[Mmap] Turn PAGESIZE into a function #59712
Conversation
`PAGESIZE` is currently fixed at compile time, but the machine where Julia runs may have a different memory page size than the one where the stdlib was compiled on. It's not worth making this a `OncePerProcess` object because the underlying function call is very cheap, cheaper than the overhead of `OncePerProcess`.
|
I think this is probably a good change, but the page size issue is also a problem for page alignment in linked files. Are we sure this will actually work for precompile files? Perhaps we should separate the concept of "you must not assume page size smaller than this" and "this is the current system's page size"? |
|
@Keno why is the linker relevant to this? That seems like a linker or patchelf bug instead? e.g. |
|
The alignment is a linker setting. The default is 64k now, so there's no problem, but my point is just that the there still is some build-time value for this, so we should possibly separate the concepts. If only to be ready for when someone decides that 1MB is the correct page granularity ;). |
|
Gotcha, that sounds unrelated to this PR in that case, and would likely be |
|
It's related in that they're currently kind of the same concept so we may want to add it and mention both in whatever transition guidance we provide. |
PAGESIZEis currently fixed at compile time, but the machine where Julia runsmay have a different memory page size than the one where the stdlib was compiled
on. It's not worth making this a
OncePerProcessobject because the underlyingfunction call is very cheap, cheaper than the overhead of
OncePerProcess.This is somewhat breaking for those packages which referred to the non-public
Mmap.PAGESIZEsymbol, luckily they are only three.