Re: [PATCH v2] gdbsupport: add path_join function

[Eli Zaretskii <eliz@gnu.org>]

> Date: Wed, 20 Apr 2022 12:52:04 +0100
> From: Pedro Alves <pedro@palves.net>
> Cc: Simon Marchi <simon.marchi@efficios.com>
> 
> I understand we don't implement the root name checks described in 1) at:
> 
>    https://en.cppreference.com/w/cpp/filesystem/path/append
> 
> however, at least recognizing "drive:/" and "drive:\" as absolute paths
> I think works as is, and I'd think is important that it works for GDB.  So I think
> we should have these tests under _WIN32:
> 
>    test_one ("C:\\bar", "C:\\foo", "C:\\bar");  // '\' sep.
>    test_one ("C:/bar", "C:\\foo", "C:/bar");    // '/' sep.
>    test_one ("D:\\bar", "C:\\foo", "D:\\bar");  // 'D:' wins.

That sounds outside of the scope of path_join, which is just a smart
concatenation function.  What you describe sounds more like
canonicalize_filename or any other function which resolves a file name
to produce an absolute one.  There's a place for those as well, but
one doesn't always want an absolute file name.  Also, resolving file
names could include symlink resolution and other stuff, and that is
really out of scope here.

How many instances do we have in our sources where path_join is used
to produce an absolute file name and arguments other than the first
one could be absolute?

> " (...) joining with an absolute path on the right hand side, like this:
>  
>    path_join ("/foo", "/bar");
>  
>  results in just the right hand side being kept.  In this example, the
>  result is "/bar".
> "

Or maybe I've lost the track of the discussion, since the above
doesn't sound consistent with what I though this function should do.
Did we switch the concept at some point?