sunlin7 / compile-db-gen Goto Github PK

I just wanted to thank you for this project. About a year ago I found myself in a situation, when I had to work on a project with an unusual build system. I tried every other compile_commands.json generation tools, but they all were failing to capture compiler invocations properly. Your idea of using strace is brilliant and worked nicely!

Some time later I decided to improve on your idea and develop compile_commands.json generator of my own. The main idea was to use ptrace directly instead or relying on parsing strace logs which are not meant to be machine-readable. Now it is more or less ready for use - https://github.com/i-ky/basset. I will appreciate if you give it a spin and test it in your usage scenarios to find any issues that I have missed.

Newer versions of strace support --seccomp-bpf option introduced in version 5.3:

Implemented usage of seccomp-bpf for stopping tracees only for filtered
syscalls. Use --seccomp-bpf option to enable.

It has huge impact on performance. According to my tests, strace command currently used by compile-db-gen adds ~100% overhead (i.e. build time doubles under strace). Simply adding --seccomp-bpf reduces overhead to ~10% making it almost unnoticeable.

I honestly don't know why strace developers did not make --seccomp-bpf a new default, but it definitely makes sense to use it in compile-db-gen, if available.

compile-db-gen is already picking only successful syscalls due to = 0 in regular expressions:

Filtering them on strace side would reduce the amount of data compile-db-gen needs to parse.