On Wed, 2020-02-26 at 15:51 -0700, Martin Sebor wrote: > GCC accepts attribute format even on functions without a prototype. > However, when such a function is subsequently redeclared with > a prototype the attribute validation runs into trouble. To avoid > the ICE I considered a) dropping the attribute on functions declared > without a prototype (like Clang does) and b) dropping the attribute > only when the function is subsequently redeclared with a prototype. > Since (b) is slightly safer in terms of retaining potentially useful > functionality the attached patch implements it and verifies that > the attribute is still accepted on functions without a prototype > and has the expected effect. > > Tested on x86_64-linux. I propose this for both GCC 10 and 9. > > Martin > > PS The solution causes a redunant warning for each instance of > a redeclaration with a prototype of a function without a prototype > but with attribute format. Since this seems like a rare problem > I don't think it's worth going to the trouble of making sure only > one warning. > > PPS I don't think it's intentional that GCC accepts attribute format > on functions without a prototype, first because it normally requires > that the format argument be a char*, and second because I could find > no tests for the "feature" in the test suite. It would make sense > to me reject such declarations but I don't think now is a good time > for such a change.
> PR c/93812 - ICE on redeclaration of an attribute format function without > protoype > > gcc/c/ChangeLog: > > PR c/93812 > * c-typeck.c (build_functype_attribute_variant): New function. > (composite_type): Call it. > > gcc/testsuite/ChangeLog: > > PR c/93812 > * gcc.dg/format/proto.c: New test. OK. If you want to drop like Clang for gcc-11, that's fine with me, but we'll definitely want to warn when we drop and document the behavior change. Thanks, Jeff
