musl: 64-bit time support #4463
Conversation
rustbot
added
A-CI
|
Hmm, seems like the style checker isn't happy with mixing |
|
Since #4433 seems close to merging, I can rebase on top of it once merged. |
This feature is enabled with independently from musl_v1_2_3 to support time64. This also sets a musl_not_time64 feature to make `cfg` statements easier to read. Defining linux_time_bits64 makes this roughly equivalent to upstream commit bminor/musl@f12bd8e.
This is equivalent to upstream commit bminor/musl@9b2921b.
This corresponds to upstream commit bminor/musl@1febd21 (most symbols) and bminor/musl@22daaea (only dlsym)
A bunch of properties were removed upstream and set to reserved. This matches upstream commit bminor/musl@827aa8f and bminor/musl@2d69fcf
xbjfk
force-pushed
the
musl-time64
branch
3 times, most recently
from
44fe8c9 to
8b32bb3
Compare
Change time_t type to i64 Change struct stat, msqid_ds and shmid_ds to reflect This commit follows upstream bminor/musl@3814333 and bminor/musl@d6dcfe4 It also implements a fix from bminor/musl@0fbd7d6
This is primarily based on a small part of bminor/musl@3814333. This also integrates bminor/musl@3c02bac, which update MSG_STAT, SEM_STAT, SEM_STAT_ANY. These are based on the value of IPC_STAT, however we can just use `cfg` as it is effectively the same.
This fixes test failures on musl.
This was incorrectly named in upstream musl and fixed in bminor/musl@cabc369
This is because the struct is different depending on whether time64 is enabled.
|
Rebased on top of the GNU changes :) |
|
Thanks! Sorry it's taken me a while, I'll try to look at this very soon |
There was a problem hiding this comment.
Thanks for working on this! Left a few comments here.
For the style check, are the changes it is suggesting possible? If not, that will probably have to be updated unfortunately (hopefully we will be able to reorganize things at some point so it's less annoying...)
| if ((musl_v1_2_3 || target_os == "loongarch64") && target_env == "musl") || target_env == "ohos" | ||
| { | ||
| set_cfg("musl_v1_2_3"); | ||
| if musl_time64 && MUSL_TIME64_ARCHS.contains(&target_arch.as_str()) { | ||
| set_cfg("musl_time64"); | ||
| set_cfg("linux_time_bits64"); | ||
| } else { | ||
| set_cfg("musl_not_time64"); | ||
| } | ||
| } else if target_env == "musl" || target_env == "ohos" { | ||
| set_cfg("musl_not_time64"); |
There was a problem hiding this comment.
Why is musl_not_time64 needed? not(musl_time64) should be usable everywhere.
| --env RUST_LIBC_UNSTABLE_GNU_FILE_OFFSET_BITS \ | ||
| --env RUST_LIBC_UNSTABLE_GNU_TIME_BITS \ | ||
| --env RUST_LIBC_UNSTABLE_MUSL_V1_2_3 \ | ||
| --env RUST_LIBC_UNSTABLE_MUSL_TIME64 \ | ||
| --env CARGO_HOME=/cargo \ | ||
| --env CARGO_TARGET_DIR=/checkout/target \ |
There was a problem hiding this comment.
Nit: this should match the indentation
There was a problem hiding this comment.
I think musl_time64 as it is used should be named something like musl_time64_on_32. We probably don't need an env to set this independently, build.rs should just set it when musl_time64 is set and target_ptr_width is 32.
| #[cfg(musl_time64)] | ||
| pub const IPC_STAT: c_int = 0x102; | ||
| #[cfg(not(musl_time64))] | ||
| pub const IPC_STAT: c_int = 2; |
There was a problem hiding this comment.
Style nit: these should be writeable as pub const IPC_STAT: c_int = if cfg!(musl_time64) { 0x102 } else { 2 } to get the else fallback and reduce the amount of code that is cfged out on some platforms.
Description
This change includes time64 support for applicable architectures (x86, arm, mips and powerpc). This is based on the previous PRs to this repo as well as the musl changelog from 1.1.24 -> 1.2. It can be enabled with the environment variable
RUST_LIBC_UNSTABLE_MUSL_TIME64only when musl_v1_2_3 is enabled and the architecture is supported.A lot of structures, especially ones with mixed endian became excessively complicated, so I used
cfg_ifto separate them. It looks like you can only nests! {}incfg_if! {}, but not vice versa.As a note, I'm considering removing
musl_not_time64, and just keeping the old logic of allowing deprecated for function definitions involvingtime_tas it introduces necessary complexity.When libc 1.0 is released, I believe the best path would be to remove the
musl_v1_2_3feature, making it unconditionally enabled, keepingmusl_time64which will expand to(musl && time64_arch).Tested through QEMU for all architectures.
Sources
Sources are located on each commit, in the form of upstream commits
Checklist
libc-test/semverhave been updated*LASTor*MAXareincluded (see rust-lang/libc#3131)
cd libc-test && cargo test --target mytarget);especially relevant for platforms that may not be checked in CI
@rustbot label +stable-nominated