MONGOCRYPT-685 update libbson to 1.27.1

[kevinAlbs]

Summary

Update bundled libbson to 1.27.1

Apply workarounds to resolve errors on RHEL 6.2 with gcc 4.4.7:

• Define __STDC_LIMIT_MACROS when compiling with C++.
• Disable GCC diagnostic macros in functions for gcc < 4.6.

Remove workarounds for older libbson:

• Replace CONCAT macro with BSON_CONCAT.
• Remove BSON_SUBTYPE_ENCRYPTED (now defined in libbson).
• Remove workaround for CDRIVER-3340.

Verified with this patch build: https://spruce.mongodb.com/version/6650b53f05094a0007522646

Background

Motivation

Updating bundled libbson is to address an issue reported in Skyk that was fixed in libbson 1.25.0: https://security.snyk.io/vuln/SNYK-UNMANAGED-MONGODBMONGOCDRIVER-6156642

Workaround: define __STDC_LIMIT_MACROS for C++ targets

Defining __STDC_LIMIT_MACROS resolves an error on this patch when compiling a C++ test target: bson-cmp.h:144: error: 'INT8_MIN' was not declared in this scope. The stdint.h header on a spawned RHEL 6.2 host includes:

/* The ISO C99 standard specifies that in C++ implementations these
   macros should only be defined if explicitly requested.  */
#if !defined __cplusplus || defined __STDC_LIMIT_MACROS

Workaround: remove diagnostic pragma on old GCC

Removing the GCC diagnostic pragma resolves an error on this patch: mongoc-aggregate.c:97: error: #pragma GCC diagnostic not allowed inside functions. I expect this is unavailable on GCC 4.6 and older. GCC 4.6 changelog notes:

Support for selectively enabling and disabling warnings via #pragma GCC diagnostic has been added

Due to the workarounds, MONGOCRYPT-688 was filed to propose dropping RHEL 6.2 support for libmongocrypt.

Adding a runtime check

Removing the workaround of CDRIVER-3340 motivated adding a compile-time check to ensure libbson is newer than 1.16.0 (to include fix of CDRIVER-3340). The check for 1.16.0 is a precaution to ensure using a system install of libbson is 1.16.0 or newer. When libbson is loaded from the system, a version check is not performed.

[rcsanchez97]

LGTM

[vector-of-bool]

Is there a reason to only set the build version if its already defined for the parent project? This could also be unset(BUILD_VERSION) to clear the value from the scope and let the sub-project infer its own value, or alternatively it can unconditionally set BUILD_VERSION to the fetch tag value.

[kevinAlbs]

No. Updated to unconditionally set BUILD_VERSION.

[vector-of-bool]

If this is moved to a separate module, I'd prefer it more module-like so that it doesn't peek at variables in the includer's scope and behave different when included multiple times. For example:

find_program(GIT_EXECUTABLE git)
find_program(PATCH_EXECUTABLE patch)

#[[
    Form a new Patch-applying command for the given inputs

    make_patch_command(
        <outvar>
        [DIRECTORY <dir>]
        [STRIP_COMPONENTS <N>]
        PATCHES [<file> ...]
    )
]]
function(make_patch_command out)
    cmake_parse_arguments(PARSE_ARGV 1 patch "" "DIRECTORY;STRIP_COMPONENTS" "PATCHES")
    if(GIT_COMMAND)
        # git ...
        set(cmd ${GIT_EXECUTABLE})

        if(patch_DIRECTORY)
            # git --work-tree=...
            list(APPEND cmd --work-tree=${patch_DIRECTORY})
        endif()
        # git ... apply ...
        list(APPEND cmd apply)
        # git ... apply -pN ...
        if(patch_STRIP_COMPONENTS)
            list(APPEND cmd -p${patch_STRIP_COMPONENTS})
        endif()
        # git ... apply ...
        foreach(p IN LISTS patch_PATCHES)
            # git ... apply ... <file>
            list(APPEND cmd ${p})
        endforeach()
    else()
        # patch ...
        set(cmd ${PATCH_EXECUTABLE})
        if(patch_DIRECTORY)
            # patch --dir=...
            list(APPEND cmd --dir=${patch_DIRECTORY})
        endif()
        # patch ... -pN ...
        if(patch_STRIP_COMPONENTS)
            list(APPEND cmd -p${patch_STRIP_COMPONENTS})
        endif()
        foreach(p IN LISTS patch_PATCHES)
            # patch ... ---input=<file>
            list(APPEND cmd --input=${p})
        endforeach()
    endif()
    set("${out}" "${cmd}" PARENT_SCOPE)
endfunction()

and then callers can use it via make_patch_command

[kevinAlbs]

I like it. Thank you for the suggestion! Updated.

[vector-of-bool]

LGTM. I discovered two simplifications to the patch command function.

[rcsanchez97]

LGTM