Am 06.03.21 um 18:52 schrieb Ævar Arnfjörð Bjarmason:

On Sat, Mar 06 2021, René Scharfe. wrote:

quoted

Am 06.03.21 um 18:27 schrieb Ævar Arnfjörð Bjarmason:

quoted

Do these sorts of rules really benefit that much from the type
v.s. expression? If yes we'll obviously need to support it, but if (and
I haven't looked closely) we can equally rewrite them with "expression"
(or it would be good enough) we could be quite a bit faster by
default...

Type information is essential for many (most?) semantic patches.
Untyped replacements could be done more easily using sed or similar.

We have 65 individual hunks in *.cocci by my count $((/ (cat *.cocci |
grep -c ^@@) 2)), but only 9 of those use the "type T" construct, which
AFAICT is the only thing affected.

I don't understand why using type variables would be the only feature
that needs headers.  Accessing struct members like in commit.cocci surely
needs their definition, no?

There's still plenty of reason to use spatch without headers being
included, it's still doing a full parse and knows that something's a
function, and otherwise cares about C syntax etc.

Anyway, I'm not saying you don't need it, just that AFAICT it's a small
minority of the rules that require the includes.

https://coccinelle.gitlabpages.inria.fr/website/docs/options.pdf says
"If the semantic patch contains type speci
fications on expression
metavariables, then the default is --local-includes. Otherwise the
default is --no-includes."   So spatch by default already avoids loading
includes unless needed, but we override that.

René