Re: [patch] Implement ISO/IEC TS 18822 C++ File system TS

[Jonathan Wakely <jwakely at redhat dot com>]

On 01/05/15 11:53 +0100, Jonathan Wakely wrote:
> On 01/05/15 10:05 +0900, Luke Allardyce wrote:
> > This fails on mingw-w64 4.02 due to
>
> Yay, thanks for trying it!
>
> > an extra typename:
> >
> > /mnt/build/native/gcc/x86_64-w64-mingw32/libstdc++-v3/include/experimental/fs_path.h:784:35:
> > error: expected nested-name-specifier
> >      using _CharAlloc = typename __alloc_rebind<_Allocator, char>;
>
> Oops, forgetting how to use my own alias.
>
>
> > readdir_r not available:
> >
> > /mnt/src/gcc/libstdc++-v3/src/filesystem/dir.cc:169:46: error:
> > 'readdir_r' was not declared in this scope
> >  if (int err = readdir_r(dirp, &ent, &result))
> >
> >
> > and calling opendir with a wchar_t*
> >
> > /mnt/src/gcc/libstdc++-v3/src/filesystem/dir.cc:108:40: error: cannot
> > convert 'const value_type* {aka const wchar_t*}' to 'const char*' for
> > argument '1' to 'DIR* opendir(const char*)'
> >    if (DIR* dirp = ::opendir(p.c_str()))
> >
> > /mnt/src/gcc/libstdc++-v3/src/filesystem/dir.cc:256:38: error: cannot
> > convert 'const value_type* {aka const wchar_t*}' to 'const char*' for
> > argument '1' to 'DIR* opendir(const char*)'
> >  if (DIR* dirp = ::opendir(p.c_str()))
>
> Ah interesting, this means mingw supports <dirent.h>, I was expecting
> it to use the dummy definitions at the top of src/filesystem/dir.cc
> (where I'd made opendir take wchar_t).
>
> I've done some reading and am going to commit some fixes and
> improvements for mingw in an hour or two. It would be great if you
> could try it again after that.

I've committed the two changes attached (only tested on linux again).

patch2.txt should fix the mingw-w64 errors above, as well as the
issues Daniel reported, and should fix the error on Solaris 10

Rainer, would you be able to test with
--enable-libstdcxx-filesystem-ts before we re-enable it to build by
default on solaris* ?


patch1.txt changes path::_M_cmpts from std::list to std::vector (as
Marc suggested months ago) and fixes the pretty printer, so paths are
shown like this in GDB:

(gdb) print p
$1 = filesystem::path ""
(gdb) print c
$2 = filesystem::path "/home/jwakely" = {[root-directory] = "/", [1] = "home", [2] = "jwakely"}
(gdb) print r
$3 = filesystem::path "/" [root-directory]
(gdb) print h
$4 = filesystem::path "home"
(gdb) print l
$5 = filesystem::path "//host/foo/bar///baz/.//xyzzy/meta/syntactic/variable" = {[root-name] = "//host", [root-directory] = "/", [2] = "foo", [3] = "bar", [4] = "baz", [5] = ".", [6] = "xyzzy", [7] = "meta", [8] = "syntactic", [9] = "variable"}

That output came from debugging this code:

 path p;
 path c = current_path();
 path r = *c.begin();
 path h = *std::next(c.begin());
 path l = "//host/foo/bar///baz/.//xyzzy/meta/syntactic/variable";


Thanks everyone for testing and feedback, it's been very helpful.

Attachment: patch1.txt
Description: Text document

Attachment: patch2.txt
Description: Text document