A failure naming a long path keeps its reason, and name refuses a component over 255 bytes

A Save of a 3000-byte path failed with the record "err 2 ctl: Save…", which
lost the reason. The first cause was reportError, which formatted into 256
bytes, so the operation's path filled the buffer before the reason was
written. The second was the waiting write's late failure, which took the
first 256 bytes of the message row, never its end. Now reportError has room
for the longest path. reportFailure also keeps the words fitted as an err is
(fitErr: the path gives up its middle, the reason stays), and the late
failure of a Save, a Dump, a shell or a ThemeFile takes those words.

fs.md already said a name holds up to 255 bytes a component, but a longer
one was taken and only failed later at Save. Now it is refused when written.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

1 files changed, 6 insertions, 0 deletions

diff --git a/src/fs.zig b/src/fs.zig
index e66a6a49..30ad9c43 100644
--- a/src/fs.zig
+++ b/src/fs.zig
@@ -1343,6 +1343,12 @@ pub const Namespace = struct {
     no_pane_slot: bool = false,
     no_pane_slot_len: u16 = 0,
     header_held: bool = false,
+    /// The newest failure said (reportFailure), fitted as an err is: its
+    /// path cut in the middle, never its reason, which the message row's
+    /// first 256 bytes lose. `said_at` is the `failures` count it was.
+    said: [256]u8 = undefined,
+    said_len: u16 = 0,
+    said_at: u32 = 0,
     late_failure: [256]u8 = undefined,
     late_failure_len: u16 = 0,
     /// A 9P write asked a language server something: the answer count it