Previously, a relative path like `..` would:
- Attempt to be normalized (i.e. remove . and .. without any path resolution), but would error with TooManyParentDirs
- This would make wToPrefixedFileW run it through `RtlGetFullPathName_U` to do the necessary path resolution, but `RtlGetFullPathName_U` always resolves relative paths relative to the CWD
Instead, when TooManyParentDirs occurs, we now look up the path of the passed in `dir` (if it's non-null) and append the relative path to it before giving it to `RtlGetFullPathName_U`. If `dir` is null, then we just give it RtlGetFullPathName_U directly and let it resolve it relative to the CWD.
Closes#16779
Before this commit, there were three issues with the makePath implementation:
1. The component iteration did not 'collapse' consecutive path separators; instead, it would treat `a/b//c` as `a/b//c`, `a/b/`, `a/b`, and `a`.
2. Trailing path separators led to redundant `makeDir` calls, e.g. with the path `a/b/` (if `a` doesn't exist), it would try to create `a/b/`, then try `a/b`, then try `a`, then try `a/b`, and finally try `a/b/` again.
3. The iteration did not treat the root of a path specially, so on Windows it could attempt to make a directory with a path like `X:` for something like `X:\a\b\c` if the `X:\` drive doesn't exist. This didn't lead to any problems that I could find, but there's no reason to try to make a drive letter as a directory (or any other root path).
This commit fixes all three issues by introducing a ComponentIterator that is root-aware and handles both sequential path separators and trailing path separators and uses it in `Dir.makePath`. This reduces the number of `makeDir` calls for paths where (1) the root of the path doesn't exist, (2) there are consecutive path separators, or (3) there are trailing path separators
As an example, here are the makeDir calls that would have been made before this commit when calling `makePath` for a relative path like `a/b//c//` (where the full path needs to be created):
a/b//c//
a/b//c/
a/b//c
a/b/
a/b
a
a/b
a/b/
a/b//c
a/b//c/
a/b//c//
And after this commit:
a/b//c
a/b
a
a/b
a/b//c
When calling NtCreateFile with a UNC path, if either `\\server` or `\\server\share` are not found, then the statuses `BAD_NETWORK_PATH` or `BAD_NETWORK_NAME` are returned (respectively).
These statuses are not translated into `error.FileNotFound` because they convey more information than the typical FileNotFound error. For example, if you were trying to call `Dir.makePath` with an absolute UNC path like `\\MyServer\MyShare\a\b\c\d`, then knowing that `\\MyServer\MyShare` was not found allows for returning after trying to create the first directory instead of then trying to create `a\b\c`, `a\b`, etc. when it's already known that they will all fail in the same way.
Most of this migration was performed automatically with `zig fmt`. There
were a few exceptions which I had to manually fix:
* `@alignCast` and `@addrSpaceCast` cannot be automatically rewritten
* `@truncate`'s fixup is incorrect for vectors
* Test cases are not formatted, and their error locations change
- fix getdents return type usize → isize
- usize ultimately forced errors to .SUCCESS in std.c.getError
New behavior in freebsd 13.2 is to return ENOENT if the directory being
iterated is deleted during iteration. We now detect this and treat it
consistent with iteration ending.
Before this commit, if Walker.next errored with e.g. `error.AccessDenied` and the caller did something like `while (true) { walker.next() catch continue; }`, then the directory that errored with AccessDenied would be continually iterated in each `next` call and error every time with AccessDenied.
After this commit, the directory that errored will be popped off the stack before the error is returned, meaning that in the subsequent `next` call, it won't be retried and the Walker will continue with whatever directories remain on its stack.
For a real example, before this commit, walking `/proc/` on my system would infinitely loop due to repeated AccessDenied errors on the same directory. After this commit, I am able to walk `/proc/` on my system fully (skipping over any directories that are unable to be iterated).
Since we are opening each directory for iteration, we know that we don't need to reset the cursor's directory before iterating. Using `iterateAssumeFirstIteration` skips the cursor resetting which eliminates an `lseek` syscall for every directory opened on non-Windows platforms.
This doesn't seem to actually matter much for performance (1.01 ± 0.02 times faster when walking /home/ on my system) but avoiding unnecessary syscalls is always nice anyway.
The idea here is that there are two ways we can reference a function at runtime:
* Through a direct call, i.e. where the function is comptime-known
* Through a function pointer
This means we can easily perform a form of rudimentary escape analysis
on functions. If we ever see a `decl_ref` or `ref` of a function, we
have a function pointer, which could "leak" into runtime code, so we
emit the function; but for a plain `decl_val`, there's no need to.
This change means that `comptime { _ = f; }` no longer forces a function
to be emitted, which was used for some things (mainly tests). These use
sites have been replaced with `_ = &f;`, which still triggers analysis
of the function body, since you're taking a pointer to the function.
Resolves: #6256Resolves: #15353
addition to #15450
createFileW does not account for failure on `LockFile`. This can result
in a file handle not being closed on failure which can be seen on test
such as `fs.test.'open file with exclusive nonblocking lock twice'`.
Also add `std.fs.has_executable_bit` for doing conditional compilation.
This adds the linux syscalls for chmod and fchmodat, as well as the
extern libc function declarations.
Only `fchmodat` is added to `std.os`, and it is not yet added to std.fs.
These functions are currently footgunny when working with pointers to
arrays and slices. They just return the stated length of the array/slice
without iterating and looking for the first sentinel, even if the
array/slice is a sentinel terminated type.
From looking at the quite small list of places in the standard
library/compiler that this change breaks existing code, the new code
looks to be more readable in all cases.
The usage of std.mem.span/len was totally unneeded in most of the cases
affected by this breaking change.
We could remove these functions entirely in favor of other existing
functions in std.mem such as std.mem.sliceTo(), but that would be a
somewhat nasty breaking change as std.mem.span() is very widely used for
converting sentinel terminated pointers to slices. It is however not at
all widely used for anything else.
Therefore I think it is better to break these few non-standard and
potentially incorrect usages of these functions now and at some later
time, if deemed worthwhile, finally remove these functions.
If we wait for at least a full release cycle so that everyone adapts to
this change first, updating for the removal could be a simple find and
replace without needing to worry about the semantics.
All but 3 callsites of this function in the standard library and
compiler were unnecessary and were removed in faf2fd18.
In this commit, the remaining 3 callsites are removed. One of them
turned out to also be unnecessary and has been replaced by slicing
directly with the length..
The 2 remaining callsites were in the very pointer-math heavy
std/os/linux/vdso.zig code which should perhaps be refactored to better
utilize slices. These 2 callsites are replaced with a plain
@ptrCast([*:0]u8, ptr) though could likely use std.mem.sliceTo() if the
surrounding code was refactored.
This fixes a bug in std.net caused during the introduction of
meta.assumeSentinel due to the unfortunate semantics of mem.span()
This leaves only 3 remaining uses of meta.assumeSentinel() in the
standard library, each of which could be a simple @ptrCast([*:0]T, foo)
instead. I think this function should likely be removed.
Can occur when trying to open a directory for iteration but the 'List folder contents' permission of the directory is set to 'Deny'.
This was found because it was being triggered during PATH searching in ChildProcess.spawnWindows if a PATH entry did not have 'List folder contents' permission, so this fixes that as well (note: the behavior on hitting this during PATH searching is to treat it as the directory not existing and therefore will fail to find any executables in a directory in the PATH without 'List folder contents' permission; this matches Windows behavior which also fails to find commands in directories that do not have 'List folder contents' permission).
* revert changes to Module because the error set is consistent across
operating systems.
* remove duplicated Stat.fromSystem code and use a less redundant name.
* make fs.Dir.statFile follow symlinks, and avoid pointless control
flow through the posix layer.
This branch largely reverts 58f961f4cb. I
would like to revisit the proposal to modify the standard library in
this way and think more carefully about it before adding isAbsolute()
checks everywhere.
Instead of checking for absolute paths and current working directories
in various file system operations, there is one simple solution: allow
overriding `std.fs.cwd` on WASI.
os.realpath is back to causing a compile error when used on WASI. This
caused a compile error in the Sema handling of `@src()`. The compiler
should never call realpath, so the commit that made this change is
reverted (95ab942184). If this breaks
debug info, a different strategy is needed to solve it other than using
realpath.
I also removed the preopens code and replaced it with something much
simpler. There is no longer any global state in the standard library.
Additionally-
* os.openat no longer does an unnecessary fstat on WASI when O.WRONLY
is not provided.
* os.chdir is back to causing a compile error on WASI.
Make the test use the minimum length and set MAX_NAME_BYTES to the maximum so that:
- the test will work on any host platform
- *and* the MAX_NAME_BYTES will be able to hold the max file name component on any host platform
Each u16 within a file name component can be encoded as up to 3 UTF-8 bytes, so we need to use MAX_NAME_BYTES to account for all possible UTF-8 encoded names.
Fixes#8268
