FILE_DISPOSITION_ON_CLOSE is used to set/clear the FILE_DELETE_ON_CLOSE,
but we do not use that anymore and FILE_DISPOSITION_POSIX_SEMANTICS
already implies unmapping of the handle and removal fo it on close.
Justification: When a file is deleted on Windows, it may not be
immediately removed from the directory. This can cause problems
with future scans of that directory, which will see the partially
deleted file. Under some workloads and system configurations,
Windows files may appear to be deleted immediately.
This is the PR with requested fixup. Thanks to @SpexGuy for the
original PR.
The majority of these are in comments, some in doc comments which might
affect the generated documentation, and a few in parameter names -
nothing that should be breaking, however.
This commit installs Zig to "build_dir/stage3" during building
so that distros' can easily find binary and use it for testing/etc.
This commit also splits "add_custom_target(stage3 ALL" and command that it invokes,
so that it won't retry it during installation,
as target will be considered not out-of-date.
See also https://www.github.com/ziglang/zig/issues/14240#issuecomment-1374642063
Fixes a regression introduced in
9295355985202c267b4326b5a6e2ad5158b48e5d that caused 32-bit builds with
`-Denable-llvm` to fail to build from source due to that common
u64/usize casting issue.
Looks like I might have messed up the wasm kernel in my recent branch causing some sporadic failures on the CI.
This file was built the following way:
1. check out d0311e28b397d173f0d60c403985047ec952a172
2. do the cmake bootstrap
3. check out 57ea6207d3cb2db706bdc06c14605e4b901736dd
4. `zig build update-zig1` and stash those modified files
5. check out 440b3df702f1c8bfddabc1c594a3f49cf0011a63 (master)
6. do the cmake bootstrap
7. `zig build update-zig1` to produce this commit
Fixes#15489
This also lays the groundwork for exposing the whether or not a function is
noinline in std.builtin.Fn as an `is_noinline: bool` field if we ever want to do that.
This is needed because bug fixes to the C backend are required in order
to actually update the standard library and compiler sources to use the
new `@memcpy` and `@memset` semantics.
This is a small change to help when reading failure logs which makes the
"exited with code 1" and similar message include the test name.
Further enhancements could do the following:
* even if one unit test crashes the process, the parent process
continues running the other unit tests
* ability to test for expected panics (#1356)
* timeouts on individual tests
The previous commit introduced an optimization to the LLVM backend that
makes `@memset` lower more optimally when the element is comptime-known
and has a repeating byte pattern.
By making these functions inline, if the element parameter is
comptime-known at the callsite, it will be comptime-known in the
`@memset` call, causing more use of the LLVM `memset` intrinsic rather
than an inline for loop when using the LLVM backend.
This affects, for example, std.crypto.argon2, which calls
appendNTimesAssumeCapacity with a `[128]u64` as the element. This is now
lowered with a single `memset` call.
When the element is comptime-known, we can check if it has a repeated
byte representation. In this case, `@memset` can be lowered with the
LLVM intrinsic rather than with a loop.
