Re: [PATCH v1/RFC 1/1] 'git clone <url> C:\cygwin\home\USER\repo' is working (again)

From: Junio C Hamano <gitster@pobox.com>
Date: Nov 27, 2018, 05:23 UTC
Message-ID: <xmqqlg5ft7pe.fsf@gitster-ct.c.googlers.com>
In-Reply-To: <CAAXzdLX4jU7+i1W1A_Q1LpPFa1D4FYVPW5rcMnqr_tDHEJn+tw@mail.gmail.com>

Steven Penny <svnpenn@gmail.com> writes:

Show 14 quoted lines

> If you strip the drive, you can still navigate within the same drive:
>
>     $ cd 'C:\Users'
>     $ pwd
>     /cygdrive/c/Users
>
>     $ cd '\Windows'
>     $ pwd
>     /cygdrive/c/Windows
>
> but you can no longer traverse drives:
>
>     $ cd '\Testing'
>     sh: cd: \Testing: No such file or directory

Sorry, but I fail to see the point the last example wants to make. If it were

    $ cd /cygdrive/d/Testing
    $ cd /cygdrive/c/Users
    $ cd ../../d/Testing

and the last step fails, then I would suspect it would make sense to treat /cygdrive/$X exactly like how we would treat $C:, because

    $ cd C:Users
    $ cd ../D:Testing

would not make sense, either, which is an indication that these two are quite similar. On the other hand, if "cd ../../d/Testing" above does not fail and does what non-DOS users would expect, then that strongly argues that treating /cygdrive/$X any specially is a mistake.

> So a good first question for me would be: why are we stripping "C:" or similar
> in the first place?

Sorry, but I do not see the connection to this question and the above example. The reason why we strip C: is because the letter that comes after that colon determines if we are talking about absolute path (in other words, the current directory does not play a role in determining which directory the path refers to), unlike the POSIX codepath where it is always the first letter in the pathname.

C:\Users is a directory whose name is Users at the top level of the
C: drive. C0\Users tells us that in the current directory, there is
a directory whose name is C0 and in it, there is a filesystem entity
whose name is Users.  So the colon that follows an alpha (in this
case, C:) is quite special, compared to other letters (in this
example, I used '0' to contrast its effect with ':').  So it is very
understandable why we want to have has_dos_prefix() and
skip_dos_prefix().

> I would say these could be merged into a "win.h" or similar. Cygwin typically
> leans toward the "/unix/style" while MINGW has been more tolerant of
> "C:\Windows\Style" and "C:/Mixed/Style" paths, i dont see that changing.

I'd defer to Windows folks to decide if a unified win.h is a good idea.

Thanks.