Add filter program internal tests#1721
Merged
Merged
Conversation
4 tasks
oferchen
added a commit
that referenced
this pull request
May 4, 2026
Add two windows-latest CI jobs to close beta-blocking gaps: - windows-iocp: builds and tests with --features iocp explicitly so the IOCP code path (PRs #1717-#1721) gets pinned coverage rather than relying on default-features inclusion masking regressions. - windows-acl-xattr: builds and tests the metadata crate on Windows with --features acl,xattr so Win32 GetSecurityInfo/SetSecurityInfo and NTFS Alternate Data Stream code paths are validated by master CI. Both jobs depend on the lint job and run on stable, matching the existing windows-test convention. Closes #1900 Closes #1869
oferchen
added a commit
that referenced
this pull request
May 4, 2026
Add two windows-latest CI jobs to close beta-blocking gaps: - windows-iocp: builds and tests with --features iocp explicitly so the IOCP code path (PRs #1717-#1721) gets pinned coverage rather than relying on default-features inclusion masking regressions. - windows-acl-xattr: builds and tests the metadata crate on Windows with --features acl,xattr so Win32 GetSecurityInfo/SetSecurityInfo and NTFS Alternate Data Stream code paths are validated by master CI. Both jobs depend on the lint job and run on stable, matching the existing windows-test convention. Closes #1900 Closes #1869
oferchen
added a commit
that referenced
this pull request
May 5, 2026
* ci: add Windows IOCP and ACL/xattr matrix entries (#1900, #1869) Add two windows-latest CI jobs to close beta-blocking gaps: - windows-iocp: builds and tests with --features iocp explicitly so the IOCP code path (PRs #1717-#1721) gets pinned coverage rather than relying on default-features inclusion masking regressions. - windows-acl-xattr: builds and tests the metadata crate on Windows with --features acl,xattr so Win32 GetSecurityInfo/SetSecurityInfo and NTFS Alternate Data Stream code paths are validated by master CI. Both jobs depend on the lint job and run on stable, matching the existing windows-test convention. Closes #1900 Closes #1869 * fix: cross-platform IOCP and metadata test build on Windows - Drop redundant `use std::io::Write as _;` from disk_batch tests; the trait is already in scope via `use std::io::{self, Write};` and the re-export through `use super::*;`. `-D warnings` flagged the duplicate. - Ungate `use super::*;` and `use std::fs;` in metadata::special tests so the `#[cfg(not(unix))]` no-op tests can resolve `create_fifo`, `create_device_node`, and `fs::File` on Windows. - Gate `test_is_device_file` with `#[cfg(unix)]` to match the helper's own gate; the test referenced `is_device_file` unconditionally. Exposed for the first time by the Windows IOCP and ACL/xattr CI matrix entries added in 42f6f15 (#1900, #1869). * fix: open IOCP test files with shared write/delete access The `open_writable` test helper in disk_batch.rs opened files with `FILE_SHARE_READ` only. `IocpDiskBatch::begin_file` calls `reopen_overlapped` which uses `ReOpenFile` to obtain a second write handle with `FILE_FLAG_OVERLAPPED`. Because the original handle did not permit shared write access, the second open failed with ERROR_SHARING_VIOLATION (Windows OS error 32), causing four tests to panic at the first `begin_file` call: - commit_with_fsync_calls_flush_file_buffers - begin_file_flushes_previous - batched_submission_submits_n_chunks - completion_ordering_independent_of_submission_order Adding `FILE_SHARE_WRITE | FILE_SHARE_DELETE` to the original handle lets `ReOpenFile` succeed and lets `tempfile::tempdir` clean up the directory after the handles drop. Refs #1900, #1869. * fix: gate Path import and apply_symlink_metadata on Unix The Windows ACL/xattr CI matrix entry compiles the metadata crate with --features acl,xattr under -D warnings. On Windows, std::path::Path was unused (only consumed by the cfg(unix) current_mode helper) and apply_symlink_metadata was unused (only the cfg(unix) symlink test calls it). Both flagged as unused-import errors. Gate both imports with cfg(unix) so they only appear on platforms that consume them. * style: rustfmt-compliant import order in timestamp_2038 test * fix(ci): unblock Windows IOCP and ACL/xattr matrix entries Two unrelated Windows-only failures surfaced by the new matrix entries: 1. IOCP reopen_overlapped requested FILE_SHARE_READ only, but the handle being reopened has FILE_GENERIC_WRITE access. Win32 sharing checks are symmetric, so the new handle's share mode must permit the original handle's access mode. Widen to FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE to match the access of the handle being converted. The negative-path test still triggers via access-mode mismatch, not share-mode mismatch. 2. metadata::apply timestamp tests used FileTime::seconds(), which returns platform-native epoch seconds (Unix epoch on Unix, Windows FILETIME 1601 epoch on Windows). Switch to unix_seconds() to match the production code in apply/timestamps.rs. Gate the sub-100ns nanosecond exact-equality test #[cfg(unix)] because NTFS FILETIME has 100ns granularity and cannot roundtrip 89 trailing nanoseconds. * fix(metadata): cross-platform timestamp tests for Windows ACL/xattr CI NTFS FILETIME has 100ns granularity, so sub-100ns nanosecond fields cannot round-trip through the filesystem. Gate the two equality-based sub-100ns tests on Unix only. filetime::FileTime::seconds() returns Windows FILETIME 1601-epoch on Windows but Unix epoch on Unix; switch the remaining mtime/atime equality assertions to unix_seconds() so they hold on both platforms. * fix(tests): IOCP socket pump unwrap and NTFS sub-100ns timestamp gate The IOCP socket tests called Arc::try_unwrap(pump) while reader/writer still held an Arc clone, causing "pump uniquely owned" panics on the new Windows IOCP CI job. Drop reader/writer before unwrapping the pump. The metadata epoch_timestamp_with_nanoseconds_is_preserved test used 123_456_789ns which truncates to 123_456_700ns on NTFS (100ns granularity); gate it Unix-only matching the other sub-100ns tests. * fix(tests): gate unix-only mmap/adaptive tests and NTFS-truncating timestamp tests map_file/tests.rs: gate 78 tests that reference MmapStrategy / AdaptiveMapStrategy / open_mmap / open_adaptive(_with_threshold|_buffered) / is_mmap / is_buffered with #[cfg(unix)]. These types and methods are unix-only in mod.rs/wrapper.rs so the unconditional references break the Windows IOCP build with E0412/E0433/E0599 (89 errors). timestamp_2038.rs: gate 10 round-trip tests that pass sub-second nanosecond values to FileTime::from_unix_time(). NTFS FILETIME stores 100-nanosecond ticks so values like 999_999_999, 123_456_789, 444_555_666 truncate and break equality on Windows. Whole-second tests (no_overflow_just_past_i32_max, negative_timestamps_are_handled_correctly) remain cross-platform. Addresses CI failures on PR #3607: - Windows IOCP --features iocp (89 compile errors) - Windows ACL/xattr (file_metadata_preserves_timestamp_at_2038_boundary and directory_metadata_preserves_timestamp_beyond_2038 panics) * fix(tests): unblock Windows IOCP and ACL/xattr CI matrix Two distinct cross-platform regressions surfaced by the new Windows matrix entries: 1. ACL/xattr matrix failed to compile on Windows: gating sub-second nanosecond timestamp tests behind cfg(unix) made apply_directory_metadata and the YEAR_*/BEFORE_*/AFTER_* named constants unused on Windows under -D dead-code -D unused-imports. Gate them behind cfg(unix) too. The two cross-platform tests (no_overflow_just_past_i32_max, negative_timestamps_are_handled_correctly) only need apply_file_metadata. 2. IOCP matrix failed in generator_merge_filters_properly_scoped: the assertion compared collected paths against literals using forward slashes, but Path::display() emits backslashes on Windows. Normalize to forward slashes before comparison; the wire format itself uses forward slashes so the comparison reflects actual user-visible paths. * fix(tests): gate Seek/Write imports for sparse tests on unix only The integration_sparse tests use Seek, SeekFrom, and Write only inside #[cfg(unix)] and #[cfg(target_os = "linux")] tests. On Windows ACL/xattr matrix builds with -D warnings, the unconditional import triggered an unused-imports error and broke the test binary. Gate the import with #[cfg(unix)] to match the test gating. * fix(daemon): gate name_converter tests to unix only The name_converter_tests module is exercised only on Unix because every test inside is #[cfg(unix)]-gated. On Windows the module expanded with a `use super::*;` import and no consumers, tripping unused-imports under -D warnings in the ACL/xattr matrix entry. Move the gate to the module itself so the whole block disappears on Windows. * fix(daemon): gate fuzzy2_generate_content helper to unix only The fuzzy level 2 daemon test is `#[cfg(unix)]`-gated, but its `fuzzy2_generate_content` helper was ungated. On Windows under `-D warnings`, the unused helper produced a dead-code error and broke the Windows ACL/xattr matrix entry. Matches the existing gating pattern of the sibling `fuzzy2_backdate_file` helper in the same file. * fix(tests): gate unix-only test items for Windows ACL/xattr CI compilation The windows-acl-xattr CI step runs `cargo nextest run --workspace --features acl,xattr -E '...'`, which compiles every test crate. Workspace defaults remain enabled (including iconv), so test code that depends on unix-only APIs (`OsStrExt::from_bytes`) must be cfg-gated even when its enclosing test fns already are - otherwise the surrounding `use` items become unused on Windows and `-D warnings` fails the build. - `crates/flist/tests/special_characters.rs`: add file-level `#![cfg(unix)]`; every test in the file builds non-UTF8 names via `OsStrExt::from_bytes`. - `crates/flist/tests/unicode_comprehensive.rs`: gate the top-level `use std::ffi::OsStr;` with `#[cfg(unix)]` (its only callsites are inside `#[cfg(unix)]` invalid-UTF8 tests). - `crates/protocol/src/flist/read/tests.rs`: tighten the iconv test mod cfg from `feature = "iconv"` to `all(feature = "iconv", unix)`; every helper inside also constructs non-UTF8 filenames via OsStrExt. * fix(tests): gate Windows-incompatible chroot defaults in daemon loads_* tests Daemon module config defaults `use chroot = true` on every platform. The validator rejects relative module paths when chroot is enabled, and on Windows `Path::is_absolute()` rejects POSIX paths like /srv/docs. The 17 loads_*_from_config chunks that pin /srv/docs (or /srv/staging) therefore panicked under the windows-acl-xattr matrix entry on PR #3607. Fix: declare `use chroot = no` inline in each affected test config so the path validator skips the absolute-path check on Windows. The single chunk that asserts `module.use_chroot()` is gated `#[cfg(unix)]` instead. This preserves existing semantics on Unix and unblocks the Windows matrix for #1900 (IOCP) and #1869 (ACL/xattr).
oferchen
added a commit
that referenced
this pull request
May 5, 2026
* ci: add Windows IOCP and ACL/xattr matrix entries (#1900, #1869) Add two windows-latest CI jobs to close beta-blocking gaps: - windows-iocp: builds and tests with --features iocp explicitly so the IOCP code path (PRs #1717-#1721) gets pinned coverage rather than relying on default-features inclusion masking regressions. - windows-acl-xattr: builds and tests the metadata crate on Windows with --features acl,xattr so Win32 GetSecurityInfo/SetSecurityInfo and NTFS Alternate Data Stream code paths are validated by master CI. Both jobs depend on the lint job and run on stable, matching the existing windows-test convention. Closes #1900 Closes #1869 * fix: cross-platform IOCP and metadata test build on Windows - Drop redundant `use std::io::Write as _;` from disk_batch tests; the trait is already in scope via `use std::io::{self, Write};` and the re-export through `use super::*;`. `-D warnings` flagged the duplicate. - Ungate `use super::*;` and `use std::fs;` in metadata::special tests so the `#[cfg(not(unix))]` no-op tests can resolve `create_fifo`, `create_device_node`, and `fs::File` on Windows. - Gate `test_is_device_file` with `#[cfg(unix)]` to match the helper's own gate; the test referenced `is_device_file` unconditionally. Exposed for the first time by the Windows IOCP and ACL/xattr CI matrix entries added in 42f6f15 (#1900, #1869). * fix: open IOCP test files with shared write/delete access The `open_writable` test helper in disk_batch.rs opened files with `FILE_SHARE_READ` only. `IocpDiskBatch::begin_file` calls `reopen_overlapped` which uses `ReOpenFile` to obtain a second write handle with `FILE_FLAG_OVERLAPPED`. Because the original handle did not permit shared write access, the second open failed with ERROR_SHARING_VIOLATION (Windows OS error 32), causing four tests to panic at the first `begin_file` call: - commit_with_fsync_calls_flush_file_buffers - begin_file_flushes_previous - batched_submission_submits_n_chunks - completion_ordering_independent_of_submission_order Adding `FILE_SHARE_WRITE | FILE_SHARE_DELETE` to the original handle lets `ReOpenFile` succeed and lets `tempfile::tempdir` clean up the directory after the handles drop. Refs #1900, #1869. * fix: gate Path import and apply_symlink_metadata on Unix The Windows ACL/xattr CI matrix entry compiles the metadata crate with --features acl,xattr under -D warnings. On Windows, std::path::Path was unused (only consumed by the cfg(unix) current_mode helper) and apply_symlink_metadata was unused (only the cfg(unix) symlink test calls it). Both flagged as unused-import errors. Gate both imports with cfg(unix) so they only appear on platforms that consume them. * style: rustfmt-compliant import order in timestamp_2038 test * fix(ci): unblock Windows IOCP and ACL/xattr matrix entries Two unrelated Windows-only failures surfaced by the new matrix entries: 1. IOCP reopen_overlapped requested FILE_SHARE_READ only, but the handle being reopened has FILE_GENERIC_WRITE access. Win32 sharing checks are symmetric, so the new handle's share mode must permit the original handle's access mode. Widen to FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE to match the access of the handle being converted. The negative-path test still triggers via access-mode mismatch, not share-mode mismatch. 2. metadata::apply timestamp tests used FileTime::seconds(), which returns platform-native epoch seconds (Unix epoch on Unix, Windows FILETIME 1601 epoch on Windows). Switch to unix_seconds() to match the production code in apply/timestamps.rs. Gate the sub-100ns nanosecond exact-equality test #[cfg(unix)] because NTFS FILETIME has 100ns granularity and cannot roundtrip 89 trailing nanoseconds. * fix(metadata): cross-platform timestamp tests for Windows ACL/xattr CI NTFS FILETIME has 100ns granularity, so sub-100ns nanosecond fields cannot round-trip through the filesystem. Gate the two equality-based sub-100ns tests on Unix only. filetime::FileTime::seconds() returns Windows FILETIME 1601-epoch on Windows but Unix epoch on Unix; switch the remaining mtime/atime equality assertions to unix_seconds() so they hold on both platforms. * fix(tests): IOCP socket pump unwrap and NTFS sub-100ns timestamp gate The IOCP socket tests called Arc::try_unwrap(pump) while reader/writer still held an Arc clone, causing "pump uniquely owned" panics on the new Windows IOCP CI job. Drop reader/writer before unwrapping the pump. The metadata epoch_timestamp_with_nanoseconds_is_preserved test used 123_456_789ns which truncates to 123_456_700ns on NTFS (100ns granularity); gate it Unix-only matching the other sub-100ns tests. * fix(tests): gate unix-only mmap/adaptive tests and NTFS-truncating timestamp tests map_file/tests.rs: gate 78 tests that reference MmapStrategy / AdaptiveMapStrategy / open_mmap / open_adaptive(_with_threshold|_buffered) / is_mmap / is_buffered with #[cfg(unix)]. These types and methods are unix-only in mod.rs/wrapper.rs so the unconditional references break the Windows IOCP build with E0412/E0433/E0599 (89 errors). timestamp_2038.rs: gate 10 round-trip tests that pass sub-second nanosecond values to FileTime::from_unix_time(). NTFS FILETIME stores 100-nanosecond ticks so values like 999_999_999, 123_456_789, 444_555_666 truncate and break equality on Windows. Whole-second tests (no_overflow_just_past_i32_max, negative_timestamps_are_handled_correctly) remain cross-platform. Addresses CI failures on PR #3607: - Windows IOCP --features iocp (89 compile errors) - Windows ACL/xattr (file_metadata_preserves_timestamp_at_2038_boundary and directory_metadata_preserves_timestamp_beyond_2038 panics) * fix(tests): unblock Windows IOCP and ACL/xattr CI matrix Two distinct cross-platform regressions surfaced by the new Windows matrix entries: 1. ACL/xattr matrix failed to compile on Windows: gating sub-second nanosecond timestamp tests behind cfg(unix) made apply_directory_metadata and the YEAR_*/BEFORE_*/AFTER_* named constants unused on Windows under -D dead-code -D unused-imports. Gate them behind cfg(unix) too. The two cross-platform tests (no_overflow_just_past_i32_max, negative_timestamps_are_handled_correctly) only need apply_file_metadata. 2. IOCP matrix failed in generator_merge_filters_properly_scoped: the assertion compared collected paths against literals using forward slashes, but Path::display() emits backslashes on Windows. Normalize to forward slashes before comparison; the wire format itself uses forward slashes so the comparison reflects actual user-visible paths. * fix(tests): gate Seek/Write imports for sparse tests on unix only The integration_sparse tests use Seek, SeekFrom, and Write only inside #[cfg(unix)] and #[cfg(target_os = "linux")] tests. On Windows ACL/xattr matrix builds with -D warnings, the unconditional import triggered an unused-imports error and broke the test binary. Gate the import with #[cfg(unix)] to match the test gating. * fix(daemon): gate name_converter tests to unix only The name_converter_tests module is exercised only on Unix because every test inside is #[cfg(unix)]-gated. On Windows the module expanded with a `use super::*;` import and no consumers, tripping unused-imports under -D warnings in the ACL/xattr matrix entry. Move the gate to the module itself so the whole block disappears on Windows. * fix(daemon): gate fuzzy2_generate_content helper to unix only The fuzzy level 2 daemon test is `#[cfg(unix)]`-gated, but its `fuzzy2_generate_content` helper was ungated. On Windows under `-D warnings`, the unused helper produced a dead-code error and broke the Windows ACL/xattr matrix entry. Matches the existing gating pattern of the sibling `fuzzy2_backdate_file` helper in the same file. * fix(tests): gate unix-only test items for Windows ACL/xattr CI compilation The windows-acl-xattr CI step runs `cargo nextest run --workspace --features acl,xattr -E '...'`, which compiles every test crate. Workspace defaults remain enabled (including iconv), so test code that depends on unix-only APIs (`OsStrExt::from_bytes`) must be cfg-gated even when its enclosing test fns already are - otherwise the surrounding `use` items become unused on Windows and `-D warnings` fails the build. - `crates/flist/tests/special_characters.rs`: add file-level `#![cfg(unix)]`; every test in the file builds non-UTF8 names via `OsStrExt::from_bytes`. - `crates/flist/tests/unicode_comprehensive.rs`: gate the top-level `use std::ffi::OsStr;` with `#[cfg(unix)]` (its only callsites are inside `#[cfg(unix)]` invalid-UTF8 tests). - `crates/protocol/src/flist/read/tests.rs`: tighten the iconv test mod cfg from `feature = "iconv"` to `all(feature = "iconv", unix)`; every helper inside also constructs non-UTF8 filenames via OsStrExt. * fix(tests): gate Windows-incompatible chroot defaults in daemon loads_* tests Daemon module config defaults `use chroot = true` on every platform. The validator rejects relative module paths when chroot is enabled, and on Windows `Path::is_absolute()` rejects POSIX paths like /srv/docs. The 17 loads_*_from_config chunks that pin /srv/docs (or /srv/staging) therefore panicked under the windows-acl-xattr matrix entry on PR #3607. Fix: declare `use chroot = no` inline in each affected test config so the path validator skips the absolute-path check on Windows. The single chunk that asserts `module.use_chroot()` is gated `#[cfg(unix)]` instead. This preserves existing semantics on Unix and unblocks the Windows matrix for #1900 (IOCP) and #1869 (ACL/xattr).
This was referenced May 5, 2026
oferchen
added a commit
that referenced
this pull request
May 7, 2026
Adds a focused plan for benchmarking the IOCP file I/O path (crates/fast_io/src/iocp/, #1717-#1721) against synchronous std::fs::File writes. Covers small (4K-64K) and large (1M-100M) file workloads with 1/4/16/64 concurrent streams, cross-references matched io_uring workloads (#1868), specifies the windows-latest runner with --features iocp via xtask windows-bench, and defines seven decision criteria gating the flip of IocpPolicy::Auto to IOCP-as-default. Refs: #1899, #1868, #1717-#1721
oferchen
added a commit
that referenced
this pull request
May 18, 2026
* ci: add Windows IOCP and ACL/xattr matrix entries (#1900, #1869) Add two windows-latest CI jobs to close beta-blocking gaps: - windows-iocp: builds and tests with --features iocp explicitly so the IOCP code path (PRs #1717-#1721) gets pinned coverage rather than relying on default-features inclusion masking regressions. - windows-acl-xattr: builds and tests the metadata crate on Windows with --features acl,xattr so Win32 GetSecurityInfo/SetSecurityInfo and NTFS Alternate Data Stream code paths are validated by master CI. Both jobs depend on the lint job and run on stable, matching the existing windows-test convention. Closes #1900 Closes #1869 * fix: cross-platform IOCP and metadata test build on Windows - Drop redundant `use std::io::Write as _;` from disk_batch tests; the trait is already in scope via `use std::io::{self, Write};` and the re-export through `use super::*;`. `-D warnings` flagged the duplicate. - Ungate `use super::*;` and `use std::fs;` in metadata::special tests so the `#[cfg(not(unix))]` no-op tests can resolve `create_fifo`, `create_device_node`, and `fs::File` on Windows. - Gate `test_is_device_file` with `#[cfg(unix)]` to match the helper's own gate; the test referenced `is_device_file` unconditionally. Exposed for the first time by the Windows IOCP and ACL/xattr CI matrix entries added in 42f6f15 (#1900, #1869). * fix: open IOCP test files with shared write/delete access The `open_writable` test helper in disk_batch.rs opened files with `FILE_SHARE_READ` only. `IocpDiskBatch::begin_file` calls `reopen_overlapped` which uses `ReOpenFile` to obtain a second write handle with `FILE_FLAG_OVERLAPPED`. Because the original handle did not permit shared write access, the second open failed with ERROR_SHARING_VIOLATION (Windows OS error 32), causing four tests to panic at the first `begin_file` call: - commit_with_fsync_calls_flush_file_buffers - begin_file_flushes_previous - batched_submission_submits_n_chunks - completion_ordering_independent_of_submission_order Adding `FILE_SHARE_WRITE | FILE_SHARE_DELETE` to the original handle lets `ReOpenFile` succeed and lets `tempfile::tempdir` clean up the directory after the handles drop. Refs #1900, #1869. * fix: gate Path import and apply_symlink_metadata on Unix The Windows ACL/xattr CI matrix entry compiles the metadata crate with --features acl,xattr under -D warnings. On Windows, std::path::Path was unused (only consumed by the cfg(unix) current_mode helper) and apply_symlink_metadata was unused (only the cfg(unix) symlink test calls it). Both flagged as unused-import errors. Gate both imports with cfg(unix) so they only appear on platforms that consume them. * style: rustfmt-compliant import order in timestamp_2038 test * fix(ci): unblock Windows IOCP and ACL/xattr matrix entries Two unrelated Windows-only failures surfaced by the new matrix entries: 1. IOCP reopen_overlapped requested FILE_SHARE_READ only, but the handle being reopened has FILE_GENERIC_WRITE access. Win32 sharing checks are symmetric, so the new handle's share mode must permit the original handle's access mode. Widen to FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE to match the access of the handle being converted. The negative-path test still triggers via access-mode mismatch, not share-mode mismatch. 2. metadata::apply timestamp tests used FileTime::seconds(), which returns platform-native epoch seconds (Unix epoch on Unix, Windows FILETIME 1601 epoch on Windows). Switch to unix_seconds() to match the production code in apply/timestamps.rs. Gate the sub-100ns nanosecond exact-equality test #[cfg(unix)] because NTFS FILETIME has 100ns granularity and cannot roundtrip 89 trailing nanoseconds. * fix(metadata): cross-platform timestamp tests for Windows ACL/xattr CI NTFS FILETIME has 100ns granularity, so sub-100ns nanosecond fields cannot round-trip through the filesystem. Gate the two equality-based sub-100ns tests on Unix only. filetime::FileTime::seconds() returns Windows FILETIME 1601-epoch on Windows but Unix epoch on Unix; switch the remaining mtime/atime equality assertions to unix_seconds() so they hold on both platforms. * fix(tests): IOCP socket pump unwrap and NTFS sub-100ns timestamp gate The IOCP socket tests called Arc::try_unwrap(pump) while reader/writer still held an Arc clone, causing "pump uniquely owned" panics on the new Windows IOCP CI job. Drop reader/writer before unwrapping the pump. The metadata epoch_timestamp_with_nanoseconds_is_preserved test used 123_456_789ns which truncates to 123_456_700ns on NTFS (100ns granularity); gate it Unix-only matching the other sub-100ns tests. * fix(tests): gate unix-only mmap/adaptive tests and NTFS-truncating timestamp tests map_file/tests.rs: gate 78 tests that reference MmapStrategy / AdaptiveMapStrategy / open_mmap / open_adaptive(_with_threshold|_buffered) / is_mmap / is_buffered with #[cfg(unix)]. These types and methods are unix-only in mod.rs/wrapper.rs so the unconditional references break the Windows IOCP build with E0412/E0433/E0599 (89 errors). timestamp_2038.rs: gate 10 round-trip tests that pass sub-second nanosecond values to FileTime::from_unix_time(). NTFS FILETIME stores 100-nanosecond ticks so values like 999_999_999, 123_456_789, 444_555_666 truncate and break equality on Windows. Whole-second tests (no_overflow_just_past_i32_max, negative_timestamps_are_handled_correctly) remain cross-platform. Addresses CI failures on PR #3607: - Windows IOCP --features iocp (89 compile errors) - Windows ACL/xattr (file_metadata_preserves_timestamp_at_2038_boundary and directory_metadata_preserves_timestamp_beyond_2038 panics) * fix(tests): unblock Windows IOCP and ACL/xattr CI matrix Two distinct cross-platform regressions surfaced by the new Windows matrix entries: 1. ACL/xattr matrix failed to compile on Windows: gating sub-second nanosecond timestamp tests behind cfg(unix) made apply_directory_metadata and the YEAR_*/BEFORE_*/AFTER_* named constants unused on Windows under -D dead-code -D unused-imports. Gate them behind cfg(unix) too. The two cross-platform tests (no_overflow_just_past_i32_max, negative_timestamps_are_handled_correctly) only need apply_file_metadata. 2. IOCP matrix failed in generator_merge_filters_properly_scoped: the assertion compared collected paths against literals using forward slashes, but Path::display() emits backslashes on Windows. Normalize to forward slashes before comparison; the wire format itself uses forward slashes so the comparison reflects actual user-visible paths. * fix(tests): gate Seek/Write imports for sparse tests on unix only The integration_sparse tests use Seek, SeekFrom, and Write only inside #[cfg(unix)] and #[cfg(target_os = "linux")] tests. On Windows ACL/xattr matrix builds with -D warnings, the unconditional import triggered an unused-imports error and broke the test binary. Gate the import with #[cfg(unix)] to match the test gating. * fix(daemon): gate name_converter tests to unix only The name_converter_tests module is exercised only on Unix because every test inside is #[cfg(unix)]-gated. On Windows the module expanded with a `use super::*;` import and no consumers, tripping unused-imports under -D warnings in the ACL/xattr matrix entry. Move the gate to the module itself so the whole block disappears on Windows. * fix(daemon): gate fuzzy2_generate_content helper to unix only The fuzzy level 2 daemon test is `#[cfg(unix)]`-gated, but its `fuzzy2_generate_content` helper was ungated. On Windows under `-D warnings`, the unused helper produced a dead-code error and broke the Windows ACL/xattr matrix entry. Matches the existing gating pattern of the sibling `fuzzy2_backdate_file` helper in the same file. * fix(tests): gate unix-only test items for Windows ACL/xattr CI compilation The windows-acl-xattr CI step runs `cargo nextest run --workspace --features acl,xattr -E '...'`, which compiles every test crate. Workspace defaults remain enabled (including iconv), so test code that depends on unix-only APIs (`OsStrExt::from_bytes`) must be cfg-gated even when its enclosing test fns already are - otherwise the surrounding `use` items become unused on Windows and `-D warnings` fails the build. - `crates/flist/tests/special_characters.rs`: add file-level `#![cfg(unix)]`; every test in the file builds non-UTF8 names via `OsStrExt::from_bytes`. - `crates/flist/tests/unicode_comprehensive.rs`: gate the top-level `use std::ffi::OsStr;` with `#[cfg(unix)]` (its only callsites are inside `#[cfg(unix)]` invalid-UTF8 tests). - `crates/protocol/src/flist/read/tests.rs`: tighten the iconv test mod cfg from `feature = "iconv"` to `all(feature = "iconv", unix)`; every helper inside also constructs non-UTF8 filenames via OsStrExt. * fix(tests): gate Windows-incompatible chroot defaults in daemon loads_* tests Daemon module config defaults `use chroot = true` on every platform. The validator rejects relative module paths when chroot is enabled, and on Windows `Path::is_absolute()` rejects POSIX paths like /srv/docs. The 17 loads_*_from_config chunks that pin /srv/docs (or /srv/staging) therefore panicked under the windows-acl-xattr matrix entry on PR #3607. Fix: declare `use chroot = no` inline in each affected test config so the path validator skips the absolute-path check on Windows. The single chunk that asserts `module.use_chroot()` is gated `#[cfg(unix)]` instead. This preserves existing semantics on Unix and unblocks the Windows matrix for #1900 (IOCP) and #1869 (ACL/xattr).
oferchen
added a commit
that referenced
this pull request
May 18, 2026
Adds a focused plan for benchmarking the IOCP file I/O path (crates/fast_io/src/iocp/, #1717-#1721) against synchronous std::fs::File writes. Covers small (4K-64K) and large (1M-100M) file workloads with 1/4/16/64 concurrent streams, cross-references matched io_uring workloads (#1868), specifies the windows-latest runner with --features iocp via xtask windows-bench, and defines seven decision criteria gating the flip of IocpPolicy::Auto to IOCP-as-default. Refs: #1899, #1868, #1717-#1721
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Testing
cargo test -p rsync-engine filter_program -- --nocapturecargo llvm-cov --workspace --summary-onlyhttps://chatgpt.com/codex/tasks/task_e_69039454bda48323af7d2b64999db052