On Wed, 10 Jul 2024 at 09:02, hellord <halouwo...@gmail.com> wrote: > > >> >> >> On Tue, Jul 9, 2024 at 10:39 PM Peter Maydell <peter.mayd...@linaro.org> >> wrote: >> Is there a particular reason to use 128MB here? If the >> runtime only fails on 4GB or more, maybe we should use >> a larger MAX_CHUNK_SIZE, like 2GB ? > > > According to current analysis, size <= 4GB all are safe, however there are > many > versions of msvcrt, this bug exists on Server 2008/2019/2022 and Windows > 11(all > with full latest updates), and it may also exist in other versions, but it is > difficult to > check each version individually. I am not sure if all versions handle > boundary sizes > like 2GB/4GB correctly. So I prefer a relatively conservative value: 128MB. > > Maybe we could use #ifdef _WIN32 to differentiate the handling between Linux > and > Windows. For Linux, it remains unchanged, while for Windows, it processes by > chunks > with max_chunk_sizeto 1GB.
I don't think it's worth making this Windows-specific. I agree that it's OK to be a bit conservative, but 128MB seems to me extremely conservative. I think we could say, for instance, 512MB or 1GB, without being at much danger of running into broken implementations here. thanks -- PMM
