iwyu recommends changing C++ headers for C equivalents`

[pnorman]

node-ram-cache.hpp should add these lines:
#include <sys/_stdint.h>                // for int64_t, int32_t
#include <sys/limits.h>                 // for INT_MIN

node-ram-cache.hpp should remove these lines:
- #include <climits>  // lines 15-15
- #include <cstdint>  // lines 16-16

These are the same headers

[ThosRTanner]

Not necessarily true (though it's difficult to tell with macros). In general #include <cxxxx> gives you stuff in the std:: namespace and #include <xxxx.h> gives you stuff in the global namespace.

The exact way behaviour is rather dependant on your compiler headers, but gcc and stlport tend to be fairly firm about this.

Also the recommendation of sys/_stdint.h seems to indicate that iwyu doesn't recognise your system very well.

[vsapsai]

Thanks for report, @pnorman. If you can include the same file multiple ways, IWYU should prefer the existing way of including. Unfortunately, as you can see, it's not always the case.

Was able to partially reproduce the issue on Mac OS X.

tests/cxx/issue226.cpp:5:3: warning: int32_t is defined in <sys/_types/_int32_t.h>, which isn't directly #included.

tests/cxx/issue226.cpp should add these lines:
#include <sys/_types/_int32_t.h>  // for int32_t

tests/cxx/issue226.cpp should remove these lines:
- #include <cstdint>  // lines 2-2

The full include-list for tests/cxx/issue226.cpp:
#include <sys/_types/_int32_t.h>  // for int32_t
#include <climits>                // for INT_MIN

[csilvers]

} These are the same headers

Not according to the standard. climits is the same as limits.h, not
sys/limits.h. (Modulo the namespace issues already mentioned.) And
cstdint is the same as stdint.h, not sys/_stdint.h.

I don't know what sys/limits.h or sys/_stdint.h are; they are not
standard headers. You don't say what system you're on, but you will
need to add a mapping to map those (internal only?) headers to the
canonical names you want them to have.

If you were to map sys/limits.h to limits.h, for instance, iwyu would
not suggest replacing limits.h with climits -- since as you point out,
they are "the same".

(I'm not saying iwyu is doing the right thing here: it should not be
suggesting you replace the includes you have, which are the correct ones
for the types/macros you're using. But the reason iwyu is tripping up
here is because it does not realize sys/_stdint.h and sys/limits.h are
internal-only headers. You'll have to tell it.)

[ptomato]

I ran into this and made a minimal example source that shows IWYU wrongly recommending the C++ include over the C include, where the source has the identifier without the std:: prefix:

#include <stdint.h>
#include <random>

template <typename T>
T foo() {
    std::mt19937_64 gen(0);
    return std::uniform_int_distribution<T>(0)(gen);
}

void bar() {
    foo<int8_t>();
}

$ include-what-you-use bug.cpp 

bug.cpp should add these lines:
#include <cstdint>  // for int8_t

bug.cpp should remove these lines:
- #include <stdint.h>  // lines 1-1

The full include-list for bug.cpp:
#include <cstdint>  // for int8_t
#include <random>   // for mt19937_64, uniform_int_distribution
---

[kimgr]

@ptomato I think in this case they're equivalent -- <cstdint> exports its symbols into the global namespace too (by questionable design). But I think that means IWYU should be more conservative; there's no use making changes for the sake of making changes. Thanks for the small repro.

[ptomato]

I can't reproduce this anymore on 0.22. It may be fixed.

[kimgr]

Thanks, closing as fixed.