[clang][Sema] Question: Why is -Wshorten-64-to-32 disjoint from -Wimplicit-int-conversion?

[whisperity]

An interesting case was encountered where conversions from larger bit-width integers to smaller bit-width ones is generally reported by -Wimplicit-int-conversion (32 to 16, 16 to 8, 128 to 64, etc.) except for 64 to 32, which was given its own warning: -Wshorten-64-to-32.

See http://godbolt.org/z/6q4GMPsxs. Lines 19 (uint64_t -> uint32_t) and 35 (int64_t -> int32_t) are not warned about, unless "Compiler 2"'s warning is enabled.

Although -Weverything enables both of them at the same time (and neither is enabled by default, AFAIK), it is strange that 64->32 is such a specific dedicated case.

Unfortunately, all I could track down was commit 263a48b where this warning was introduced, and that is ancient, but history runs dry at that point.

CC @dwblaikie, @rjmccall. Your names appeared in the relevant code path.

---

What would be the cons against enabling -Wshorten-64-to-32 under -Wimplicit-int-conversion implicitly?

[llvmbot]

@llvm/issue-subscribers-c-1

Author: None (whisperity)

An interesting case was encountered where conversions from larger bit-width integers to smaller bit-width ones is *generally* reported by `-Wimplicit-int-conversion` (32 to 16, 16 to 8, 128 to 64, etc.) **except** for 64 to 32, which was given its own warning: `-Wshorten-64-to-32`.

See http://godbolt.org/z/6q4GMPsxs. Lines 19 (uint64_t -> uint32_t) and 35 (int64_t -> int32_t) are not warned about, unless "Compiler 2"'s warning is enabled.

Although -Weverything enables both of them at the same time (and neither is enabled by default, AFAIK), it is strange that 64->32 is such a specific dedicated case.

Unfortunately, all I could track down was commit 263a48b where this warning was introduced, and that is ancient, but history runs dry at that point.

> CC @dwblaikie, @rjmccall. Your names appeared in the relevant code path.

---

What would be the cons against enabling -Wshorten-64-to-32 under -Wimplicit-int-conversion implicitly?

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: None (whisperity)

An interesting case was encountered where conversions from larger bit-width integers to smaller bit-width ones is *generally* reported by `-Wimplicit-int-conversion` (32 to 16, 16 to 8, 128 to 64, etc.) **except** for 64 to 32, which was given its own warning: `-Wshorten-64-to-32`.

See http://godbolt.org/z/6q4GMPsxs. Lines 19 (uint64_t -> uint32_t) and 35 (int64_t -> int32_t) are not warned about, unless "Compiler 2"'s warning is enabled.

Although -Weverything enables both of them at the same time (and neither is enabled by default, AFAIK), it is strange that 64->32 is such a specific dedicated case.

Unfortunately, all I could track down was commit 263a48b where this warning was introduced, and that is ancient, but history runs dry at that point.

> CC @dwblaikie, @rjmccall. Your names appeared in the relevant code path.

---

What would be the cons against enabling -Wshorten-64-to-32 under -Wimplicit-int-conversion implicitly?

[rjmccall]

IIRC, -Wshorten-64-to-32 was an earlier warning implemented specifically to help with 32-to-64-bit portability problems, and then the other warnings were generalizations of that logic. I can't think of any reason why they should be disjoint, though; the more general groups should include -Wshorten-64-to-32.