On Thu, May 28, 2026 at 12:32:48PM +0200, Tomas Glozar wrote: > [ CC to linux-perf-users for the libsubcmd code changes ] > > rtla currently uses its own implementation that uses getopt_long() to > parse command-line arguments. > > Migrate rtla to use libsubcmd for command line argument parsing, > similarly to what is already done by other tools like perf, bpftool, > and objtool. Among other benefits, this allows help messages to be > generated automatically rather than having to be typed out manually > for each tool. > > libsubcmd is extended with a flag to parse optarg from separate > argument if a new flag is turned on. Without the flag, the old behavior > is preserved. That keeps the parsing working for tools that use > positional arguments, and allows RTLA to keep its flexible syntax for -C > and -t options and their long variants, --cgroup and --trace-output. > Another flag is added to disable automatic definition of --no-xy for > every option --xy and vice versa, which overlaps for RTLA's --irq and > --thread options. > > The new implementation is moved into a separate file, cli.c, together > with a tiny header counterpart, cli.h. This helps separate the parsing > logic, which has little in common with the rest of RTLA, in a separate > module. Another new file, cli_p.h, is used as a private header to contain > macros and static function declarations that are also used by unit tests > next to cli.c, but should not be imported from elsewhere. > > Macros to generate struct option array fields for libsubcmd's > parse_args() are used to preserve the consolidation of argument parsing > code across different RTLA tools. Kernel and user threads are, as > an exception, treated as common, although they are currently implemented > for timerlat only, in line with earlier consolidation changes. > > The test suite is expanded to include two levels of unit tests, one testing > the already existing tool_parse_args() functions, one tests option callbacks, > which are a new level of the CLI parser added in this patchset. This helps > to verify that no regressions are caused by this refactoring. > > I expect more improvements to the code being possible in the future, > like creating macros for option groups to further deduplicate the code, > reducing the amount of extra code in the _parse_args() functions, or > implementing support for unsetting options (which is currently only > supported for those that do not use a custom callback). > > Base commit: > - > https://git.kernel.org/pub/scm/linux/kernel/git/tglozar/linux.git/commit/?h=rtla-for-next&id=f03a59f949176ce4312cb466245d1243aaf40389 > > Dependencies: > - > https://lore.kernel.org/linux-trace-kernel/[email protected]/T/ > - > https://lore.kernel.org/linux-trace-kernel/[email protected]/ > - > https://lore.kernel.org/linux-trace-kernel/[email protected]/ > (apply in reverse order, alternatively, use base commit above) > > v3 changes (all in cover letter or first commit): > - Add FORCE to all targets that feature a make subcommand to ensure changes > in the dependency will trigger rebuild correctly. > - Convert all dependencies on directory targets LIBSUBCMD_OUTPUT and > LIB_OUTPUT into order-only prerequisites to prevent modifications of > the directory content triggering rebuild of targets inside it. > - Properly mention depedencies of the patchset in the cover letter, as well > as the base commit. > > v2: > https://lore.kernel.org/linux-trace-kernel/[email protected]/T/ >
Oops. I mistakenly added the reviewed-by tag to the v2 patch series. Reviewed-by: Wander Lairson Costa <[email protected]>
