Windows: board-identity errors print KiCad's open boards as \\?\ verbatim paths beside a raw requested path

[pauliuszaleckas]

Summary

On Windows, the board-identity errors print KiCad's open boards as
extended-length ("verbatim") paths while printing the requested board as the
caller typed it, so one message shows the same directory in two spellings:

requested board 'C:\Users\me\proj\board.kicad_pcb' is not open in KiCad
  (open boards: \\?\C:\Users\me\proj\other.kicad_pcb)

Board identity is unaffected — the comparison is correct and this is a
display problem only. But these errors exist so a caller can see which board
KiCad actually holds and retry against it, and a \\?\ path is a poor thing to
hand back for that purpose.

Reported by

@mixelpixx, reviewing #656 on KiCad 10.0.5 / Windows 11, against a release
build driving a standalone pcbnew over stdio. Quoting the review:

On Windows the wrong_document message prints the open board with a \\?\
prefix. Cosmetic, probably older than this PR; worth a follow-up issue, not a
change here.

Filing it as asked. Not reproduced by me — I am on Fedora Linux, where
canonicalize returns a plain absolute path and the defect cannot appear. What
follows is a code trace, not a second measurement, and a Windows user
confirming the exact rendering would be welcome.

Mechanism

In select_requested_board (crates/konnect-ipc/src/client.rs) the two halves
of the message come from different forms of the path:

• requested is the raw argument — requested.display().to_string() (:4202).
• open_documents is built from the identities (:4273), which are
  comparable_identity() output (:4335) — i.e. path.canonicalize().

std::fs::canonicalize on Windows returns a verbatim path (\\?\C:\…); on
Unix it returns a plain absolute path. Hence the asymmetry, and hence
Windows-only.

Scope

board_document_label (:4186) is the shared source, so this is wider than
wrong_document. Every error kind built from an open-document identity shows
verbatim paths on Windows:

Error kind	Field
wrong_document	open_documents
ambiguous_target	candidates
stale_target	previously_bound, and the open-document list in reason
ambiguous_open_board	the "reports '…' open more than once" reason (:4265)

requested is raw in all of them, so every one of these can show two spellings
of one directory side by side.

Age

comparable_identity landed in f8139e8 (2026-08-31, "tell a KiCad without this
board open apart from a refusal"). There is no verbatim-prefix handling
anywhere in the workspace today. So this predates #656, which touches
client.rs only in the enabled-layer methods and none of the identity or label
code.

Expected

Render these paths in the form a user typed and can paste back, while leaving
the comparison on the canonical form.

Two constraints for whoever picks this up:

1. Display only. Canonicalized paths must keep flowing into
   select_requested_board's matching. Normalising before the comparison would
   reintroduce exactly the identity defect fix(pcb): tell a KiCad without this board open apart from a refusal #407 fixed.
2. \\?\UNC\ is not a prefix you can strip. \\?\UNC\server\share\x has
   to render as \\server\share\x, not UNC\server\share\x. A naive
   strip_prefix(r"\\?\") silently corrupts every network path.

Worth a test matrix covering a drive path, a UNC path, a path that does not
exist (which takes comparable_identity's lexical branch and so has no prefix
to begin with), and a Unix path that must pass through untouched.

Related

#537 also concerns how Konnect reports paths, but is a different defect —
which path gets resolved and whether the error names it. This one is about
how an already-resolved absolute path is rendered.

[pauliuszaleckas]

Correction: I first credited this report to @neusse. The \\?\ observation and
the quoted line are @mixelpixx's, from the #656 review; @neusse wrote the
earlier comments on #542 and I carried that name over without re-checking the
author. The body is fixed — apologies to both.