From: Guy Harris <gharris@sonic.net>
Date: Mon, 7 Apr 2025 21:00:24 +0000 (-0700)
Subject: Suppress bogus C4738 from MSVC.
X-Git-Url: https://git.tcpdump.org/tcpdump/commitdiff_plain/d8078c7fcb4d8260ca2e9ebb78f43d71894be639

Suppress bogus C4738 from MSVC.

Our trick to extract a big-endian IEEE floating-point number from a
packet confuses MSVC into thinking we're doing something we're not
doing, or maybe just that we might do so somewhere that it can't see.
Suppress the resulting warning.

This should fix issue #1318.
---

diff --git a/CMakeLists.txt b/CMakeLists.txt
index c4536674..05bfdce0 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1158,6 +1158,39 @@ if(EXISTS ${CMAKE_SOURCE_DIR}/.devel OR EXISTS ${CMAKE_BINARY_DIR}/.devel)
         # file in tcpdump, gets a warning.
         #
         check_and_add_compiler_option(-wd4324)
+        #
+        # Suppress warnings about a "possible loss of performance"
+        # due to "storing 32-bit float result in memory" in the
+        # "fetch a 32-bit floating-point variable, in a particular
+        # byte order, from a packet" routines.
+        #
+        # What those routines do is fetch a 32-bit unsigned integral
+        # value in the appropriate byte order, store it into the
+        # "32-bit unsigned integer value" member of a union, and then
+        # return the "32-bit floating-point value" member of that union,
+        # in order to reinterpret the bits of the value as if it were
+        # an IEEE floating-point value. (Yes, we know, VAX floating
+        # point; at least the IBM mainframes have supported IEEE
+        # floating-point since the G4 S/390s.)
+        #
+        # The Microsoft page on that warning:
+        #
+        #    https://learn.microsoft.com/en-us/cpp/error-messages/compiler-warnings/compiler-warning-level-3-c4738?view=msvc-170
+        #
+        # says that
+        #
+        #    C4738 warns that the result of an assignment, cast, passed
+        #    argument, or other operation may need to be rounded or
+        #    that the operation ran out of registers and needed to
+        #    use memory (spilling). This can result in performance loss.
+        #
+        # That suggests that the performance issues would be due to
+        # conversion between double and float, which we're not doing,
+        # or due to a calculation somehow running out of registers,
+        # but we're not doing a calculation.  Thus, we just suppress
+        # the warning.
+        #
+        check_and_add_compiler_option(-wd4738)
     else()
         #
         # Other compilers, including MSVC with a Clang front end and