I’m new to clang world, here I have one question about libclang or libtooling, I would be very appreciate if anyone can help me.
Currently I want to develop a new language, which is C like, just some new syntax sugars on top of C language, I can modify clang lexer code to make that, or I can use translator, but if this requirement can be done by libclang or libtooling, that’s perfect simple for me.
My direct impression is there have no way to change language syntax by libclang/libtooling, libclang/libtooling only can parse and analysis on C family language, no chance to support other language by library, is this correct?
Yes, you’d have to modify clang to lex/parse/sema your keywords/extensions/whatever.
Yes, you'd have to modify clang to lex/parse/sema your
Is there have a simple solution without change to much on clang?
In my case, I want to parse script which have custom language based on
C. The script just like below:
Actually there just have few new keywords(like trace, trace_end) in
this new language, all others in C.
Any guide on this?
And you want to DO what, exactly with it, after you have parsed it?
And does your script use ALL of C's functionality?
Some functions will be compiled to bpf bytecode and inject into
kernel(run in kernel), some functions
in one script will be compiled to x86_64 instruction which run in userspace.
(There have one project called ktap, www.ktap.org, mainly purpose for
kernel dynamic tracing, probably you knows systemtap,
ktap is similar like systemtap. There will have a bpf backend for
llvm, now I'm just think how to integrating ktap and bpf)
Some basic requirement of this new frontend is:
1). support new keywords, like trace, trace_end, argstr.
2). support table syntax, also multiple key table. C language don't
support table syntax at all.
3). data structure info retrieve from debuginfo(like vmlinux), not
defined in script or header file.
Do you think my design can achieve by change some part of clang(lex,
etc) and reuse rest library of clang? or have better solution?
If you just want C and not C++ or other language extensions, then I think writing your own C frontend IR is probably easier than reusing Clang’s. All the complexity in Clang comes from C++, really, and the frontend isn’t designed to be pluggable like the optimizer is.
Yes, I just need C, completely don‘t need any C++ and other extension, clang internal looks have more complexity than I need.
Do you have any guide on how to write C frontend based on LLVM? any project can take reference other than clang? I looked into tcc before, but it seems it’s not fully C completely. actually I’m totally new to compiler, but I really want to build that tracing tool based on bpf.
Yes, I just need C, completely don‘t need any C++ and other extension,
clang internal looks have more complexity than I need.
Do you have any guide on how to write C frontend based on LLVM? any
project can take reference other than clang?
FYI, clang started out as a C compiler before it grew C++ support. Check
back (way back) in the history.
-- Sean Silva