[Enhancement] Import\configure project through compile_commands.json file

[pidgeon777]

[Enhancement] Import\configure project through compile_commands.json file

[ruben2020]

@pidgeon777 Currently CodeQuery is not using libclang, so it doesn't really need compilation commands and flags. cscope and ctags or similar tools only need a list of files. Nevertheless this can be considered a possible feature enhancement. I thought about writing a tool to use libclang as another frontend but I haven't got around to it yet. But I think it will be a lot slower than cscope and ctags. However, it will be more accurate.

[pidgeon777]

This is an example of compile_commands.json file widely used by clangd and ccls:

[
{
  "directory": "C:/Work/Projects/Linux/Programs/Hello_World",
  "command": "C:/CrossComp/bin/aarch64-none-linux-gnu-gcc.exe -Wall -Wextra -I. -isystem C:/CrossComp/aarch64-none-linux-gnu/libc/usr/include -O3 -DNDEBUG -std=c99 -o main.o -c main.c",
  "file": "C:/Work/Projects/Linux/Programs/Hello_World/main.c"
},
{
  "directory": "XXX",
  "command": "YYY",
  "file": "ZZZ"
},
...
]

The main key elements are:

• -I/path/to/include/headers: this argument specifies the folder of include headers needed to compile the source file specified by the associated "file" parameter. For each source file to be compiled, multiple -Ixxx arguments could be specified.
• -isystem /path/to/system/headers: this argument specifies the folder of system headers needed to compile the source file specified by the associated "file" parameter. For each source file to be compiled, multiple -isystem arguments could be specified.
• "file": "/path/to/file.c": the source file to compile.

To generate the cscope database, a cscope.files file is used, which list all of the C and H files to be parsed. Also, this could be used for generating a tags file.

This means that by parsing a compile_commands.json file, it could be possible to do something like this:

1. Scan all of the -Ixxx argument paths for header files, recursively add them to the list.
2. Scan all of the -isystemYYY argument paths for system headers files, recursively add them to the list (optional?).
3. Add all of the C sources specified by the "file" to the list.
4. Sort the list and remove duplicated elements.

By doing so, a cscope.files could be generated, like the following:

...
C:\CrossComp\aarch64-none-linux-gnu\libc\usr\include\stdio.h
C:\Work\Projects\Linux\Programs\Hello_World\include.h
C:\Work\Projects\Linux\Programs\Hello_World\main.c
...

and it would include exactly the same files parsed through the compile_commands.json file by clangd or ccls.

So, the resulting cscope/CodeQuery database would be the most complete and accurate one.

I thought about writing a tool to use libclang as another frontend but I haven't got around to it yet. But I think it will be a lot slower than cscope and ctags. However, it will be more accurate.

This would be great and I encourage you to do so. Even better if it could work as an LSP client, thus compatible with almost every programming language.
LSP is the future of programming languages, tags based analysis is quick and fast but it is impossible to obtain the same level of accuracy and completeness provided by LSP-based solutions.