32/11.2.0/../../../../x86_64-w64-mingw32/include --target=x86_64-w64-mingw32 "-resource-dir=C:\\Program Files\\clangd_15.0.6\\lib\\clang\\15.0.6" -- "C:\\Users\\lenovo\\helloc\\main.c"
I[10:00:32.384] Parsing command...

if I remove spaces and keep the symbol -, I also have the same problem.

C:\Program Files\clangd_15.0.6\bin>clangd --check=C:\Users\lenovo\helloc\main.c --query-driver="C:\PROGRA~1\RedPanda-Cpp\MinGW64\bin\gcc.exe" --log=verbose
I[10:48:24.066] clangd version 15.0.6 (https://github.com/llvm/llvm-project 088f33605d8a61ff519c580a71b1dd57d16a03f8)
I[10:48:24.068] Features: windows+grpc
I[10:48:24.069] PID: 612
I[10:48:24.071] Working directory: C:\Program Files\clangd_15.0.6\bin
I[10:48:24.073] argv[0]: clangd
I[10:48:24.075] argv[1]: --check=C:\Users\lenovo\helloc\main.c
I[10:48:24.079] argv[2]: --query-driver=C:\PROGRA~1\RedPanda-Cpp\MinGW64\bin\gcc.exe
I[10:48:24.080] argv[3]: --log=verbose
V[10:48:24.086] User config file is C:\Users\lenovo\AppData\Local\clangd\config.yaml
I[10:48:24.087] Entering check mode (no LSP server)
I[10:48:24.088] Testing on source file C:\Users\lenovo\helloc\main.c
I[10:48:24.091] Loading compilation database...
I[10:48:24.096] Loaded compilation database from C:\Users\lenovo\helloc\build/compile_commands.json
V[10:48:24.096] Broadcasting compilation database from C:\Users\lenovo\helloc
V[10:48:24.097] System include extraction: not allowed driver C:\PROGRA~1\REDPAN~1\MinGW64\bin\gcc.exe
I[10:48:24.099] Compile command from CDB is: "C:\\Program Files\\RedPanda-Cpp\\MinGW64\\bin\\gcc.exe" -g -std=gnu99 -o "CMakeFiles\\helloc.dir\\main.c.obj" -c "-resource-dir=C:\\Program Files\\clangd_15.0.6\\lib\\clang\\15.0.6" -- "C:\\Users\\lenovo\\helloc\\main.c"
I[10:48:24.102] Parsing command...

System information
Win10 LTSC 2019 in vmware

Output of clangd --version:

clangd version 15.0.6 (https://github.com/llvm/llvm-project 088f33605d8a61ff519c580a71b1dd57d16a03f8)
Features: windows+grpc
Platform: x86_64-pc-windows-msvc

Editor/LSP plugin:
vscode + clangd + cmake tools
Operating system:

[HighCommander4]

The issue isn't so much that the argument to --query-driver can't have spaces in it, it's that this line from the logs:

V[09:52:18.694] System include extraction: not allowed driver C:\PROGRA~1\REDPAN~1\MinGW64\bin\gcc.exe

tells us that the actual driver path that clangd is trying to use, which it gets from the compilation database, is C:\PROGRA~1\REDPAN~1\MinGW64\bin\gcc.exe. It is this path that clangd then tries to compare against the argument of --query-driver, and the comparison fails.

The comparison is currently a plain string comparison, it's not aware of the fact that e.g. PROGRA~1 is a shorthand for Program Files.

Could the comparison be reworked to understand that? Probably, though I'm not familiar with the relevant path-related APIs (possibly Windows-specific APIs would need to be used).

Anyways, a workaround is to look in the logs for "not allowed driver" message, and use the path from there in --query-driver. You can also use a glob, e.g. --query-driver=C:\**\MinGW64\bin\gcc.exe.

[ReasonJinMing]

The issue isn't so much that the argument to --query-driver can't have spaces in it, it's that this line from the logs:

V[09:52:18.694] System include extraction: not allowed driver C:\PROGRA~1\REDPAN~1\MinGW64\bin\gcc.exe

tells us that the actual driver path that clangd is trying to use, which it gets from the compilation database, is C:\PROGRA~1\REDPAN~1\MinGW64\bin\gcc.exe. It is this path that clangd then tries to compare against the argument of --query-driver, and the comparison fails.

The comparison is currently a plain string comparison, it's not aware of the fact that e.g. PROGRA~1 is a shorthand for Program Files.

Could the comparison be reworked to understand that? Probably, though I'm not familiar with the relevant path-related APIs (possibly Windows-specific APIs would need to be used).

Anyways, a workaround is to look in the logs for "not allowed driver" message, and use the path from there in --query-driver. You can also use a glob, e.g. --query-driver=C:\**\MinGW64\bin\gcc.exe.

Thank you for your precise answers.
I think the problem is that the not allowed driver hint might give us the wrong information.When I first met it, my immediate thoughts were that I had the wrong compiler.I can hardly imagine that the problem is the mismatch of path of complier between compilation database and --query-driver.

[dogtopus]

The comparison is currently a plain string comparison, it's not aware of the fact that e.g. PROGRA~1 is a shorthand for Program Files.

This also causes other issues like patterns with mixed forward/backward slashes fail to match anything, breaks setup that invokes toolchain from a workspace-relative directory under vscode. (e.g. ${workspaceFolder}/toolchain/bin/arm-none-eabi-* on vscode-clangd matches fine on *NIX but fails on Windows due to ${workspaceFolder} being expanded to something like C:\Users\user\Documents\my_project causing the whole pattern to be interpreted as something like C:\Users\user\Documents\my_project/toolchain/bin/arm-none-eabi-* when the actual compile DB contains canonical paths only, causing the match to fail).