[PATCH] PR21482: long paths on Windows

[Michael Spencer]

On Tue, Nov 11, 2014 at 8:52 PM, Robinson, Paul
<Paul_Robinson at playstation.sony.com> wrote:
> New patch attached.  Threw together . and .. elimination for Michael.
> Extra test for Rafael, although I had to #include <windows.h> in the
> unittest to make the necessary Windows APIs available. (Testing a relative
> path requires setting the current directory to a known writeable place,
> which means calling Windows APIs because path/fs don't have them.)  If it's
> going overboard to do that, let me know and I can take the relative-path
> test back out; but actually I think it's useful to have it.
> --paulr

LGTM. Thanks for fixing this.

- Michael Spencer

>
>> -----Original Message-----
>> From: llvm-commits-bounces at cs.uiuc.edu [mailto:llvm-commits-
>> bounces at cs.uiuc.edu] On Behalf Of Robinson, Paul
>> Sent: Tuesday, November 11, 2014 2:13 PM
>> To: Michael Spencer
>> Cc: llvm-commits at cs.uiuc.edu; Aaron Ballman
>> Subject: RE: [PATCH] PR21482: long paths on Windows
>>
>> > -----Original Message-----
>> > From: Michael Spencer [mailto:bigcheesegs at gmail.com]
>> > Sent: Tuesday, November 11, 2014 12:36 PM
>> > To: Robinson, Paul
>> > Cc: Aaron Ballman; Rafael Espíndola; llvm-commits at cs.uiuc.edu
>> > Subject: Re: [PATCH] PR21482: long paths on Windows
>> >
>> > On Tue, Nov 11, 2014 at 7:27 AM, Robinson, Paul
>> > <Paul_Robinson at playstation.sony.com> wrote:
>> > > New patch attached.
>> > >
>> > > I ended up not using make_absolute because it caused more string
>> > copying.
>> > > If there's some reason to consider it safer than current_path and
>> doing
>> > > my own pasting, I can do that instead, but only if y'all think it has
>> > > actual value.
>> > >
>> > > This changed the specific error returned by another fs test, which
>> > > seems reasonable.
>> > > --paulr
>> >
>> > This patch doesn't seem to be able to handle "..\foo". \\?\ disables
>> > expansion of . and ..
>>
>> Feh.
>>
>> I'm not able to find a pre-Windows8 API that does this canonicalization
>> on sufficiently long path names.  If anybody else has one I'm all ears.
>> Otherwise I guess I have to write my own.
>> --paulr
>>
>> >
>> > Other than that it looks fine.
>> >
>> > - Michael Spencer
>> >
>> > >
>> > >> -----Original Message-----
>> > >> From: Michael Spencer [mailto:bigcheesegs at gmail.com]
>> > >> Sent: Monday, November 10, 2014 4:04 PM
>> > >> To: Robinson, Paul
>> > >> Cc: Aaron Ballman; Rafael Espíndola; llvm-commits at cs.uiuc.edu
>> > >> Subject: Re: [PATCH] PR21482: long paths on Windows
>> > >>
>> > >> On Mon, Nov 10, 2014 at 3:05 PM, Robinson, Paul
>> > >> <Paul_Robinson at playstation.sony.com> wrote:
>> > >> >> -----Original Message-----
>> > >> >> From: Aaron Ballman [mailto:aaron.ballman at gmail.com]
>> > >> >> Sent: Monday, November 10, 2014 2:47 PM
>> > >> >> To: Rafael Espíndola
>> > >> >> Cc: Robinson, Paul; llvm-commits at cs.uiuc.edu
>> > >> >> Subject: Re: [PATCH] PR21482: long paths on Windows
>> > >> >>
>> > >> >> On Mon, Nov 10, 2014 at 5:43 PM, Rafael Espíndola
>> > >> >> <rafael.espindola at gmail.com> wrote:
>> > >> >> > static std::error_code widen_path(const Twine &Path8,
>> > >> >> >
>> > >> >> > widenPath
>> > >> >> >
>> > >> >> > Using  '\\?\'  also disables using / as a path separator, no?
>> > Don't
>> > >> we
>> > >> >> > have to check/assert that there is no / in Path8?
>> > >> >
>> > >> > So for paranoia I should translate all / to \ in the widened path?
>> I
>> > >> don't
>> > >> > see any existing utility to do that, but easy enough to invent one.
>> > >>
>> > >> sys::path::native(). in Path.h
>> > >>
>> > >> We also have make_absolute in FileSystem.h
>> > >>
>> > >> - Michael Spencer
>> > >>
>> > >> >
>> > >> >> >
>> > >> >> > +  size_t NLevels = ((248 - TmpLen) / 10) + 1;
>> > >> >> > +  const char *OneDir = "\\123456789";
>> > >> >> >
>> > >> >> > Use the size of OneDir instead of hard codding the 10.
>> > >> >> >
>> > >> >> > Aaron, this only uses '\\?\' to create long paths. They can
>> still
>> > be
>> > >> >> > accessed by regular tools by changing the current directory, no?
>> > >> >>
>> > >> >> Yes, relative paths aren't an issue. Just tools using absolute
>> paths
>> > >> >> that don't support the extended path namespace (which can happen
>> if
>> > >> >> the app assumes MAX_PATH is actually the max).
>> > >> >>
>> > >> >> ~Aaron
>> > >> >>
>> > >> >> >
>> > >> >> > On 10 November 2014 16:01, Robinson, Paul
>> > >> >> > <Paul_Robinson at playstation.sony.com> wrote:
>> > >> >> >> Support directory names longer than 248 characters on Windows.
>> > >> >> >>
>> > >> >> >> The normal Windows path limit is 260 characters; the limit for
>> > >> >> directory
>> > >> >> >> names is 248, to leave room for an 8.3 filename at the end.
>> > Adding
>> > >> the
>> > >> >> >> '\\?\' prefix greatly expands these limits.  Intercept path
>> names
>> > >> that
>> > >> >> >> we are about to pass to the Windows APIs and add the prefix if
>> > >> >> necessary.
>> > >> >> >>
>> > >> >> >> Also correct the comment about the state of the temporary
>> > directory
>> > >> >> used
>> > >> >> >> by the Support unittests.
>> > >> >> >> --paulr
>> > >> >> >>
>> > >> >> >>
>> > >> >> >> _______________________________________________
>> > >> >> >> llvm-commits mailing list
>> > >> >> >> llvm-commits at cs.uiuc.edu
>> > >> >> >> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>> > >> >> >>
>> > >> >
>> > >> > _______________________________________________
>> > >> > llvm-commits mailing list
>> > >> > llvm-commits at cs.uiuc.edu
>> > >> > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits