On Mon, Mar 22, 2021 at 07:05:06PM +0100, René Scharfe. wrote:

quoted

This change speeds up the runtime of "make -j8 coccicheck" from ~1m50s
to ~1m20s for me. See [2] for more timings.

Without this patch, /usr/bin/time -l reports:

       95.08 real       598.29 user        32.62 sys
           103727104  maximum resident set size

With --include-headers-for-types, I get this:

       76.37 real       483.83 user        26.77 sys
            97107968  maximum resident set size

This is similar to your numbers.  And adding that option to the demo
script in your [1] gives consistent results for all xargs -n values,
so that option alone fixes the subtle bug you refer to above, right?

However, with the second part of this patch also applied (adding %.h
to FOUND_C_SOURCES), I get this:

       90.38 real       558.38 user        38.32 sys
           100073472  maximum resident set size

Is this still necessary?

If I understand what the spatch options are doing (and I'm not at all
sure that I do), then with --include-headers-for-types, aren't we
avoiding looking for transformations in the included files?

In which case not adding *.h to the list of files we pass means that
we'd fail to find instances in the header files that need to be
mentioned in the output.

-Peff