Expose sync() for ResizableFile?

[mcourteaux]

I'm building a tiny database system on top of the ResizableFile system. It seems beneficial to have self-defined sync() calls whenever I finish an operation on the database. I'd rather not want to wait until the very end of my application to write back the changes to disk.

This function is accessible through MappedMemory, however this struct member is private in ResizableFile. Would it be a bad idea to expose this sync() primitive in ResizableFile?

[pknowles]

Good point. This is needed!

• For linux that's straight forward
• I would like it to remain cross platform, so I'll have to get an equivalent for windows too. Looks like I'll want NtFlushVirtualMemory
• I'm also thinking I should find a way to limit the exposed calls - e.g. if someone were to use sync() on linux at the moment it'd fail to compile on windows. The simplest might be a wrapper class rather than the current using and concepts.

I guess people might want to be able to flush a subrange too.

[mcourteaux]

• I'm also thinking I should find a way to limit the exposed calls - e.g. if someone were to use sync() on linux at the moment it'd fail to compile on windows. The simplest might be a wrapper class rather than the current using and concepts.

I'd say avoiding wrappers is the way to go. C++ is expressive enough to simply have a static_assert take care of you not calling this function on Windows, if it doesn't exist yet.

[pknowles]

Added a .sync() and .sync(offset, size) in #14. Thanks for the suggestion!

[mcourteaux]

Wauw! Thanks a lot for taking the time to do this. Happy to suggest things like this 😄