-Wall has a new nuisance warning

[luser-dr00g]

I think we should add -Wno-cast-function-type to the CFLAGS but I forget where to do it.

I looked at configure.ac and Makefile.am and various Makefile.mk files but I didn't see other warning flags.

[vtorri]

what is Xpost_Op_Func has var args ?

[vtorri]

like

typedef int (Xpost_Op_Func)(Xpost_Context ctx,...);

it seems to compile without warnings but the devices must be tested

[vtorri]

ha no, error :

src/lib/xpost_operator.c: In function 'xpost_operator_cons':
src/lib/xpost_operator.c:483:23: error: assignment to 'Xpost_Op_Func' {aka 'int (*)(struct _Xpost_Context *, ...)'} from incompatible pointer type 'int (*)(Xpost_Context *)' {aka 'int (*)(struct _Xpost_Context *)'} [-Wincompatible-pointer-types]
  483 |             sp[si].fp = (int(*)(Xpost_Context *))fp;
      |                       ^

[vtorri]

and with

            sp[si].fp = (int(*)(Xpost_Context *, ...))fp;

it seems good

[vtorri]

i let you test and commit ?

[luser-dr00g]

Hmm. I don't know. If we're changing the code instead of the warning flag, then I'm inclined to change the stored type to just int(*)() with no args hence no prototype. It's always casted to the correct type before calling. But I think going back to no prototype will be less prone to tripping future "helpful warnings" from gcc.

[vtorri]

It's up to you :-)

[luser-dr00g]

Changing to unprototyped produces a different warning:

  CC       src/lib/src_lib_libxpost_la-xpost_dev_bgr.lo
In file included from src/lib/xpost_dev_bgr.c:52:
src/lib/xpost_operator.h:68:1: warning: function declaration isn’t a prototype [-Wstrict-prototypes]
   68 | typedef int (*Xpost_Op_Func)();
      | ^~~~~~~

But this happens only once per file instead of every operator function.

[vtorri]

do you consider that xpost should be c89 compliant, or c99 ?

[vtorri]

hmm, c99 according to configure.ac

[luser-dr00g]

Hmm. I forget! :) At one point we were trying to stay at c89. That's where autotools really gave a big help in getting alloca() to work portably.

I guess we're going for c99 except for VLAs, so it can still compile with MVSC.

[vtorri]

alloca check has been simplified : https://www.gnu.org/savannah-checkouts/gnu/autoconf/manual/autoconf-2.70/autoconf.html#Particular-Functions

do you want me to simplify this in xpost code ?

[vtorri]

alloca simplification is done

[vtorri]

@luser-dr00g so, what about

typedef int (*Xpost_Op_Func)(Xpost_Context *ctx,...);

? no warning, and we keep the warning flag

[luser-dr00g]

Excellent.

[luser-dr00g]

Hmm. I don't know about specifying a varargs function. That feel more like lying to the compiler. With the unprototyped function it's more like withholding information from the compiler because it would only freak out about it. Calling it a varargs functions seems like we're making an implementation assumption which may not be guaranteed by the standard.

[vtorri]

maybe that is something that we should ask to gcc devs

[vtorri]

according to guy in #c (irc chan), cast to int (*)() or int (*)(Xpost_Context *ctx,...) is correct