You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
conformance shim: SPEC_TMP_BASE is /tmp, a symlink on darwin — 11 realpath/realdirpath examples compare a resolved path with an unresolved one #630
Real mspec roots tmp() at the spec repo's own tmp/ directory, not at /tmp, so it never meets this.
Cost measured against the pinned corpus: 5 of 5 remaining failures in core/file/realpath_spec.rb and 6 of 7 in core/file/realdirpath_spec.rb. (The seventh, "raises Errno::ENOENT if the symlink points to an absent directory", was a genuine VM defect and is fixed in #629.)
Fix: point SPEC_TMP_BASE at a directory that is not reached through a symlink — e.g. resolve it once at load time with File.realpath, or root it under the corpus checkout.
Two other shim gaps cost examples in the same area and could go in the same change:
tmp("") returns a uniquified non-existent path (/tmp/rbgo_spec_tmp/file-1-...) where mspec returns the tmp directory. core/file/stat/blockdev_spec.rb, chardev_spec.rb and the shared open_directory examples in core/file/{open,new}_spec.rb all call tmp("") and get Errno::ENOENT.
as_real_superuser and with_block_device are undefined, so the collection blocks that use them fail to load in core/filetest/{executable,readable,writable}_real_spec.rb and core/file/{writable,readable,executable}_real_spec.rb and core/file/stat/ftype_spec.rb.
Activity
README: the subsystems that landed since the feature list was written…
README: the subsystems that landed since the feature list was written…
README: the subsystems that landed since the feature list was written…
README: measured ruby/spec conformance (21,982 vs the 15,400+ claimed…