Re: [PATCH 1/9] Fix possible Cygwin build problem

[Jon Turney <jon.turney@dronecode.org.uk>]

On 26/05/2022 20:04, Eli Zaretskii wrote:
>> From: Tom Tromey <tromey@adacore.com>
>> Cc: Jon Turney <jon.turney@dronecode.org.uk>,  Eli Zaretskii <eliz@gnu.org>,
>>    tromey@adacore.com
>> Date: Thu, 26 May 2022 11:00:45 -0600
>>
>>>>>>> "Eli" == Eli Zaretskii via Gdb-patches <gdb-patches@sourceware.org> writes:
>>
>> Eli> __USEWIDE in native Windows programming assumes the use of wchar_t for
>> Eli> character and character strings, something that GDB sources aren't
>> Eli> equipped for doing, at least not easily.  AFAIK, Cygwin alleviates
>> Eli> that by using UTF-8 encoded strings, but native Windows programming
>> Eli> cannot (yet) do that safely enough on all supported Windows versions.
>>
>> It looks like windows-nat.c defines __USEWIDE only for Cygwin, and then
>> is careful to use the *W forms of various Windows APIs.  So maybe it
>> works ok in practice?
> 
> I doubt it, in general, because GDB uses 'char *' for strings, and
> those won't work in *W APIs without an explicit conversion to wchar_t.
> 
> However, I see only 2 calls to *W APIs in windows-nat.c, and they are
> both under the __CYGWIN__ condition.  So I think we are good.
> 
>> Eli> As long as MinGW builds of GDB support only the so-called "ANSI"
>> Eli> encoding (i.e. the current system codepage), there's no reason to use
>> Eli> __USEWIDE in the MinGW build of GDB, and every reason not to use it.
>>
>> Yeah, I think __USEWIDE is not used for mingw.  However I wonder if we
>> could remove it for Cygwin as well... that would clean up some code.
>> Not a huge amount, but every Cygwin divergence is a bit of a pain.
> 
> This will need a Cygwin expert, which I'm not.

I think that would be a step backwards.

The reasons for this code seem to have been discussed, back when it was 
added [1].

I think we must use the *W() API variants on cygwin, so that the paths 
can be losslessly converted char * encoded in cygwin's conception of the 
locale (which is not the Windows codepage, and may not be representable 
as one).

The only way I can think of this to be unified is to always use the *W() 
API variants, keep that data which is internal to windows-nat in WCHAR, 
and convert to a displayable name for the rest of gdb using wcstombs() 
(on cygwin) or WideCharToMultiByte() on Windows.

[1] https://sourceware.org/pipermail/gdb-patches/2010-February/072173.html