For the record, something I do not understand is why we never come across this problem when compiling with msvc64.

I think (I can't find it properly stated anywhere) that MASM automatically aligns DATA segments to PARAGRAPH (so ALIGN 16) which means you wouldn't see the effect?

I think (I can't find it properly stated anywhere) that MASM automatically aligns DATA segments to PARAGRAPH (so ALIGN 16) which means you wouldn't see the effect?

Spot on! https://docs.microsoft.com/en-us/cpp/assembler/masm/segment?view=vs-2019 says that PARA is used by default.

If you agree, do you care approving the PR? I'll add a Changes entry.

Just to confirm - spotted on mingw64, but your larger fail case didn’t apply on msvc64, but did apply on amd64 on another arch?

Just to confirm - spotted on mingw64, but your larger fail case didn’t apply on msvc64, but did apply on amd64 on another arch?

No, just mingw64.

It just seems the right thing to do to emit the alignment for all amd64 backends.

I rewrote the Changes entry to be more accurate.

Thanks!

4.09 aa70fea

Should this go to 4.08 ?

I see no harm in cherry-picking to 4.08. We don't currently have a bugfix-release planned, but you can never know, and some people may in the future want to use the head of the branch anyway.

OK, thanks! I added a new heading for a hypothetical 4.08.1 release: d665e78

For 4.06 (for example) we used:

4.06 maintenance branch:
------------------------

which I think is slightly better unless a release is actually planned.

Thanks, I fixed it as suggested.