Create a pass which, given a class with bases, tries to replace uses of it with uses of it's bases #261
Comments
|
yes that would be desirable. I'll just add that we're not too likely to get to this anytime soon since we don't have people working on C-Reduce at the moment. |
|
There is already a very similar pass, currently restricted to base class templates: @regehr what do you think of making this pass more generic? Or would you prefer a separate pass? |
|
if there's no loss of reduction power, I think keeping it all in the same pass is great. thanks! |
|
Yesterday I actually overlooked that there is already another pass removing base classes: creduce/clang_delta/RemoveBaseClass.cpp Line 26 in 9250247 On the example above it does remove @mizvekov could you try to run c-reduce again on your example? I'm surprised that it didn't manage to produce a slightly smaller file based on the pass I found. Could it be that your interestingness test is actually preventing it from getting smaller? |
|
I have used creduce 2.11.0, provided by the msys2 environment. Here is the full resulting reduced file: template <bool, class> using a = int;
template <int b> struct c {
static const int d = b;
};
template <size_t> struct e {};
template <size_t f> e<f> g;
struct h {
template <class ak> static auto j(ak) {
auto a = l<ak>;
a();
}
template <class> static auto l() {}
};
class m {
public:
template <size_t i> m(e<i>) {}
};
class n : m {
using bj = m;
public:
using bj::bj;
~n() {
h::j([] {});
}
};
class o : n {
using bj = n;
using bj::bj;
};
class p : o {
using bj = o;
using bj::bj;
};
class J : p {
using bj = p;
using bj::bj;
};
class q : J {
using bj = J;
using bj::bj;
};
class r : q {
using bj = q;
using bj::bj;
};
class s : r {
using bj = r;
using bj::bj;
};
class t : s {
using bj = s;
using bj::bj;
};
class F {
public:
template <a<c<1>::d, int> = 0> F() : bt(g<0>) {}
t bt;
};
int main() { F k; }Running creduce another time on it produces no changes. Here is the full output of that: You can see that it produced a single failure on Here is my interestingness test: #!/usr/bin/fish
export LLVM_SYMBOLIZER_PATH=/usr/bin/true
set -l clang "C:/PROGRA~1/LLVM/bin/CLANG_~1.EXE"
set -l args "-D_LIBCPP_NO_AUTO_LINK" "-std=gnu++23" "--target=x86_64-pc-windows-msvc" "-fno-exceptions" "-fsanitize=function" "-flto=thin" "-ggdb3" "-O0" "-fuse-ld=lld" "-o" "test.exe"
if ! $clang $args "test.cc" 2>&1;
exit 1
end
set output (./test.exe 2>&1)
if test $status != 0;
exit 1
end
if ! string match --quiet --regex "through pointer to incorrect function type" "$output";
exit 1
endThis is running on llvm 17.0.1, but the bug this is looking for is present on llvm master as of today. FWIW, I reduced this further manually to: static auto l() {}
int main() {
auto a = l;
a();
}And this is the MR that fixes this bug: llvm/llvm-project#66816 |
|
let's tag in @chenyang78 here on the off chance he has a minute to think about this! |
|
Hi folks, thank you for the discussion! It's probably due to that |
|
I found one bug in the Since I currently don't have a Windows setup ready with c-reduce I could not test if that also improves the reduction result in the case above. |
|
I tested again with newer version including #263 , but I get the same results. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I have observed that sometimes reduction results in deep class hierarchies:
I think the simplest thing here would be to try replacing uses of a class with uses of the bases of that class.
The text was updated successfully, but these errors were encountered: