#pragma clang diagnostic ignored "-Wpadded" leaves scope

[tomvierjahn]

Description

I am using clang and -Wpadded.

#include "foo.hpp"
#include "catch/catch.hpp

and clang complains about padding in class Foo, as desired.

Whereas

#include "catch/catch.hpp
#include "foo.hpp"

leads to no warnings.

It seems that some #pragma clang diagnostic ignored "-Wpadded" are not enclosed by push/pop in the single header version.

Steps to reproduce

see above

Extra information

• Catch version: v1.6.0
• Operating System: MacOs X 10.12.3
• Compiler+version: Apple LLVM version 8.0.0 (clang-800.0.42.1)

[horenmar]

Can you also check with current Catch? I suspect it will still be there, but ...

Also it might be intentional -- Catch generates some code in the test files, so we have to leak suppressions to the test file as well -- but in this case I don't think it is and will have to check.

[tomvierjahn]

It still does.

• Catch version: v2.0.0-develop.3 (git hash afc0b31)

[horenmar]

@tomvierjahn That is not the current version. We probably should have deleted that branch when we went back to fixing Catch Classic, but it was left around in case we wanted to backport some changes.

Using Catch v1.8.2 and compiling this file

struct foobar {
    bool b;
    int i;
};

#define CATCH_CONFIG_MAIN
#include "catch.hpp"

TEST_CASE("aa") {
    REQUIRE(1 == 2);
}

void foo() {
    foobar fb;
}

Triggers -Wpadded, so I am going to close this issue.

xarn@DESKTOP-A79J11O:~/scratch$ clang++-3.8 test.cpp -Weverything -Wno-missing-prototypes -Wno-exit-time-destructors
test.cpp:5:9: warning: padding struct 'foobar' with 3 bytes to align 'i' [-Wpadded]
    int i;
        ^
1 warning generated.

[onqtam]

I think the problem is when the struct is defined after the catch header and not before, but the issue seems fixed - online compiler - triggers the warning as desired

[horenmar]

@onqtam My brief testing showed that Clang cares about instantiation of the structs, not definition, when issuing -Wpadded.

[tomvierjahn]

Unfortunately, it is not fixed yet.
Furthermore, it shows "strange" behaviour …

Setup

• Catch version (from catch.hpp), git hash 0c015aa:

*  Catch v1.8.2
*  Generated: 2017-03-13 21:18:33.619572

• clang: Apple LLVM version 8.1.0 (clang-802.0.38)
• warning levels:

-Werror
-Weverything
-pedantic
-pedantic-errors
-Wextra-tokens
-Wambiguous-member-template
-Wbind-to-temporary-copy
-Wno-c++98-compat

Case 1:

• single file only: test_foo.cpp
• shows expected behaviour

variant A

test_foo.cpp:

#define CATCH_CONFIG_MAIN
#include "catch/catch.hpp"

class A {
public:
  float f;
  bool i;
};

#pragma clang diagnostic ignored "-Wexit-time-destructors"
TEST_CASE("padding", "[padding]") {
  A a;
  a.i = true;
  CHECK(a.i);
}

warning:

padding size of 'A' with 3 bytes to alignment boundary [-Werror,-Wpadded]

variant B

test_foo.cpp:

#define CATCH_CONFIG_MAIN

class A {
public:
  float f;
  bool i;
};

#include "catch/catch.hpp"

#pragma clang diagnostic ignored "-Wexit-time-destructors"
TEST_CASE("padding", "[padding]") {
  A a;
  a.i = true;
  CHECK(a.i);
}

warning:

padding size of 'A' with 3 bytes to alignment boundary [-Werror,-Wpadded]

variant C

test_foo.cpp:

class A {
public:
  float f;
  bool i;
};

#define CATCH_CONFIG_MAIN
#include "catch/catch.hpp"

#pragma clang diagnostic ignored "-Wexit-time-destructors"
TEST_CASE("padding", "[padding]") {
  A a;
  a.i = true;
  CHECK(a.i);
}

warning:

padding size of 'A' with 3 bytes to alignment boundary [-Werror,-Wpadded]

Case 2

• separate files test.cpp and test_foo.cpp
• behaviour depends on the position of `#include "catch/catch.hpp"

test.cpp

#define CATCH_CONFIG_MAIN
#include "catch/catch.hpp"

variant A:

test_foo.cpp:

#include "catch/catch.hpp"

class A {
public:
  float f;
  bool i;
};

#pragma clang diagnostic ignored "-Wexit-time-destructors"
TEST_CASE("padding", "[padding]") {
  A a;
  a.i = true;
  CHECK(a.i);
}

no warning

variant B:

test_foo.cpp:

class A {
public:
  float f;
  bool i;
};

#include "catch/catch.hpp"

#pragma clang diagnostic ignored "-Wexit-time-destructors"
TEST_CASE("padding", "[padding]") {
  A a;
  a.i = true;
  CHECK(a.i);
}

warning:

padding size of 'A' with 3 bytes to alignment boundary [-Werror,-Wpadded]

[horenmar]

I finally got back to this and it should now be fixed in master (not in the single include yet).

It turned out that this was by interplay between superfluous include, and our single header generator, making it so that a pragma diagnostic pop that was supposed to be at the end of the single include header, was instead inside code that is guarded by CATCH_IMPL define. This meant that in test files, there was 1 more pragma diagnostic push than there were pops.