Extend .ccls

[MaskRay]

Fix #115 @madscientist

• Add %h for C header files (the suffix .h is considered a C header, not a C++ header)
• Add %hpp for C++ header files
• If .ccls exists, it provides full command line for files not specified by compile_commands.json (before, compile_commands.json was ignored)
• If the first line of .ccls is %compile_commands.json, it appends flags to compile_commands.json "arguments", instead of overriding.
  Files not specified by compile_commands.json will not be added to folder.entries, but their command line can be inferred from other files.

[madscientist]

Wow thanks for the quick response! I'm testing now but something is not quite right. At first I was getting an error in Messages (maybe while ccls was rebuilding the index?) but now I get no errors but the header doesn't appear to be included. I do see the --include flag mentioned in the log though. I'll do some debugging provide more details.

[madscientist]

I enabled lsp-print-io but it didn't seem to help me understand what's happening. Basically, I can see from the log that the %h option is being found; I see:

project.cc:247 I use /tmp/cclstest/.ccls: %h --include=global.h

but ccls still can't find these types.

I've created a test environment and attached it as a tar file here. Unpack it in /tmp (there are some hardcoded paths in compile_commands.json) then run emacs and visit the header file src/header.h and you'll see that size_t is not recognized as a valid type:

cd /tmp
tar xzf cclstest.tar.gz
emacs -nw -q -l ccls-test.el
C-x C-f cclstest/src/header.h

If I do the equivalent thing from the command line it works:

cd src
clang --include=global.h -I/tmp/cclstest/inc -o source.o -c source.c

even if I comment out the #include "global.h" from source.c, so this syntax does work properly from the command line.

cclstest.tar.gz

[madscientist]

OK more details: it appears to be related to the combination of compile_commands.json and .ccls. If I remove compile_commands.json and change my .ccls to have all the flags:

%c
-I/tmp/cclstest/inc
%h
--include=global.h

then it works! If I put back compile_commands.json, even with the above .ccls, it doesn't work again.

BTW, I think a more formal definition of the syntax of .ccls in the wiki would be helpful; while the above works, if I put the option on the same line as %h

%h --include=global.h

then it doesn't work... is that intended? Using %c -I/tmp/cclstest/inc on one line does seem to work.

[MaskRay]

I pushed a commit to make -v=1 work: `ccls -log-file=/tmp/ls.log -v=1 "$@" will now log:

18:26:44 preamble     sema_manager.cc:706 I create session for /tmp/cclstest/src/source.c
18:26:44 preamble     sema_manager.cc:711 I  args: gcc -I/tmp/cclstest/inc -o source.o -c /tmp/cclstest/src/source.c --gcc-toolchain=/usr -working-directory=/tmp/cclstest/src

It should make it easy to observe what is going on.

I have fixed %h. It required some more work.

Note %c %h are modifiers that get stripped as the line prefix. You can't place the modified flag on the next line.

With a change to .ccls (it overrides compile_commands.json entries, instead of augmenting its command line with more flags), your example works for me:

% cat /tmp/cclstest/.ccls
%clang
%h --include=inc/global.h

[madscientist]

Awesome, I'll give it a try.

I was confused about the precise syntax of the .ccls file; the https://github.com/MaskRay/ccls/wiki/Getting-started wiki page shows this syntax:

%clang
%c -std=gnu11
%cpp -std=gnu++14
%objective-c %objective-cpp -DGNUSTEP
%c %cpp -DFOO
-pthread

# Includes
-I/work/ccls/third_party
-I/work/ccls/another_third_party
# -I space_is_not_allowed

which implies that flags on a line by themselves are valid...? Maybe I misunderstood it.

[MaskRay]

With your comment #115 (comment) in mind, I've changed the semantic a bit.

For you /tmp/cclstest example to work, you may also use:

% cat .ccls
%compile_commands.json
%h --include=global.h

See the updated description of the commit.

The relevant log when src/source.c and src/header.h (inferred from src/source.c) are opened:

22:32:54 preamble     sema_manager.cc:712 I create session for /tmp/cclstest/src/source.c
  gcc -I/tmp/cclstest/inc -o source.o -c /tmp/cclstest/src/source.c --gcc-toolchain=/usr -working-directory=/tmp/cclstest/src
22:33:42 preamble     sema_manager.cc:712 I create session for /tmp/cclstest/src/header.h
  gcc -I/tmp/cclstest/inc -o source.o -c /tmp/cclstest/src/header.h --include=global.h --gcc-toolchain=/usr -working-directory=/tmp/cclstest/src

I was confused about the precise syntax of the .ccls file

The source is still the best way to understand what's going on :) project.cc

      StringRef A(arg);
      if (A == "%clang") {
        args.push_back(lang == LanguageId::Cpp || lang == LanguageId::ObjCpp
                           ? "clang++"
                           : "clang");
      } else if (A[0] == '%') {
        bool ok = false;
        for (;;) {
          if (A.consume_front("%c "))
            ok |= lang == LanguageId::C;
          else if (A.consume_front("%h "))
            ok |= lang == LanguageId::C && header;
          else if (A.consume_front("%cpp "))
            ok |= lang == LanguageId::Cpp;
          else if (A.consume_front("%hpp "))
            ok |= lang == LanguageId::Cpp && header;
          else if (A.consume_front("%objective-c "))
            ok |= lang == LanguageId::ObjC;
          else if (A.consume_front("%objective-cpp "))
            ok |= lang == LanguageId::ObjCpp;
          else
            break;
        }
        if (ok)
          args.push_back(A.data());
      } else if (!excludeArgs.count(A)) {
        args.push_back(arg);
      }

[MaskRay]

@Riatre @oblitum @bstaletic

[oblitum]

This didn't fix inclusion of headers at %cpp -I/opt/src/range-v3/include for me :-/ (#160).

[oblitum]

Can you give an example how %h and %hpp is supposed to be used? Is it for include directories or files?

[oblitum]

Can you give an example how %h and %hpp is supposed to be used? Is it for include directories or files?

Nevermind, I've noticed from discussion it's for files.

[madscientist]

@MaskRay OK tested in my real code... it works great!! Now all my headers have full type information. Love it!

@oblitum Right, for files. It's for extra flags to be given to the compiler when ccls "compiles" header files alone, rather than as part of a source file. I don't expect it will be the most commonly-used option but it's critical in some situations.

[MaskRay]

@oblitum For clang < 8, include_complete.cc (inherited from cquery) is used. I've commented at #160 (comment) The latest project branch should address #160