Automation roots: same_file doesn't open a path outside the roots, native NT link targets and names Windows trims are refused, and the docs say network paths are refused

This commit is contained in:
Adithya Shankaran committed 2026-11-15 18:24:05 +05:30
1 parent bad07e4433
commit d4e2641fa0
3 files changed
+145 -4

No files matched your search

+8 -2
View File
@@ -94,8 +94,14 @@ Each may be given more than once, as `--flag DIR` or `--flag=DIR`:
(writing through it would create its target, wherever that is). A path outside the roots gets
the same refusal whatever exists there, so a refusal doesn't tell what is outside. Folders are
compared by whole names: `/work2/x` is not inside `/work`. On Windows the case comes from the disk (`C:\WORK\x` is
inside `C:\work`), `\\?\` and UNC (`\\server\share\…`) paths are resolved like drive-letter ones,
and new names that are devices (`CON`, `NUL`, `COM1`…) or streams (`a.png:x`) are refused.
inside `C:\work`), `\\?\C:\…` paths are resolved like drive-letter ones, and
new names that are devices (`CON`, `NUL`, `COM1`…) or streams (`a.png:x`) are refused.
A path on another computer (`\\server\share\…`, `\\?\UNC\…`, WebDAV's `\\host@SSL\share\…`) or a device
path (`\\.\pipe\…`, `\\?\GLOBALROOT\…`) is refused by its spelling before anything looks it up, since looking
it up would contact that host, unless a root is on the same share; a link whose target is one of those is
refused too. A local volume's `\\?\Volume{…}\` path is let through. A link whose text is a native NT name
(`\Device\…`, `\??\…`, `\GLOBAL??\…`) is refused as a device path, and so is a name that ends in a period or a space, in a
path or as a link's target, since Windows trims it where it looks a name up.
- **The network:** a confined call reaches no FigCraft or plugin server, so it can't send a
document anywhere through one. Hosting, joining, publishing and sharing a file (`collab.host`,
`collab.join`, `collab.publish`, `collab.share` …), team libraries on a server (`library.check`