Add support for compilation database generated by CMake. #135
Conversation
The compilation database file name (usually <build folder>/compilation_commands.json) can be automatically generated by CMake, when generating the makefiles, by setting the CMake variable CMAKE_EXPORT_COMPILE_COMMANDS to 1. This can be done from inside the CMakeLists.txt or from the command line. Then, the compilation database can be provided to SublimeClang using the "compilation_database" parameter in the SublimeClang configuration. The compilation database format is specified by the LLVM/Clang project: http://clang.llvm.org/docs/JSONCompilationDatabase.html The file contains, for each source file built, the complete command line used to built it, which allows then SublimeClang to have the exact flags to use to parse the file, without having to add all the include folders manually for each project. The parsing of each command line is currently done using a regex, which tries to match any option starting with -D, -I, -O, -U, -W, -f, -g, or -s. This is not ideal, and shlex python module could be used instead to have a more clean split of the arguments, but it's much faster to parse the database using this regex. Also, the database parsing is currently done asynchronously, when the first source file is loaded. This might be improved by doing it the earliest possible.
|
Actually, I've just realised that this does not work for header files... only source files have a command associated. |
The flags retrieved from the compilation database, are merged together for each folder containing a source file. This adds support for header files located in the same folder as a source file, however it does not solve the missing flags for headers located in folder which do not contain any source. Also, this merges the compilation flags together, which is maybe not desired. Maybe a configuration parameter to control this would be appropriate.
|
This new commit partially solves the issue, at least for header files located in the same folder as the sources. However, it uses the union of all the flags of all the sources inside the folder for any source or header file located in the same folder. This is maybe not optimal, as we loose the possibility to use the exact flags that are used to compile each source file. Also it does not solve the issue of missing flags for headers located elsewhere, where no source file is located (for example in a separated include/ folder). |
|
I'm leaning towards not accepting this pull request as it's awfully CMake specific and I see no reason the same task couldn't be solved via the already implemented options script mechanism. |
|
Well, CMake is able to generate such database. However the database format and mechanism is specified by Clang (see the link in the first comment). I even think the Clang developers are responsible for the CMake support, as Clang is built using it. By the way, the database is currently parsed by the plugin, but it seems that the clang python bindings have a direct support for it since Clang 3.2. http://clang.llvm.org/doxygen/group__COMPILATIONDB.html I haven't looked how to use this correctly. Anyway, as the implementation is not complete and has some obvious issues that I did not foresee, I would say this require some more work. The script is maybe one better solution, and would simply parse the database, aggregating all the flags together... Although I liked the possibility to have the exact flags used to build each file (which could vary from file to file). |
The filename is provided to the options script so you don't necessarily lose that functionality. |
|
If I may add an alternative suggestion: |
|
Option scripts are so very dependent on the build system used which IMO makes them better handled via gists and a link to the gist in the documentation, similar to how there are a couple of example configurations provided as gists. I'd rather not bloat the code unnecessarily, especially with code paths I do not use myself and thus wouldn't notice when it breaks. I'm going to close this for now, but if anyone strongly disagrees with me feel free to continue the argumentation here :) |
|
FYI, I turned this into a gist at https://gist.github.com/3944250, with a link provided in the settings file. |
The compilation database file name (usually /compilation_commands.json) can
be automatically generated by CMake, when generating the makefiles, by setting the CMake
variable CMAKE_EXPORT_COMPILE_COMMANDS to 1. This can be done from inside the CMakeLists.txt
or from the command line. Then, the compilation database can be provided to SublimeClang
using the "compilation_database" parameter in the SublimeClang configuration.
The compilation database format is specified by the LLVM/Clang project:
http://clang.llvm.org/docs/JSONCompilationDatabase.html
The file contains, for each source file built, the complete command line used to built it, which
allows then SublimeClang to have the exact flags to use to parse the file, without having to
add all the include folders manually for each project.
The parsing of each command line is currently done using a regex, which tries to match any
option starting with -D, -I, -O, -U, -W, -f, -g, or -s. This is not ideal, and shlex python
module could be used instead to have a more clean split of the arguments, but it's much
faster to parse the database using this regex.
Also, the database parsing is currently done asynchronously, when the first source file is loaded.
This might be improved by doing it the earliest possible.